Tag Archives: machine learning

TSMixer: An all-MLP architecture for time series forecasting

Posted by Si-An Chen, Student Researcher, Cloud AI Team, and Chun-Liang Li, Research Scientist, Cloud AI Team

Time series forecasting is critical to various real-world applications, from demand forecasting to pandemic spread prediction. In multivariate time series forecasting (forecasting multiple variants at the same time), one can split existing methods into two categories: univariate models and multivariate models. Univariate models focus on inter-series interactions or temporal patterns that encompass trends and seasonal patterns on a time series with a single variable. Examples of such trends and seasonal patterns might be the way mortgage rates increase due to inflation, and how traffic peaks during rush hour. In addition to inter-series patterns, multivariate models process intra-series features, known as cross-variate information, which is especially useful when one series is an advanced indicator of another series. For example, a rise in body weight may cause an increase in blood pressure, and increasing the price of a product may lead to a decrease in sales. Multivariate models have recently become popular solutions for multivariate forecasting as practitioners believe their capability of handling cross-variate information may lead to better performance.

In recent years, deep learning Transformer-based architectures have become a popular choice for multivariate forecasting models due to their superior performance on sequence tasks. However, advanced multivariate models perform surprisingly worse than simple univariate linear models on commonly-used long-term forecasting benchmarks, such as Electricity Transformer Temperature (ETT), Electricity, Traffic, and Weather. These results raise two questions:

Does cross-variate information benefit time series forecasting?
When cross-variate information is not beneficial, can multivariate models still perform as well as univariate models?

In “TSMixer: An All-MLP Architecture for Time Series Forecasting”, we analyze the advantages of univariate linear models and reveal their effectiveness. Insights from this analysis lead us to develop Time-Series Mixer (TSMixer), an advanced multivariate model that leverages linear model characteristics and performs well on long-term forecasting benchmarks. To the best of our knowledge, TSMixer is the first multivariate model that performs as well as state-of-the-art univariate models on long-term forecasting benchmarks, where we show that cross-variate information is less beneficial. To demonstrate the importance of cross-variate information, we evaluate a more challenging real-world application, M5. Finally, empirical results show that TSMixer outperforms state-of-the-art models, such as PatchTST, Fedformer, Autoformer, DeepAR and TFT.

TSMixer architecture

A key difference between linear models and Transformers is how they capture temporal patterns. On one hand, linear models apply fixed and time-step-dependent weights to capture static temporal patterns, and are unable to process cross-variate information. On the other hand, Transformers use attention mechanisms that apply dynamic and data-dependent weights at each time step, capturing dynamic temporal patterns and enabling them to process cross-variate information.

In our analysis, we show that under common assumptions of temporal patterns, linear models have naïve solutions to perfectly recover the time series or place bounds on the error, which means they are great solutions for learning static temporal patterns of univariate time series more effectively. In contrast, it is non-trivial to find similar solutions for attention mechanisms, as the weights applied to each time step are dynamic. Consequently, we develop a new architecture by replacing Transformer attention layers with linear layers. The resulting TSMixer model, which is similar to the computer vision MLP-Mixer method, alternates between applications of the multi-layer perceptron in different directions, which we call time-mixing and feature-mixing, respectively. The TSMixer architecture efficiently captures both temporal patterns and cross-variate information, as shown in the figure below. The residual designs ensure that TSMixer retains the capacity of temporal linear models while still being able to exploit cross-variate information.

Transformer block and TSMixer block architectures. TSMixer replaces the multi-head attention layer with time-mixing, a linear model applied on the time dimension.

Comparison between data-dependent (attention mechanisms) and time-step-dependent (linear models). This is an example of forecasting the next time step by learning the weights of the previous three time steps.

Evaluation on long-term forecasting benchmarks

We evaluate TSMixer using seven popular long-term forecasting datasets (ETTm1, ETTm2, ETTh1, ETTh2, Electricity, Traffic, and Weather), where recent research has shown that univariate linear models outperform advanced multivariate models with large margins. We compare TSMixer with state-of-the-art multivariate models (TFT, FEDformer, Autoformer, Informer), and univariate models, including linear models and PatchTST. The figure below shows the average improvement of mean squared error (MSE) by TSMixer compared with others. The average is calculated across datasets and multiple forecasting horizons. We demonstrate that TSMixer significantly outperforms other multivariate models and performs on par with state-of-the-art univariate models. These results show that multivariate models are capable of performing as well as univariate models.

The average MSE improvement of TSMixer compared with other baselines. The red bars show multivariate methods and the blue bars show univariate methods. TSMixer achieves significant improvement over other multivariate models and achieves comparable results to univariate models.

Ablation study

We performed an ablation study to compare TSMixer with TMix-Only, a TSMixer variant that consists of time mixing layers only. The results show that TMix-Only performs almost the same as TSMixer, which means the additional feature mixing layers do not improve the performance and confirms that cross-variate information is less beneficial on popular benchmarks. The results validate the superior univariate model performance shown in previous research. However, existing long-term forecasting benchmarks are not well representative of the need for cross-variate information in some real-world applications where time series may be intermittent or sparse, hence temporal patterns may not be sufficient for forecasting. Therefore, it may be inappropriate to evaluate multivariate forecasting models solely on these benchmarks.

Evaluation on M5: Effectiveness of cross-variate information

To further demonstrate the benefit of multivariate models, we evaluate TSMixer on the challenging M5 benchmark, a large-scale retail dataset containing crucial cross-variate interactions. M5 contains the information of 30,490 products collected over 5 years. Each product description includes time series data, like daily sales, sell price, promotional event information, and static (non-time-series) features, such as store location and product category. The goal is to forecast the daily sales of each product for the next 28 days, evaluated using the weighted root mean square scaled error (WRMSSE) from the M5 competition. The complicated nature of retail makes it more challenging to forecast solely using univariate models that focus on temporal patterns, so multivariate models with cross-variate information and even auxiliary features are more essential.

First, we compare TSMixer to other methods only considering the historical data, such as daily sales and historical sell prices. The results show that multivariate models outperforms univariate models significantly, indicating the usefulness of cross-variate information. And among all compared methods, TSMixer effectively leverages the cross-variate information and achieves the best performance.

Additionally, to leverage more information, such as static features (e.g., store location, product category) and future time series (e.g., a promotional event scheduled in coming days) provided in M5, we propose a principle design to extend TSMixer. The extended TSMixer aligns different types of features into the same length, and then applies multiple mixing layers to the concatenated features to make predictions. The extended TSMixer architecture outperforms models popular in industrial applications, including DeepAR and TFT, showcasing its strong potential for real-world impact.

The architecture of the extended TSMixer. In the first stage (align stage), it aligns the different types of features into the same length before concatenating them. In the second stage (mixing stage) it applies multiple mixing layers conditioned with static features.

The WRMSSE on M5. The first three methods (blue) are univariate models. The middle three methods (orange) are multivariate models that consider only historical features. The last three methods (red) are multivariate models that consider historical, future, and static features.

Conclusion

We present TSMixer, an advanced multivariate model that leverages linear model characteristics and performs as well as state-of-the-art univariate models on long-term forecasting benchmarks. TSMixer creates new possibilities for the development of time series forecasting architectures by providing insights into the importance of cross-variate and auxiliary information in real-world scenarios. The empirical results highlight the need to consider more realistic benchmarks for multivariate forecasting models in future research. We hope that this work will inspire further exploration in the field of time series forecasting, and lead to the development of more powerful and effective models that can be applied to real-world applications.

Acknowledgements

This research was conducted by Si-An Chen, Chun-Liang Li, Nate Yoder, Sercan O. Arik, and Tomas Pfister.

Source: Google AI Blog

SayTap: Language to quadrupedal locomotion

Posted by Yujin Tang and Wenhao Yu, Research Scientists, Google

Simple and effective interaction between human and quadrupedal robots paves the way towards creating intelligent and capable helper robots, forging a future where technology enhances our lives in ways beyond our imagination. Key to such human-robot interaction systems is enabling quadrupedal robots to respond to natural language instructions. Recent developments in large language models (LLMs) have demonstrated the potential to perform high-level planning. Yet, it remains a challenge for LLMs to comprehend low-level commands, such as joint angle targets or motor torques, especially for inherently unstable legged robots, necessitating high-frequency control signals. Consequently, most existing work presumes the provision of high-level APIs for LLMs to dictate robot behavior, inherently limiting the system’s expressive capabilities.

In “SayTap: Language to Quadrupedal Locomotion”, we propose an approach that uses foot contact patterns (which refer to the sequence and manner in which a four-legged agent places its feet on the ground while moving) as an interface to bridge human commands in natural language and a locomotion controller that outputs low-level commands. This results in an interactive quadrupedal robot system that allows users to flexibly craft diverse locomotion behaviors (e.g., a user can ask the robot to walk, run, jump or make other movements using simple language). We contribute an LLM prompt design, a reward function, and a method to expose the SayTap controller to the feasible distribution of contact patterns. We demonstrate that SayTap is a controller capable of achieving diverse locomotion patterns that can be transferred to real robot hardware.

SayTap method

The SayTap approach uses a contact pattern template, which is a 4 X T matrix of 0s and 1s, with 0s representing an agent’s feet in the air and 1s for feet on the ground. From top to bottom, each row in the matrix gives the foot contact patterns of the front left (FL), front right (FR), rear left (RL) and rear right (RR) feet. SayTap’s control frequency is 50 Hz, so each 0 or 1 lasts 0.02 seconds. In this work, a desired foot contact pattern is defined by a cyclic sliding window of size L_w and of shape 4 X L_w. The sliding window extracts from the contact pattern template four foot ground contact flags, which indicate if a foot is on the ground or in the air between t + 1 and t + L_w. The figure below provides an overview of the SayTap method.

SayTap introduces these desired foot contact patterns as a new interface between natural language user commands and the locomotion controller. The locomotion controller is used to complete the main task (e.g., following specified velocities) and to place the robot’s feet on the ground at the specified time, such that the realized foot contact patterns are as close to the desired contact patterns as possible. To achieve this, the locomotion controller takes the desired foot contact pattern at each time step as its input in addition to the robot’s proprioceptive sensory data (e.g., joint positions and velocities) and task-related inputs (e.g., user-specified velocity commands). We use deep reinforcement learning to train the locomotion controller and represent it as a deep neural network. During controller training, a random generator samples the desired foot contact patterns, the policy is then optimized to output low-level robot actions to achieve the desired foot contact pattern. Then at test time a LLM translates user commands into foot contact patterns.

SayTap approach overview.

SayTap uses foot contact patterns (e.g., 0 and 1 sequences for each foot in the inset, where 0s are foot in the air and 1s are foot on the ground) as an interface that bridges natural language user commands and low-level control commands. With a reinforcement learning-based locomotion controller that is trained to realize the desired contact patterns, SayTap allows a quadrupedal robot to take both simple and direct instructions (e.g., “Trot forward slowly.”) as well as vague user commands (e.g., “Good news, we are going to a picnic this weekend!”) and react accordingly.

We demonstrate that the LLM is capable of accurately mapping user commands into foot contact pattern templates in specified formats when given properly designed prompts, even in cases when the commands are unstructured or vague. In training, we use a random pattern generator to produce contact pattern templates that are of various pattern lengths T, foot-ground contact ratios within a cycle based on a given gait type G, so that the locomotion controller gets to learn on a wide distribution of movements leading to better generalization. See the paper for more details.

Results

With a simple prompt that contains only three in-context examples of commonly seen foot contact patterns, an LLM can translate various human commands accurately into contact patterns and even generalize to those that do not explicitly specify how the robot should react.

SayTap prompts are concise and consist of four components: (1) general instruction that describes the tasks the LLM should accomplish; (2) gait definition that reminds the LLM of basic knowledge about quadrupedal gaits and how they can be related to emotions; (3) output format definition; and (4) examples that give the LLM chances to learn in-context. We also specify five velocities that allow a robot to move forward or backward, fast or slow, or remain still.

General instruction block
You are a dog foot contact pattern expert.
Your job is to give a velocity and a foot contact pattern based on the input.
You will always give the output in the correct format no matter what the input is.

Gait definition block
The following are description about gaits:
1. Trotting is a gait where two diagonally opposite legs strike the ground at the same time.
2. Pacing is a gait where the two legs on the left/right side of the body strike the ground at the same time.
3. Bounding is a gait where the two front/rear legs strike the ground at the same time. It has a longer suspension phase where all feet are off the ground, for example, for at least 25% of the cycle length. This gait also gives a happy feeling.

Output format definition block
The following are rules for describing the velocity and foot contact patterns:
1. You should first output the velocity, then the foot contact pattern.
2. There are five velocities to choose from: [-1.0, -0.5, 0.0, 0.5, 1.0].
3. A pattern has 4 lines, each of which represents the foot contact pattern of a leg.
4. Each line has a label. "FL" is front left leg, "FR" is front right leg, "RL" is rear left leg, and "RR" is rear right leg.
5. In each line, "0" represents foot in the air, "1" represents foot on the ground.

Example block
Input: Trot slowly
Output: 0.5
FL: 11111111111111111000000000
FR: 00000000011111111111111111
RL: 00000000011111111111111111
RR: 11111111111111111000000000

Input: Bound in place
Output: 0.0
FL: 11111111111100000000000000
FR: 11111111111100000000000000
RL: 00000011111111111100000000
RR: 00000011111111111100000000

Input: Pace backward fast
Output: -1.0
FL: 11111111100001111111110000
FR: 00001111111110000111111111
RL: 11111111100001111111110000
RR: 00001111111110000111111111

Input:

SayTap prompt to the LLM. Texts in blue are used for illustration and are not input to LLM.

Following simple and direct commands

We demonstrate in the videos below that the SayTap system can successfully perform tasks where the commands are direct and clear. Although some commands are not covered by the three in-context examples, we are able to guide the LLM to express its internal knowledge from the pre-training phase via the “Gait definition block” (see the second block in our prompt above) in the prompt.

Following unstructured or vague commands

But what is more interesting is SayTap’s ability to process unstructured and vague instructions. With only a little hint in the prompt to connect certain gaits with general impressions of emotions, the robot bounds up and down when hearing exciting messages, like “We are going to a picnic!” Furthermore, it also presents the scenes accurately (e.g., moving quickly with its feet barely touching the ground when told the ground is very hot).

Conclusion and future work

We present SayTap, an interactive system for quadrupedal robots that allows users to flexibly craft diverse locomotion behaviors. SayTap introduces desired foot contact patterns as a new interface between natural language and the low-level controller. This new interface is straightforward and flexible, moreover, it allows a robot to follow both direct instructions and commands that do not explicitly state how the robot should react.

One interesting direction for future work is to test if commands that imply a specific feeling will allow the LLM to output a desired gait. In the gait definition block shown in the results section above, we provide a sentence that connects a happy mood with bounding gaits. We believe that providing more information can augment the LLM’s interpretations (e.g., implied feelings). In our evaluation, the connection between a happy feeling and a bounding gait led the robot to act vividly when following vague human commands. Another interesting direction for future work is to introduce multi-modal inputs, such as videos and audio. Foot contact patterns translated from those signals will, in theory, still work with our pipeline and will unlock many more interesting use cases.

Acknowledgements

Yujin Tang, Wenhao Yu, Jie Tan, Heiga Zen, Aleksandra Faust and Tatsuya Harada conducted this research. This work was conceived and performed while the team was in Google Research and will be continued at Google DeepMind. The authors would like to thank Tingnan Zhang, Linda Luu, Kuang-Huei Lee, Vincent Vanhoucke and Douglas Eck for their valuable discussions and technical support in the experiments.

Source: Google AI Blog

Language to rewards for robotic skill synthesis

Posted by Wenhao Yu and Fei Xia, Research Scientists, Google

Empowering end-users to interactively teach robots to perform novel tasks is a crucial capability for their successful integration into real-world applications. For example, a user may want to teach a robot dog to perform a new trick, or teach a manipulator robot how to organize a lunch box based on user preferences. The recent advancements in large language models (LLMs) pre-trained on extensive internet data have shown a promising path towards achieving this goal. Indeed, researchers have explored diverse ways of leveraging LLMs for robotics, from step-by-step planning and goal-oriented dialogue to robot-code-writing agents.

While these methods impart new modes of compositional generalization, they focus on using language to link together new behaviors from an existing library of control primitives that are either manually engineered or learned a priori. Despite having internal knowledge about robot motions, LLMs struggle to directly output low-level robot commands due to the limited availability of relevant training data. As a result, the expression of these methods are bottlenecked by the breadth of the available primitives, the design of which often requires extensive expert knowledge or massive data collection.

In “Language to Rewards for Robotic Skill Synthesis”, we propose an approach to enable users to teach robots novel actions through natural language input. To do so, we leverage reward functions as an interface that bridges the gap between language and low-level robot actions. We posit that reward functions provide an ideal interface for such tasks given their richness in semantics, modularity, and interpretability. They also provide a direct connection to low-level policies through black-box optimization or reinforcement learning (RL). We developed a language-to-reward system that leverages LLMs to translate natural language user instructions into reward-specifying code and then applies MuJoCo MPC to find optimal low-level robot actions that maximize the generated reward function. We demonstrate our language-to-reward system on a variety of robotic control tasks in simulation using a quadruped robot and a dexterous manipulator robot. We further validate our method on a physical robot manipulator.

The language-to-reward system consists of two core components: (1) a Reward Translator, and (2) a Motion Controller. The Reward Translator maps natural language instruction from users to reward functions represented as python code. The Motion Controller optimizes the given reward function using receding horizon optimization to find the optimal low-level robot actions, such as the amount of torque that should be applied to each robot motor.

LLMs cannot directly generate low-level robotic actions due to lack of data in pre-training dataset. We propose to use reward functions to bridge the gap between language and low-level robot actions, and enable novel complex robot motions from natural language instructions.

Reward Translator: Translating user instructions to reward functions

The Reward Translator module was built with the goal of mapping natural language user instructions to reward functions. Reward tuning is highly domain-specific and requires expert knowledge, so it was not surprising to us when we found that LLMs trained on generic language datasets are unable to directly generate a reward function for a specific hardware. To address this, we apply the in-context learning ability of LLMs. Furthermore, we split the Reward Translator into two sub-modules: Motion Descriptor and Reward Coder.

Motion Descriptor

First, we design a Motion Descriptor that interprets input from a user and expands it into a natural language description of the desired robot motion following a predefined template. This Motion Descriptor turns potentially ambiguous or vague user instructions into more specific and descriptive robot motions, making the reward coding task more stable. Moreover, users interact with the system through the motion description field, so this also provides a more interpretable interface for users compared to directly showing the reward function.

To create the Motion Descriptor, we use an LLM to translate the user input into a detailed description of the desired robot motion. We design prompts that guide the LLMs to output the motion description with the right amount of details and format. By translating a vague user instruction into a more detailed description, we are able to more reliably generate the reward function with our system. This idea can also be potentially applied more generally beyond robotics tasks, and is relevant to Inner-Monologue and chain-of-thought prompting.

Reward Coder

In the second stage, we use the same LLM from Motion Descriptor for Reward Coder, which translates generated motion description into the reward function. Reward functions are represented using python code to benefit from the LLMs’ knowledge of reward, coding, and code structure.

Ideally, we would like to use an LLM to directly generate a reward function R (s, t) that maps the robot state s and time t into a scalar reward value. However, generating the correct reward function from scratch is still a challenging problem for LLMs and correcting the errors requires the user to understand the generated code to provide the right feedback. As such, we pre-define a set of reward terms that are commonly used for the robot of interest and allow LLMs to composite different reward terms to formulate the final reward function. To achieve this, we design a prompt that specifies the reward terms and guide the LLM to generate the correct reward function for the task.

The internal structure of the Reward Translator, which is tasked to map user inputs to reward functions.

Motion Controller: Translating reward functions to robot actions

The Motion Controller takes the reward function generated by the Reward Translator and synthesizes a controller that maps robot observation to low-level robot actions. To do this, we formulate the controller synthesis problem as a Markov decision process (MDP), which can be solved using different strategies, including RL, offline trajectory optimization, or model predictive control (MPC). Specifically, we use an open-source implementation based on the MuJoCo MPC (MJPC).

MJPC has demonstrated the interactive creation of diverse behaviors, such as legged locomotion, grasping, and finger-gaiting, while supporting multiple planning algorithms, such as iterative linear–quadratic–Gaussian (iLQG) and predictive sampling. More importantly, the frequent re-planning in MJPC empowers its robustness to uncertainties in the system and enables an interactive motion synthesis and correction system when combined with LLMs.

Examples

Robot dog

In the first example, we apply the language-to-reward system to a simulated quadruped robot and teach it to perform various skills. For each skill, the user will provide a concise instruction to the system, which will then synthesize the robot motion by using reward functions as an intermediate interface.

Dexterous manipulator

We then apply the language-to-reward system to a dexterous manipulator robot to perform a variety of manipulation tasks. The dexterous manipulator has 27 degrees of freedom, which is very challenging to control. Many of these tasks require manipulation skills beyond grasping, making it difficult for pre-designed primitives to work. We also include an example where the user can interactively instruct the robot to place an apple inside a drawer.

Validation on real robots

We also validate the language-to-reward method using a real-world manipulation robot to perform tasks such as picking up objects and opening a drawer. To perform the optimization in Motion Controller, we use AprilTag, a fiducial marker system, and F-VLM, an open-vocabulary object detection tool, to identify the position of the table and objects being manipulated.

Conclusion

In this work, we describe a new paradigm for interfacing an LLM with a robot through reward functions, powered by a low-level model predictive control tool, MuJoCo MPC. Using reward functions as the interface enables LLMs to work in a semantic-rich space that plays to the strengths of LLMs, while ensuring the expressiveness of the resulting controller. To further improve the performance of the system, we propose to use a structured motion description template to better extract internal knowledge about robot motions from LLMs. We demonstrate our proposed system on two simulated robot platforms and one real robot for both locomotion and manipulation tasks.

Acknowledgements

We would like to thank our co-authors Nimrod Gileadi, Chuyuan Fu, Sean Kirmani, Kuang-Huei Lee, Montse Gonzalez Arenas, Hao-Tien Lewis Chiang, Tom Erez, Leonard Hasenclever, Brian Ichter, Ted Xiao, Peng Xu, Andy Zeng, Tingnan Zhang, Nicolas Heess, Dorsa Sadigh, Jie Tan, and Yuval Tassa for their help and support in various aspects of the project. We would also like to acknowledge Ken Caluwaerts, Kristian Hartikainen, Steven Bohez, Carolina Parada, Marc Toussaint, and the greater teams at Google DeepMind for their feedback and contributions.

Source: Google AI Blog

Autonomous visual information seeking with large language models

Posted by Ziniu Hu, Student Researcher, and Alireza Fathi, Research Scientist, Google Research, Perception Team

There has been great progress towards adapting large language models (LLMs) to accommodate multimodal inputs for tasks including image captioning, visual question answering (VQA), and open vocabulary recognition. Despite such achievements, current state-of-the-art visual language models (VLMs) perform inadequately on visual information seeking datasets, such as Infoseek and OK-VQA, where external knowledge is required to answer the questions.

Examples of visual information seeking queries where external knowledge is required to answer the question. Images are taken from the OK-VQA dataset.

In “AVIS: Autonomous Visual Information Seeking with Large Language Models”, we introduce a novel method that achieves state-of-the-art results on visual information seeking tasks. Our method integrates LLMs with three types of tools: (i) computer vision tools for extracting visual information from images, (ii) a web search tool for retrieving open world knowledge and facts, and (iii) an image search tool to glean relevant information from metadata associated with visually similar images. AVIS employs an LLM-powered planner to choose tools and queries at each step. It also uses an LLM-powered reasoner to analyze tool outputs and extract key information. A working memory component retains information throughout the process.

An example of AVIS’s generated workflow for answering a challenging visual information seeking question. The input image is taken from the Infoseek dataset.

Comparison to previous work

Recent studies (e.g., Chameleon, ViperGPT and MM-ReAct) explored adding tools to LLMs for multimodal inputs. These systems follow a two-stage process: planning (breaking down questions into structured programs or instructions) and execution (using tools to gather information). Despite success in basic tasks, this approach often falters in complex real-world scenarios.

There has also been a surge of interest in applying LLMs as autonomous agents (e.g., WebGPT and ReAct). These agents interact with their environment, adapt based on real-time feedback, and achieve goals. However, these methods do not restrict the tools that can be invoked at each stage, leading to an immense search space. Consequently, even the most advanced LLMs today can fall into infinite loops or propagate errors. AVIS tackles this via guided LLM use, influenced by human decisions from a user study.

Informing LLM decision making with a user study

Many of the visual questions in datasets such as Infoseek and OK-VQA pose a challenge even for humans, often requiring the assistance of various tools and APIs. An example question from the OK-VQA dataset is shown below. We conducted a user study to understand human decision-making when using external tools.

We conducted a user study to understand human decision-making when using external tools. Image is taken from the OK-VQA dataset.

The users were equipped with an identical set of tools as our method, including PALI, PaLM, and web search. They received input images, questions, detected object crops, and buttons linked to image search results. These buttons offered diverse information about the detected object crops, such as knowledge graph entities, similar image captions, related product titles, and identical image captions.

We record user actions and outputs and use it as a guide for our system in two key ways. First, we construct a transition graph (shown below) by analyzing the sequence of decisions made by users. This graph defines distinct states and restricts the available set of actions at each state. For example, at the start state, the system can take only one of these three actions: PALI caption, PALI VQA, or object detection. Second, we use the examples of human decision-making to guide our planner and reasoner with relevant contextual instances to enhance the performance and effectiveness of our system.

AVIS transition graph.

General framework

Our approach employs a dynamic decision-making strategy designed to respond to visual information-seeking queries. Our system has three primary components. First, we have a planner to determine the subsequent action, including the appropriate API call and the query it needs to process. Second, we have a working memory that retains information about the results obtained from API executions. Last, we have a reasoner, whose role is to process the outputs from the API calls. It determines whether the obtained information is sufficient to produce the final response, or if additional data retrieval is required.

The planner undertakes a series of steps each time a decision is required regarding which tool to employ and what query to send to it. Based on the present state, the planner provides a range of potential subsequent actions. The potential action space may be so large that it makes the search space intractable. To address this issue, the planner refers to the transition graph to eliminate irrelevant actions. The planner also excludes the actions that have already been taken before and are stored in the working memory.

Next, the planner collects a set of relevant in-context examples that are assembled from the decisions previously made by humans during the user study. With these examples and the working memory that holds data collected from past tool interactions, the planner formulates a prompt. The prompt is then sent to the LLM, which returns a structured answer, determining the next tool to be activated and the query to be dispatched to it. This design allows the planner to be invoked multiple times throughout the process, thereby facilitating dynamic decision-making that gradually leads to answering the input query.

We employ a reasoner to analyze the output of the tool execution, extract the useful information and decide into which category the tool output falls: informative, uninformative, or final answer. Our method utilizes the LLM with appropriate prompting and in-context examples to perform the reasoning. If the reasoner concludes that it’s ready to provide an answer, it will output the final response, thus concluding the task. If it determines that the tool output is uninformative, it will revert back to the planner to select another action based on the current state. If it finds the tool output to be useful, it will modify the state and transfer control back to the planner to make a new decision at the new state.

AVIS employs a dynamic decision-making strategy to respond to visual information-seeking queries.

Results

We evaluate AVIS on Infoseek and OK-VQA datasets. As shown below, even robust visual-language models, such as OFA and PaLI, fail to yield high accuracy when fine-tuned on Infoseek. Our approach (AVIS), without fine-tuning, achieves 50.7% accuracy on the unseen entity split of this dataset.

AVIS visual question answering results on Infoseek dataset. AVIS achieves higher accuracy in comparison to previous baselines based on PaLI, PaLM and OFA.

Our results on the OK-VQA dataset are shown below. AVIS with few-shot in-context examples achieves an accuracy of 60.2%, higher than most of the previous works. AVIS achieves lower but comparable accuracy in comparison to the PALI model fine-tuned on OK-VQA. This difference, compared to Infoseek where AVIS outperforms fine-tuned PALI, is due to the fact that most question-answer examples in OK-VQA rely on common sense knowledge rather than on fine-grained knowledge. Therefore, PaLI is able to encode such generic knowledge in the model parameters and doesn’t require external knowledge.

Visual question answering results on A-OKVQA. AVIS achieves higher accuracy in comparison to previous works that use few-shot or zero-shot learning, including Flamingo, PaLI and ViperGPT. AVIS also achieves higher accuracy than most of the previous works that are fine-tuned on OK-VQA dataset, including REVEAL, ReVIVE, KAT and KRISP, and achieves results that are close to the fine-tuned PaLI model.

Conclusion

We present a novel approach that equips LLMs with the ability to use a variety of tools for answering knowledge-intensive visual questions. Our methodology, anchored in human decision-making data collected from a user study, employs a structured framework that uses an LLM-powered planner to dynamically decide on tool selection and query formation. An LLM-powered reasoner is tasked with processing and extracting key information from the output of the selected tool. Our method iteratively employs the planner and reasoner to leverage different tools until all necessary information required to answer the visual question is amassed.

Acknowledgements

This research was conducted by Ziniu Hu, Ahmet Iscen, Chen Sun, Kai-Wei Chang, Yizhou Sun, David A. Ross, Cordelia Schmid and Alireza Fathi.

Source: Google AI Blog

Neural network pruning with combinatorial optimization

Posted by Hussein Hazimeh, Research Scientist, Athena Team, and Riade Benbaki, Graduate Student at MIT

Modern neural networks have achieved impressive performance across a variety of applications, such as language, mathematical reasoning, and vision. However, these networks often use large architectures that require lots of computational resources. This can make it impractical to serve such models to users, especially in resource-constrained environments like wearables and smartphones. A widely used approach to mitigate the inference costs of pre-trained networks is to prune them by removing some of their weights, in a way that doesn’t significantly affect utility. In standard neural networks, each weight defines a connection between two neurons. So after weights are pruned, the input will propagate through a smaller set of connections and thus requires less computational resources.

Original network vs. a pruned network.

Pruning methods can be applied at different stages of the network’s training process: post, during, or before training (i.e., immediately after weight initialization). In this post, we focus on the post-training setting: given a pre-trained network, how can we determine which weights should be pruned? One popular method is magnitude pruning, which removes weights with the smallest magnitude. While efficient, this method doesn’t directly consider the effect of removing weights on the network’s performance. Another popular paradigm is optimization-based pruning, which removes weights based on how much their removal impacts the loss function. Although conceptually appealing, most existing optimization-based approaches seem to face a serious tradeoff between performance and computational requirements. Methods that make crude approximations (e.g., assuming a diagonal Hessian matrix) can scale well, but have relatively low performance. On the other hand, while methods that make fewer approximations tend to perform better, they appear to be much less scalable.

In “Fast as CHITA: Neural Network Pruning with Combinatorial Optimization”, presented at ICML 2023, we describe how we developed an optimization-based approach for pruning pre-trained neural networks at scale. CHITA (which stands for “Combinatorial Hessian-free Iterative Thresholding Algorithm”) outperforms existing pruning methods in terms of scalability and performance tradeoffs, and it does so by leveraging advances from several fields, including high-dimensional statistics, combinatorial optimization, and neural network pruning. For example, CHITA can be 20x to 1000x faster than state-of-the-art methods for pruning ResNet and improves accuracy by over 10% in many settings.

Overview of contributions

CHITA has two notable technical improvements over popular methods:

Efficient use of second-order information: Pruning methods that use second-order information (i.e., relating to second derivatives) achieve the state of the art in many settings. In the literature, this information is typically used by computing the Hessian matrix or its inverse, an operation that is very difficult to scale because the Hessian size is quadratic with respect to the number of weights. Through careful reformulation, CHITA uses second-order information without having to compute or store the Hessian matrix explicitly, thus allowing for more scalability.
Combinatorial optimization: Popular optimization-based methods use a simple optimization technique that prunes weights in isolation, i.e., when deciding to prune a certain weight they don’t take into account whether other weights have been pruned. This could lead to pruning important weights because weights deemed unimportant in isolation may become important when other weights are pruned. CHITA avoids this issue by using a more advanced, combinatorial optimization algorithm that takes into account how pruning one weight impacts others.

In the sections below, we discuss CHITA’s pruning formulation and algorithms.

A computation-friendly pruning formulation

There are many possible pruning candidates, which are obtained by retaining only a subset of the weights from the original network. Let k be a user-specified parameter that denotes the number of weights to retain. Pruning can be naturally formulated as a best-subset selection (BSS) problem: among all possible pruning candidates (i.e., subsets of weights) with only k weights retained, the candidate that has the smallest loss is selected.

Pruning as a BSS problem: among all possible pruning candidates with the same total number of weights, the best candidate is defined as the one with the least loss. This illustration shows four candidates, but this number is generally much larger.

Solving the pruning BSS problem on the original loss function is generally computationally intractable. Thus, similar to previous work, such as OBD and OBS, we approximate the loss with a quadratic function by using a second-order Taylor series, where the Hessian is estimated with the empirical Fisher information matrix. While gradients can be typically computed efficiently, computing and storing the Hessian matrix is prohibitively expensive due to its sheer size. In the literature, it is common to deal with this challenge by making restrictive assumptions on the Hessian (e.g., diagonal matrix) and also on the algorithm (e.g., pruning weights in isolation).

CHITA uses an efficient reformulation of the pruning problem (BSS using the quadratic loss) that avoids explicitly computing the Hessian matrix, while still using all the information from this matrix. This is made possible by exploiting the low-rank structure of the empirical Fisher information matrix. This reformulation can be viewed as a sparse linear regression problem, where each regression coefficient corresponds to a certain weight in the neural network. After obtaining a solution to this regression problem, coefficients set to zero will correspond to weights that should be pruned. Our regression data matrix is (n x p), where n is the batch (sub-sample) size and p is the number of weights in the original network. Typically n << p, so storing and operating with this data matrix is much more scalable than common pruning approaches that operate with the (p x p) Hessian.

CHITA reformulates the quadratic loss approximation, which requires an expensive Hessian matrix, as a linear regression (LR) problem. The LR’s data matrix is linear in p, which makes the reformulation more scalable than the original quadratic approximation.

Scalable optimization algorithms

CHITA reduces pruning to a linear regression problem under the following sparsity constraint: at most k regression coefficients can be nonzero. To obtain a solution to this problem, we consider a modification of the well-known iterative hard thresholding (IHT) algorithm. IHT performs gradient descent where after each update the following post-processing step is performed: all regression coefficients outside the Top-k (i.e., the k coefficients with the largest magnitude) are set to zero. IHT typically delivers a good solution to the problem, and it does so iteratively exploring different pruning candidates and jointly optimizing over the weights.

Due to the scale of the problem, standard IHT with constant learning rate can suffer from very slow convergence. For faster convergence, we developed a new line-search method that exploits the problem structure to find a suitable learning rate, i.e., one that leads to a sufficiently large decrease in the loss. We also employed several computational schemes to improve CHITA’s efficiency and the quality of the second-order approximation, leading to an improved version that we call CHITA++.

Experiments

We compare CHITA’s run time and accuracy with several state-of-the-art pruning methods using different architectures, including ResNet and MobileNet.

Run time: CHITA is much more scalable than comparable methods that perform joint optimization (as opposed to pruning weights in isolation). For example, CHITA’s speed-up can reach over 1000x when pruning ResNet.

Post-pruning accuracy: Below, we compare the performance of CHITA and CHITA++ with magnitude pruning (MP), Woodfisher (WF), and Combinatorial Brain Surgeon (CBS), for pruning 70% of the model weights. Overall, we see good improvements from CHITA and CHITA++.

Post-pruning accuracy of various methods on ResNet20. Results are reported for pruning 70% of the model weights.

Post-pruning accuracy of various methods on MobileNet. Results are reported for pruning 70% of the model weights.

Next, we report results for pruning a larger network: ResNet50 (on this network, some of the methods listed in the ResNet20 figure couldn’t scale). Here we compare with magnitude pruning and M-FAC. The figure below shows that CHITA achieves better test accuracy for a wide range of sparsity levels.

Test accuracy of pruned networks, obtained using different methods.

Conclusion, limitations, and future work

We presented CHITA, an optimization-based approach for pruning pre-trained neural networks. CHITA offers scalability and competitive performance by efficiently using second-order information and drawing on ideas from combinatorial optimization and high-dimensional statistics.

CHITA is designed for unstructured pruning in which any weight can be removed. In theory, unstructured pruning can significantly reduce computational requirements. However, realizing these reductions in practice requires special software (and possibly hardware) that support sparse computations. In contrast, structured pruning, which removes whole structures like neurons, may offer improvements that are easier to attain on general-purpose software and hardware. It would be interesting to extend CHITA to structured pruning.

Acknowledgements

This work is part of a research collaboration between Google and MIT. Thanks to Rahul Mazumder, Natalia Ponomareva, Wenyu Chen, Xiang Meng, Zhe Zhao, and Sergei Vassilvitskii for their help in preparing this post and the paper. Also thanks to John Guilyard for creating the graphics in this post.

Source: Google AI Blog

Champion Innovator David Cardozo, based in Victoriaville, Quebec

Posted by Max Saltonstall, Developer Relations Engineer

Google Cloud Champion Innovators are a global network of more than 500 non-Google professionals, who are technical experts in Google Cloud products and services. Each Champion specializes in one of nine different technical categories: cloud AI/ML, data analytics, hybrid multi-cloud, modern architecture, security and networking, serverless app development, storage, Workspace and databases.

In our ongoing interview series we sit down with Champion Innovators across the world to learn more about their journeys, their technology focus, and what excites them.

Today we're talking to David Cardozo, a Machine Learning Scientist, Kubeflow Community member and ML GDE.

What tech area has you most fascinated right now, and why?

I love all the creative ways people are using Machine Learning (ML) to solve problems. There are a ton of cool applications that I see through my consulting work – counting cranberries from drone footage, tallying fish in fish farms, classifying plastics for recycling – and there's great stuff going on in both the public and private sector.

I'm also digging into the Kubeflow community right now, learning from that group. It's a melting pot of languages: Go, Python, etc. By participating in the working group and meetings I'm understanding so much more about current issues, blockers to progress, and get a deeper understanding of the technology itself. I love gaining that insight.

How do you like to learn new services, tools, and applications?

I read a lot: engineering blogs, books, documentation. Right now I'm learning system design from a variety of Google blogs, which helps me learn how to scale up the things I design. I'm also learning how to make ML models, and how to improve the ones I've deployed.

I'm passionate about contributing to the open source community and actively participate in various projects. Right now with friends in the community we developed Elegy – a high level API for Deep Learning in JAX.

Writing about a topic also helps me learn. Right now, I am working on blogs focused on Kubeflow pipelines in version 2.0 and Vertex AI in Google Cloud.

When I'm diving into a brand new technology I try to join the working groups that are furthering its development, so I get an inside look at how things are moving. Those working groups, their discussions and notes, teach me a ton. I also use the Google Cloud Forum and StackOverflow communities to deepen my knowledge.

What are some exciting projects you have in flight right now?

Getting to play with Generative AI within Vertex (on Google Cloud) has been very fun. I like hearing about what the other Innovators are making; it's a very smart, creative group with cool projects. Learning more about the cutting edge of ML is very exciting.

I'm doing a bit more with Open Source in my free time, trying to understand more around Kubernetes and Kubeflow.

What engages you outside of the technology world?

I stay active: swimming, lots of soccer. I also have been learning about option trading, testing out the waters of active investing. The complexity of those economic systems stimulates my curiosity. I really want to understand how it works, and how to make it useful.

My background is in the social sciences, I'm a bit of a frustrated historian. My interest in school was history, but my family said that I shouldn't focus on social science, so I majored in Math and Physics, but never finished my degree. Right now, after a few life and career pivots, I'm working on completing my Bachelor's through Coursera via the University of London, and earning a history degree requires a lot of reading. This has inspired me to make an AI project that summarizes the knowledge from very long documents, making history research more accessible by giving people a format that's easier to consume.

What brought you into the Innovators program?

I started as one of the Google Developer Experts, but I always wanted more opportunities to talk with Google engineers and get more feedback on the cloud architectures I was building, for myself or my clients. I also wanted to be more involved in the Cloud community.

When I see members of the community encountering challenges, struggling as I did, I feel the pull to help them. As a native Spanish speaker I wanted to make more content in Spanish for folks like myself. I didn't have a mentor as I was learning, and I'd like to fill that gap for others.

So I began organizing meetups in Latin America, and in Spanish speaking communities. I sought out more data scientists. And I went through Qwiklabs and Cloud Skills Boost to learn to improve my own skills.

After I joined the Innovators program, I've had the chance to play with new AI technologies, work more closely with Google experts and received credits for more Cloud experimentation.

What's one thing our readers should do next?

I recommend using some of the open, public teaching resources in Computer Science (CS), especially if you're like me and didn't focus on CS in school. For me, computers came very late to Colombia and I didn't have a chance to major in CS as a student, so I got into it via Math, then information security.

I also suggest taking a look at Elegy, and being involved in solving first issues, providing feedback and also some pull requests :)

I've liked Stanford's course on Neural Networks (CS 231n), as well as MIT's open courseware classes and ML videos on YouTube by Joel Grus.

Each Champion Innovator is not affiliated with Google nor do they offer services on behalf of Google.

Source: Google for Developers Blog - News about Web, Mobile, AI and Cloud

Machine Learning Communities: Q2 ‘23 highlights and achievements

Posted by Nari Yoon, Bitnoori Keum, Hee Jung, DevRel Community Manager / Soonson Kwon, DevRel Program Manager

Let’s explore highlights and accomplishments of vast Google Machine Learning communities over the second quarter of 2023. We are enthusiastic and grateful about all the activities by the global network of ML communities. Here are the highlights!

ML Training Campaigns Summary

More than 35 communities around the world have hosted ML Campaigns distributed by the ML Developer Programs team during the first half of the year. Thank you all for your training efforts for the entire ML community!

ML Study Jams: TFUG Bauchi, GDSC Uninter, TFUG Abidjan, MLAct, Universitas Pendidikan Indonesia, National Institute of Technology (Kosen), Kumamoto College, GDG Assiut, GDG Bassam, GDG Cloud Abidjan, GDG Antananarivo, Madan Mohan Malaviya University of Technology - Gorakhpur, Université d'Abomey-Calavi (UAC), ABES Engineering College - Ghaziabad, ABV-IIITM, Vishwakarma University - Pune, Pimpri Chinchwad College of Engineering and Research - Pune, GDG Cloud Edmonton, GDG Cocody, GDG Cloud Wilmington, University of Lay Adventist of Kigali

ML Paper Reading Clubs: GalsenAI, TFUG Dhaka, Pseudo Lab, TFUG Durg, TFUG Ibadan, Universidad Nacional de Ingeniería, GDG Karaganda, Western University, GDG Raipur, University College Dublin
ML Math Clubs: TFUG Dhaka, TFUG Hajipur, GDG Yangon, GalsenAI

Community Highlights

Keras

Image Segmentation using Composable Fully-Convolutional Networks by ML GDE Suvaditya Mukherjee (India) is a Kears.io example explaining how to implement a fully-convolutional network with a VGG-16 backend and how to use it for performing image segmentation. His presentation, KerasCV for the Young and Restless (slides | video) at TFUG Malaysia and TFUG Kolkata was an introduction to KerasCV. He discussed how basic computer vision components work, why Keras is an important tool, and how KerasCV builds on top of the established TFX and Keras ecosystem.

[ML Story] My Keras Chronicles by ML GDE Aritra Roy Gosthipaty (India) summarized his story of getting into deep learning with Keras. He included pointers as to how one could get into the open source community. Plus, his Kaggle notebook, [0.11] keras starter: unet + tf data pipeline is a starter guide for Vesuvius Challenge. He and Subvaditya also shared Keras implementation of Temporal Latent Bottleneck Networks, proposed in the paper.

KerasFuse by ML GDE Ayse Ayyuce Demirbas (Portugal) is a Python library that combines the power of TensorFlow and Keras with various computer vision techniques for medical image analysis tasks. It provides a collection of modules and functions to facilitate the development of deep learning models in TensorFlow & Keras for tasks such as image segmentation, classification, and more.

TensorFlow at Google I/O 23: A Preview of the New Features and Tools by TFUG Ibadan explored the preview of the latest features and tools in TensorFlow. They covered a wide range of topics including Dtensor, KerasCV & KerasNLP, TF quantization API, and JAX2TF.

StableDiffusion - Textual-Inversion implementation app by ML GDE Dimitre Oliveira (Brazil) is an example of how to implement code from research and fine-tunes it using the Textual Inversion process. It also provides relevant use cases for valuable tools and frameworks such as HuggingFace, Gradio, TensorFlow serving, and KerasCV.

In Understanding Gradient Descent and Building an Image Classifier in TF From Scratch, ML GDE Tanmay Bakshi (Canada) talked about how to develop a solid intuition for the fundamentals backing ML tech, and actually built a real image classification system for dogs and cats, from scratch in TF.Keras.

TensorFlow and Keras Implementation of the CVPR 2023 paper by Usha Rengaraju (India) is a research paper implementation of BiFormer: Vision Transformer with Bi-Level Routing Attention.

Smile Detection with Python, OpenCV, and Deep Learning by Rouizi Yacine is a tutorial explaining how to use deep learning to build a more robust smile detector using TensorFlow, Keras, and OpenCV.

Kaggle

Screengrab of ML Olympiad for Students - TopVistos USA

ML Olympiad for Students by GDSC UNINTER was for students and aspiring ML practitioners who want to improve their ML skills. It consisted of a challenge of predicting US working visa applications. 320+ attendees registered for the opening event, 700+ views on YouTube, 66 teams competed, and the winner got a 71% F1-score.

ICR | EDA & Baseline by ML GDE Ertuğrul Demir (Turkey) is a starter notebook for newcomers interested in the latest featured code competition on Kaggle. It got 200+ Upvotes and 490+ forks.

Screengrab of Compete More Effectively on Kaggle using Weights and Biases showing participants in the video call

Compete More Effectively on Kaggle using Weights and Biases by TFUG Hajipur was a meetup to explore techniques using Weights and Biases to improve model performance in Kaggle competitions. Usha Rengaraju (India) joined as a speaker and delivered her insights on Kaggle and strategies to win competitions. She shared tips and tricks and demonstrated how to set up a W&B account and how to integrate with Google Colab and Kaggle.

Skeleton Based Action Recognition: A failed attempt by ML GDE Ayush Thakur (India) is a discussion post about documenting his learnings from competing in the Kaggle competition, Google - Isolated Sign Language Recognition. He shared his repository, training logs, and ideas he approached in the competition. Plus, his article Keras Dense Layer: How to Use It Correctly) explored what the dense layer in Keras is and how it works in practice.

On-device ML

Add Machine Learning to your Android App by ML GDE Pankaj Rai (India) at Tech Talks for Educators was a session on on-device ML and how to add ML capabilities to Android apps such as object detection and gesture detection. He explained capabilities of ML Kit, MediaPipe, TF Lite and how to use these tools. 700+ people registered for his talk.

In MediaPipe with a bit of Bard at I/O Extended Singapore 2023, ML GDE Martin Andrews (Singapore) shared how MediaPipe fits into the ecosystem, and showed 4 different demonstrations of MediaPipe functionality: audio classification, facial landmarks, interactive segmentation, and text classification.

Adding ML to our apps with Google ML Kit and MediaPipe by ML GDE Juan Guillermo Gomez Torres (Bolivia) introduced ML Kit & MediaPipe, and the benefits of on-device ML. In Startup Academy México (Google for Startups), he shared how to increase the value for clients with ML and MediaPipe.

LLM

Introduction to Google's PaLM 2 API by ML GDE Hannes Hapke (United States) introduced how to use PaLM2 and summarized major advantages of it. His another article The role of ML Engineering in the time of GPT-4 & PaLM 2 explains the role of ML experts in finding the right balance and alignment among stakeholders to optimally navigate the opportunities and challenges posed by this emerging technology. He did presentations under the same title at North America Connect 2023 and the GDG Portland event.

Image of a cellphone with ChatBard on the display in front of a computer display with Firebase PaLM in Cloud Firestore

ChatBard : An Intelligent Customer Service Center App by ML GDE Ruqiya Bin Safi (Saudi Arabia) is an intelligent customer service center app powered by generative AI and LLMs using PaLM2 APIs.

Bard can now code and put that code in Colab for you by ML GDE Sam Witteveen (Singapore) showed how Bard makes code. He runs a Youtube channel exploring ML and AI, with playlists such as Generative AI, Paper Reviews, LLMs, and LangChain.

Google’s Bard Can Write Code by ML GDE Bhavesh Bhatt (India) shows the coding capabilities of Bard, how to create a 2048 game with it, and how to add some basic features to the game. He also uploaded videos about LangChain in a playlist and introduced Google Cloud’s new course on Generative AI in this video.

Screengrab of GDG Deep Learning Course Attention Mechanisms and Transformers led by Ruqiya Bin Safi ML GDE & WTM Ambassador, @Ru0Sa

Attention Mechanisms and Transformers by GDG Cloud Saudi talked about Attention and Transformer in NLP and ML GDE Ruqiya Bin Safi (Saudi Arabia) participated as a speaker. Another event, Hands-on with the PaLM2 API to create smart apps(Jeddah) explored what LLMs, PaLM2, and Bard are, how to use PaLM2 API, and how to create smart apps using PaLM2 API.

Hands-on with Generative AI: Google I/O Extended [Virtual] by ML GDE Henry Ruiz (United States) and Web GDE Rabimba Karanjai (United States) was a workshop on generative AI showing hands-on demons of how to get started using tools such as PaLM API, Hugging Face Transformers, and LangChain framework.

Generative AI with Google PaLM and MakerSuite by ML GDE Kuan Hoong (Malaysia) at Google I/O Extended George Town 2023 was a talk about LLMs with Google PaLM and MakerSuite. The event hosted by GDG George Town and also included ML topics such as LLMs, responsible AI, and MLOps.

Intor to Gen AI with PaLM API and MakerSuite led by GUS Luis Gustavo and Tensorflow User Group Sao Paolo

Intro to Gen AI with PaLM API and MakerSuite by TFUG São Paulo was for people who want to learn generative AI and how Google tools can help with adoption and value creation. They covered how to start prototyping Gen AI ideas with MakerSuite and how to access advanced features of PaLM2 and PaLM API. The group also hosted Opening Pandora's box: Understanding the paper that revolutionized the field of NLP (video) and ML GDE Pedro Gengo (Brazil) and ML GDE Vinicius Caridá (Brazil) shared the secret behind the famous LLM and other Gen AI models.The group members studied Attention Is All You Need paper together and learned the full potential that the technology can offer.

Language models which PaLM can speak, see, move, and understand by GDG Cloud Taipei was for those who want to understand the concept and application of PaLM. ML GED Jerry Wu (Taiwan) shared the PaLM’s main characteristics, functions, and etc.

Flow chart illustrating flexible serving structure of stable diffusion

Serving With TF and GKE: Stable Diffusion by ML GDE Chansung Park (Korea) and ML GDE Sayak Paul (India) discusses how TF Serving and Kubernetes Engine can serve a system with online deployment. They broke down Stable Diffusion into main components and how they influence the subsequent consideration for deployment. Then they also covered the deployment-specific bits such as TF Serving deployment and k8s cluster configuration.

TFX + W&B Integration by ML GDE Chansung Park (Korea) shows how KerasTuner can be used with W&B’s experiment tracking feature within the TFX Tuner component. He developed a custom TFX component to push a full-trained model to the W&B Artifact store and publish a working application on Hugging Face Space with the current version of the model. Also, his talk titled, ML Infra and High Level Framework in Google Cloud Platform, delivered what MLOps is, why it is hard, why cloud + TFX is a good starter, and how TFX is seamlessly integrated with Vertex AI and Dataflow. He shared use cases from the past projects that he and ML GDE Sayak Paul (India) have done in the last 2 years.

Open and Collaborative MLOps by ML GDE Sayak Paul (India) was a talk about why openness and collaboration are two important aspects of MLOps. He gave an overview of Hugging Face Hub and how it integrates well with TFX to promote openness and collaboration in MLOps workflows.

ML Research

Paper review: PaLM 2 Technical Report by ML GDE Grigory Sapunov (UK) looked into the details of PaLM2 and the paper. He shares reviews of papers related to Google and DeepMind through his social channels and here are some of them: Model evaluation for extreme risks (paper), Faster sorting algorithms discovered using deep reinforcement learning (paper), Power-seeking can be probable and predictive for trained agents (paper).

Learning JAX in 2023: Part 3 — A Step-by-Step Guide to Training Your First Machine Learning Model with JAX by ML GDE Aritra Roy Gosthipaty (India) and ML GDE Ritwik Raha (India) shows how JAX can train linear and nonlinear regression models and the usage of PyTrees library to train a multilayer perceptron model. In addition, at May 2023 Meetup hosted by TFUG Mumbai, they gave a talk titled Decoding End to End Object Detection with Transformers and covered the architecture of the mode and the various components that led to DETR’s inception.

20 steps to train a deployed version of the GPT model on TPU by ML GDE Jerry Wu (Taiwan) shared how to use JAX and TPU to train and infer Chinese question-answering data.

Photo of the audience from the back of the room at Developer Space @Google Singapore during Multimodal Transformers - Custom LLMs, ViTs & BLIPs

Multimodal Transformers - Custom LLMs, ViTs & BLIPs by TFUG Singapore looked at what models, systems, and techniques have come out recently related to multimodal tasks. ML GDE Sam Witteveen (Singapore) looked into various multimodal models and systems and how you can build your own with the PaLM2 Model. In June, this group invited Blaise Agüera y Arcas (VP and Fellow at Google Research) and shared the Cerebra project and the research going on at Google DeepMind including the current and future developments in generative AI and emerging trends.

TensorFlow

Training a recommendation model with dynamic embeddings by ML GDE Thushan Ganegedara (Australia) explains how to build a movie recommender model by leveraging TensorFlow Recommenders (TFRS) and TensorFlow Recommenders Addons (TFRA). The primary focus was to show how the dynamic embeddings provided in the TFRA library can be used to dynamically grow and shrink the size of the embedding tables in the recommendation setting.

Screengrab of a tweet by Mathis Hammel showcasing his talk, 'How I built the most efficient deepfake detector in the world for $100'

How I built the most efficient deepfake detector in the world for $100 by ML GDE Mathis Hammel (France) was a talk exploring a method to detect images generated via ThisPersonDoesNotExist.com and even a way to know the exact time the photo was produced. Plus, his Twitter thread, OSINT Investigation on LinkedIn, investigated a network of fake companies on LinkedIn. He used a homemade tool based on a TensorFlow model and hosted it on Google Cloud. Technical explanations of generative neural networks were also included. More than 701K people viewed this thread and it got 1200+ RTs and 3100+ Likes.

Screengrab of Few-shot learning: Creating a real-time object detection using TensorFlow and python by ML GDE Hugo Zanini

Few-shot learning: Creating a real-time object detection using TensorFlow and Python by ML GDE Hugo Zanini (Brazil) shows how to take pictures of an object using a webcam, label the images, and train a few-shot learning model to run in real-time. Also, his article, Custom YOLOv7 Object Detection with TensorFlow.js explains how he trained a custom YOLOv7 model to run it directly in the browser in real time and offline with TensorFlow.js.

The Lord of the Words Transformation of a Sequence Encoder/Decoder Attention

The Lord of the Words : The Return of the experiments with DVC (slides) by ML GDE Gema Parreno Piqueras (Spain) was a talk explaining Transformers in the neural machine learning scenario, and how to use Tensorflow and DVC. In the project, she used Tensorflow Datasets translation catalog to load data from various languages, and TensorFlow Transformers library to train several models.

Accelerate your TensorFlow models with XLA (slides) and Ship faster TensorFlow models with XLA by ML GDE Sayak Paul (India) shared how to accelerate TensorFlow models with XLA in Cloud Community Days Kolkata 2023 and Cloud Community Days Pune 2023.

Setup of NVIDIA Merlin and Tensorflow for Recommendation Models by ML GDE Rubens Zimbres (Brazil) presented a review of recommendation algorithms as well as the Two Towers algorithm, and setup of NVIDIA Merlin on premises and on Vertex AI.

Cloud

AutoML pipeline for tabular data on VertexAI in Go by ML GDE Paolo Galeone (Italy) delved into the development and deployment of tabular models using VertexAI and AutoML with Go, showcasing the actual Go code and sharing insights gained through trial & error and extensive Google research to overcome documentation limitations.

Beyond images: searching information in videos using AI (slides) by ML GDE Pedro Gengo (Brazil) and ML GDE Vinicius Caridá (Brazil) showed how to create a search engine where you can search for information in videos. They presented an architecture where they transcribe the audio and caption the frames, convert this text into embeddings, and save them in a vector DB to be able to search given a user query.

The secret sauce to creating amazing ML experiences for developers by ML GDE Gant Laborde (United States) was a podcast sharing his “aha” moment, 20 years of experience in ML, and the secret to creating enjoyable and meaningful experiences for developers.

What's inside Google’s Generative AI Studio? by ML GDE Gad Benram (Portugal) shared the preview of the new features and what you can expect from it. Additionally, in How to pitch Vertex AI in 2023, he shared the six simple and honest sales pitch points for Google Cloud representatives on how to convince customers that Vertex AI is the right platform.

In How to build a conversational AI Augmented Reality Experience with Sachin Kumar, ML GDE Sachin Kumar (Qatar) talked about how to build an AR app combining multiple technologies like Google Cloud AI, Unity, and etc. The session walked through the step-by-step process of building the app from scratch.

Machine Learning on Google Cloud Platform led by Nitin Tiwari, Google Developer Expert - Machine Learning, Software Engineer @LTMIMindtree

Machine Learning on Google Cloud Platform by ML GDE Nitin Tiwari (India) was a mentoring aiming to provide students with an in-depth understanding of the processes involved in training an ML model and deploying it using GCP. In Building robust ML solutions with TensorFlow and GCP, he shared how to leverage the capabilities of GCP and TensorFlow for ML solutions and deploy custom ML models.

Data to AI on Google cloud: Auto ML, Gen AI, and more by TFUG Prayagraj educated students on how to leverage Google Cloud’s advanced AI technologies, including AutoML and generative AI.

Source: Google for Developers Blog - News about Web, Mobile, AI and Cloud

Kubeflow joins the CNCF family

We are thrilled to announce a major milestone in the journey of the Kubeflow project. After a comprehensive review process and several months of meticulous preparation, Kubeflow has been accepted by the Cloud Native Computing Foundation (CNCF) as an incubating project. This momentous step marks a new chapter in our collaborative and open approach to accelerating machine learning (ML) in the cloud native ecosystem.

The acceptance of Kubeflow into the incubation stage by the CNCF reflects not just the project's maturity, but also its widespread adoption and expanding user base. It underscores the tremendous value of the diverse suite of components that Kubeflow provides, including Notebooks, Pipelines, Training Operators, Katib, Central Dashboard, Manifests, and many more. These tools have been instrumental in creating a cohesive, end-to-end ML platform that streamlines the development and deployment of ML workflows.

Furthermore, the alignment of Kubeflow with the CNCF acknowledges the project's foundational reliance on several existing CNCF projects such as Argo, Cert-Manager, and Istio. The joining of Kubeflow with the CNCF will serve to strengthen these existing relationships and foster greater collaboration among cloud native projects, leading to even more robust and innovative solutions for users.

Looking ahead, Google and the Kubeflow community are eager to collaborate with the CNCF on the transition process. Rest assured, our commitment to Kubeflow's ongoing development remains unwavering during this transition. We will continue to support new feature development, plan and execute upcoming releases, and strive to deliver further improvements to the Kubeflow project.

We extend our heartfelt thanks to the CNCF Technical Oversight Committee and the wider CNCF community for their support and recognition of the Kubeflow project. We look forward to this exciting new phase in our shared journey towards advancing machine learning in the cloud native landscape.

As Kubeflow continues to evolve, we invite developers, data scientists, ML engineers, and all other interested individuals to join us in shaping the future of cloud native machine learning. Let's innovate together, with Kubeflow and the CNCF, to make machine learning workflows more accessible, manageable, and scalable than ever before!

By James Liu – GCP Cloud AI

Source: Google Open Source Blog

SimPer: Simple self-supervised learning of periodic targets

Posted by Daniel McDuff, Staff Research Scientist, and Yuzhe Yang, Student Researcher, Google

Learning from periodic data (signals that repeat, such as a heart beat or the daily temperature changes on Earth’s surface) is crucial for many real-world applications, from monitoring weather systems to detecting vital signs. For example, in the environmental remote sensing domain, periodic learning is often needed to enable nowcasting of environmental changes, such as precipitation patterns or land surface temperature. In the health domain, learning from video measurement has shown to extract (quasi-)periodic vital signs such as atrial fibrillation and sleep apnea episodes.

Approaches like RepNet highlight the importance of these types of tasks, and present a solution that recognizes repetitive activities within a single video. However, these are supervised approaches that require a significant amount of data to capture repetitive activities, all labeled to indicate the number of times an action was repeated. Labeling such data is often challenging and resource-intensive, requiring researchers to manually capture gold-standard temporal measurements that are synchronized with the modality of interest (e.g., video or satellite imagery).

Alternatively, self-supervised learning (SSL) methods (e.g., SimCLR and MoCo v2), which leverage a large amount of unlabeled data to learn representations that capture periodic or quasi-periodic temporal dynamics, have demonstrated success in solving classification tasks. However, they overlook the intrinsic periodicity (i.e., the ability to identify if a frame is part of a periodic process) in data and fail to learn robust representations that capture periodic or frequency attributes. This is because periodic learning exhibits characteristics that are distinct from prevailing learning tasks.

Feature similarity is different in the context of periodic representations as compared to static features (e.g., images). For example, videos that are offset by short time delays or are reversed should be similar to the original sample, whereas videos that have been upsampled or downsampled by a factor x should be different from the original sample by a factor of x.

To address these challenges, in “SimPer: Simple Self-Supervised Learning of Periodic Targets”, published at the eleventh International Conference on Learning Representations (ICLR 2023), we introduced a self-supervised contrastive framework for learning periodic information in data. Specifically, SimPer leverages the temporal properties of periodic targets using temporal self-contrastive learning, where positive and negative samples are obtained through periodicity-invariant and periodicity-variant augmentations from the same input instance. We propose periodic feature similarity that explicitly defines how to measure similarity in the context of periodic learning. Moreover, we design a generalized contrastive loss that extends the classic InfoNCE loss to a soft regression variant that enables contrasting over continuous labels (frequency). Next, we demonstrate that SimPer effectively learns period feature representations compared to state-of-the-art SSL methods, highlighting its intriguing properties including better data efficiency, robustness to spurious correlations, and generalization to distribution shifts. Finally, we are excited to release the SimPer code repo with the research community.

The SimPer framework

SimPer introduces a temporal self-contrastive learning framework. Positive and negative samples are obtained through periodicity-invariant and periodicity-variant augmentations from the same input instance. For temporal video examples, periodicity-invariant changes are cropping, rotation or flipping, whereas periodicity-variant changes involve increasing or decreasing the speed of a video.

To explicitly define how to measure similarity in the context of periodic learning, SimPer proposes periodic feature similarity. This construction allows us to formulate training as a contrastive learning task. A model can be trained with data without any labels and then fine-tuned if necessary to map the learned features to specific frequency values.

Given an input sequence x, we know there’s an underlying associated periodic signal. We then transform x to create a series of speed or frequency altered samples, which changes the underlying periodic target, thus creating different negative views. Although the original frequency is unknown, we effectively devise pseudo- speed or frequency labels for the unlabeled input x.

Conventional similarity measures such as cosine similarity emphasize strict proximity between two feature vectors, and are sensitive to index shifted features (which represent different time stamps), reversed features, and features with changed frequencies. In contrast, periodic feature similarity should be high for samples with small temporal shifts and or reversed indexes, while capturing a continuous similarity change when the feature frequency varies. This can be achieved via a similarity metric in the frequency domain, such as the distance between two Fourier transforms.

To harness the intrinsic continuity of augmented samples in the frequency domain, SimPer designs a generalized contrastive loss that extends the classic InfoNCE loss to a soft regression variant that enables contrasting over continuous labels (frequency). This makes it suitable for regression tasks, where the goal is to recover a continuous signal, such as a heart beat.

SimPer constructs negative views of data through transformations in the frequency domain. The input sequence x has an underlying associated periodic signal. SimPer transforms x to create a series of speed or frequency altered samples, which changes the underlying periodic target, thus creating different negative views. Although the original frequency is unknown, we effectively devise pseudo speed or frequency labels for unlabeled input x (periodicity-variant augmentations τ). SimPer takes transformations that do not change the identity of the input and defines these as periodicity-invariant augmentations σ, thus creating different positive views of the sample. Then, it sends these augmented views to the encoder f, which extracts corresponding features.

Results

To evaluate SimPer's performance, we benchmarked it against state-of-the-art SSL schemes (e.g., SimCLR, MoCo v2, BYOL, CVRL) on a set of six diverse periodic learning datasets for common real-world tasks in human behavior analysis, environmental remote sensing, and healthcare. Specifically, below we present results on heart rate measurement and exercise repetition counting from video. The results show that SimPer outperforms the state-of-the-art SSL schemes across all six datasets, highlighting its superior performance in terms of data efficiency, robustness to spurious correlations, and generalization to unseen targets.

Here we show quantitative results on two representative datasets using SimPer pre-trained using various SSL methods and fine-tuned on the labeled data. First, we pre-train SimPer using the Univ. Bourgogne Franche-Comté Remote PhotoPlethysmoGraphy (UBFC) dataset, a human photoplethysmography and heart rate prediction dataset, and compare its performance to state-of-the-art SSL methods. We observe that SimPer outperforms SimCLR, MoCo v2, BYOL, and CVRL methods. The results on the human action counting dataset, Countix, further confirm the benefits of SimPer over others methods as it notably outperforms the supervised baseline. For the feature evaluation results and performance on other datasets, please refer to the paper.

Results of SimCLR, MoCo v2, BYOL, CVRL and SimPer on the Univ. Bourgogne Franche-Comté Remote PhotoPlethysmoGraphy (UBFC) and Countix datasets. Heart rate and repetition count performance is reported as mean absolute error (MAE).

Conclusion and applications

We present SimPer, a self-supervised contrastive framework for learning periodic information in data. We demonstrate that by combining a temporal self-contrastive learning framework, periodicity-invariant and periodicity-variant augmentations, and continuous periodic feature similarity, SimPer provides an intuitive and flexible approach for learning strong feature representations for periodic signals. Moreover, SimPer can be applied to various fields, ranging from environmental remote sensing to healthcare.

Acknowledgements

We would like to thank Yuzhe Yang, Xin Liu, Ming-Zher Poh, Jiang Wu, Silviu Borac, and Dina Katabi for their contributions to this work.

Source: Google AI Blog

Symbol tuning improves in-context learning in language models

Posted by Jerry Wei, Student Researcher, and Denny Zhou, Principal Scientist, Google Research

A key feature of human intelligence is that humans can learn to perform new tasks by reasoning using only a few examples. Scaling up language models has unlocked a range of new applications and paradigms in machine learning, including the ability to perform challenging reasoning tasks via in-context learning. Language models, however, are still sensitive to the way that prompts are given, indicating that they are not reasoning in a robust manner. For instance, language models often require heavy prompt engineering or phrasing tasks as instructions, and they exhibit unexpected behaviors such as performance on tasks being unaffected even when shown incorrect labels.

In “Symbol tuning improves in-context learning in language models”, we propose a simple fine-tuning procedure that we call symbol tuning, which can improve in-context learning by emphasizing input–label mappings. We experiment with symbol tuning across Flan-PaLM models and observe benefits across various settings.

Symbol tuning boosts performance on unseen in-context learning tasks and is much more robust to underspecified prompts, such as those without instructions or without natural language labels.
Symbol-tuned models are much stronger at algorithmic reasoning tasks.
Finally, symbol-tuned models show large improvements in following flipped-labels presented in-context, meaning that they are more capable of using in-context information to override prior knowledge.

An overview of symbol tuning, where models are fine-tuned on tasks where natural language labels are replaced with arbitrary symbols. Symbol tuning relies on the intuition that when instruction and relevant labels are not available, models must use in-context examples to learn the task.

Motivation

Instruction tuning is a common fine-tuning method that has been shown to improve performance and allow models to better follow in-context examples. One shortcoming, however, is that models are not forced to learn to use the examples because the task is redundantly defined in the evaluation example via instructions and natural language labels. For example, on the left in the figure above, although the examples can help the model understand the task (sentiment analysis), they are not strictly necessary since the model could ignore the examples and just read the instruction that indicates what the task is.

In symbol tuning, the model is fine-tuned on examples where the instructions are removed and natural language labels are replaced with semantically-unrelated labels (e.g., “Foo,” “Bar,” etc.). In this setup, the task is unclear without looking at the in-context examples. For example, on the right in the figure above, multiple in-context examples would be needed to figure out the task. Because symbol tuning teaches the model to reason over the in-context examples, symbol-tuned models should have better performance on tasks that require reasoning between in-context examples and their labels.

Datasets and task types used for symbol tuning.

Symbol-tuning procedure

We selected 22 publicly-available natural language processing (NLP) datasets that we use for our symbol-tuning procedure. These tasks have been widely used in the past, and we only chose classification-type tasks since our method requires discrete labels. We then remap labels to a random label from a set of ~30K arbitrary labels selected from one of three categories: integers, character combinations, and words.

For our experiments, we symbol tune Flan-PaLM, the instruction-tuned variants of PaLM. We use three different sizes of Flan-PaLM models: Flan-PaLM-8B, Flan-PaLM-62B, and Flan-PaLM-540B. We also tested Flan-cont-PaLM-62B (Flan-PaLM-62B at 1.3T tokens instead of 780B tokens), which we abbreviate as 62B-c.

We use a set of ∼300K arbitrary symbols from three categories (integers, character combinations, and words). ∼30K symbols are used during tuning and the rest are held out for evaluation.

Experimental setup

We want to evaluate a model’s ability to perform unseen tasks, so we cannot evaluate on tasks used in symbol tuning (22 datasets) or used during instruction tuning (1.8K tasks). Hence, we choose 11 NLP datasets that were not used during fine-tuning.

In-context learning

In the symbol-tuning procedure, models must learn to reason with in-context examples in order to successfully perform tasks because prompts are modified to ensure that tasks cannot simply be learned from relevant labels or instructions. Symbol-tuned models should perform better in settings where tasks are unclear and require reasoning between in-context examples and their labels. To explore these settings, we define four in-context learning settings that vary the amount of reasoning required between inputs and labels in order to learn the task (based on the availability of instructions/relevant labels)

Depending on the availability of instructions and relevant natural language labels, models may need to do varying amounts of reasoning with in-context examples. When these features are not available, models must reason with the given in-context examples to successfully perform the task.

Symbol tuning improves performance across all settings for models 62B and larger, with small improvements in settings with relevant natural language labels (+0.8% to +4.2%) and substantial improvements in settings without relevant natural language labels (+5.5% to +15.5%). Strikingly, when relevant labels are unavailable, symbol-tuned Flan-PaLM-8B outperforms FlanPaLM-62B, and symbol-tuned Flan-PaLM-62B outperforms Flan-PaLM-540B. This performance difference suggests that symbol tuning can allow much smaller models to perform as well as large models on these tasks (effectively saving ∼10X inference compute).

Large-enough symbol-tuned models are better at in-context learning than baselines, especially in settings where relevant labels are not available. Performance is shown as average model accuracy (%) across eleven tasks.

Algorithmic reasoning

We also experiment on algorithmic reasoning tasks from BIG-Bench. There are two main groups of tasks: 1) List functions — identify a transformation function (e.g., remove the last element in a list) between input and output lists containing non-negative integers; and 2) simple turing concepts — reason with binary strings to learn the concept that maps an input to an output (e.g., swapping 0s and 1s in a string).

On the list function and simple turing concept tasks, symbol tuning results in an average performance improvement of 18.2% and 15.3%, respectively. Additionally, Flan-cont-PaLM-62B with symbol tuning outperforms Flan-PaLM-540B on the list function tasks on average, which is equivalent to a ∼10x reduction in inference compute. These improvements suggest that symbol tuning strengthens the model’s ability to learn in-context for unseen task types, as symbol tuning did not include any algorithmic data.

Symbol-tuned models achieve higher performance on list function tasks and simple turing concept tasks. (A–E): categories of list functions tasks. (F): simple turing concepts task.

Flipped labels

In the flipped-label experiment, labels of in-context and evaluation examples are flipped, meaning that prior knowledge and input-label mappings disagree (e.g., sentences containing positive sentiment labeled as “negative sentiment”), thereby allowing us to study whether models can override prior knowledge. Previous work has shown that while pre-trained models (without instruction tuning) can, to some extent, follow flipped labels presented in-context, instruction tuning degraded this ability.

We see that there is a similar trend across all model sizes — symbol-tuned models are much more capable of following flipped labels than instruction-tuned models. We found that after symbol tuning, Flan-PaLM-8B sees an average improvement across all datasets of 26.5%, Flan-PaLM-62B sees an improvement of 33.7%, and Flan-PaLM-540B sees an improvement of 34.0%. Additionally, symbol-tuned models achieve similar or better than average performance as pre-training–only models.

Symbol-tuned models are much better at following flipped labels presented in-context than instruction-tuned models are.

Conclusion

We presented symbol tuning, a new method of tuning models on tasks where natural language labels are remapped to arbitrary symbols. Symbol tuning is based off of the intuition that when models cannot use instructions or relevant labels to determine a presented task, it must do so by instead learning from in-context examples. We tuned four language models using our symbol-tuning procedure, utilizing a tuning mixture of 22 datasets and approximately 30K arbitrary symbols as labels.

We first showed that symbol tuning improves performance on unseen in-context learning tasks, especially when prompts do not contain instructions or relevant labels. We also found that symbol-tuned models were much better at algorithmic reasoning tasks, despite the lack of numerical or algorithmic data in the symbol-tuning procedure. Finally, in an in-context learning setting where inputs have flipped labels, symbol tuning (for some datasets) restores the ability to follow flipped labels that was lost during instruction tuning.

Future work

Through symbol tuning, we aim to increase the degree to which models can examine and learn from input–label mappings during in-context learning. We hope that our results encourage further work towards improving language models’ ability to reason over symbols presented in-context.

Acknowledgements

The authors of this post are now part of Google DeepMind. This work was conducted by Jerry Wei, Le Hou, Andrew Lampinen, Xiangning Chen, Da Huang, Yi Tay, Xinyun Chen, Yifeng Lu, Denny Zhou, Tengyu Ma, and Quoc V. Le. We would like to thank our colleagues at Google Research and Google DeepMind for their advice and helpful discussions.