Oftentimes, machine learning (ML) model developers begin their design using a generic backbone model that is trained at scale and with capabilities transferable to a wide range of downstream tasks. In natural language processing, a number of popular backbone models, including BERT, T5, GPT-3 (sometimes also referred to as “foundation models”), are pre-trained on web-scale data and have demonstrated generic multi-tasking capabilities through zero-shot, few-shot or transfer learning. Compared with training over-specialized individual models, pre-training backbone models for a large number of downstream tasks can amortize the training costs, allowing one to overcome resource limitations when building large scale models.
In computer vision, pioneering work has shown the effectiveness of single-encoder models pre-trained for image classification to capture generic visual representations that are effective for other downstream tasks. More recently, contrastive dual-encoder (CLIP, ALIGN, Florence) and generative encoder-decoder (SimVLM) approaches trained using web-scale noisy image-text pairs have been explored. Dual-encoder models exhibit remarkable zero-shot image classification capabilities but are less effective for joint vision-language understanding. On the other hand, encoder-decoder methods are good at image captioning and visual question answering but cannot perform retrieval-style tasks.
In “CoCa: Contrastive Captioners are Image-Text Foundation Models”, we present a unified vision backbone model called Contrastive Captioner (CoCa). Our model is a novel encoder-decoder approach that simultaneously produces aligned unimodal image and text embeddings and joint multimodal representations, making it flexible enough to be directly applicable for all types of downstream tasks. Specifically, CoCa achieves state-of-the-art results on a series of vision and vision-language tasks spanning vision recognition, cross-modal alignment, and multimodal understanding. Furthermore, it learns highly generic representations so that it can perform as well or better than fully fine-tuned models with zero-shot learning or frozen encoders.
Method We propose CoCa, a unified training framework that combines contrastive loss and captioning loss on a single training data stream consisting of image annotations and noisy image-text pairs, effectively merging single-encoder, dual-encoder and encoder-decoder paradigms.
To this end, we present a novel encoder-decoder architecture where the encoder is a vision transformer (ViT), and the text decoder transformer is decoupled into two parts, a unimodal text decoder and a multimodal text decoder. We skip cross-attention in unimodal decoder layers to encode text-only representations for contrastive loss, and cascade multimodal decoder layers with cross-attention to image encoder outputs to learn multimodal image-text representations for captioning loss. This design maximizes the model's flexibility and universality in accommodating a wide spectrum of tasks, and at the same time, it can be efficiently trained with a single forward and backward propagation for both training objectives, resulting in minimal computational overhead. Thus, the model can be trained end-to-end from scratch with training costs comparable to a naïve encoder-decoder model.
Benchmark Results The CoCa model can be directly fine-tuned on many tasks with minimal adaptation. By doing so, our model achieves a series of state-of-the-art results on popular vision and multimodal benchmarks, including (1) visual recognition: ImageNet, Kinetics-400/600/700, and MiT; (2) cross-modal alignment: MS-COCO, Flickr30K, and MSR-VTT; and (3) multimodal understanding: VQA, SNLI-VE, NLVR2, and NoCaps.
It is noteworthy that CoCa attains these results as a single model adapted for all tasks while often lighter than prior top-performing specialized models. For example, CoCa obtains 91.0% ImageNet top-1 accuracy while using less than half the parameters of prior state-of-the-art models. In addition, CoCa also obtains strong generative capability of high-quality image captions.
Zero-Shot Performance Besides achieving excellent performance with fine-tuning, CoCa also outperforms previous state-of-the-art models on zero-shot learning tasks, including image classification,and cross-modal retrieval. CoCa obtains 86.3% zero-shot accuracy on ImageNet while also robustly outperforming prior models on challenging variant benchmarks, such as ImageNet-A, ImageNet-R, ImageNet-V2, and ImageNet-Sketch. As shown in the figure below, CoCa obtains better zero-shot accuracy with smaller model sizes compared to prior methods.
Frozen Encoder Representation One particularly exciting observation is that CoCa achieves results comparable to the best fine-tuned models using only a frozen visual encoder, in which features extracted after model training are used to train a classifier, rather than the more computationally intensive effort of fine-tuning a model. On ImageNet, a frozen CoCa encoder with a learned classification head obtains 90.6% top-1 accuracy, which is better than the fully fine-tuned performance of existing backbone models (90.1%). We also find this setup to work extremely well for video recognition. We feed sampled video frames into the CoCa frozen image encoder individually, and fuse output features by attentional pooling before applying a learned classifier. This simple approach using a CoCa frozen image encoder achieves video action recognition top-1 accuracy of 88.0% on Kinetics-400 dataset and demonstrates that CoCa learns a highly generic visual representation with the combined training objectives.
Conclusion We present Contrastive Captioner (CoCa), a novel pre-training paradigm for image-text backbone models. This simple method is widely applicable to many types of vision and vision-language downstream tasks, and obtains state-of-the-art performance with minimal or even no task-specific adaptations.
Acknowledgements We would like to thank our co-authors Vijay Vasudevan, Legg Yeung, Mojtaba Seyedhosseini, and Yonghui Wu who have been involved in all aspects of the project. We also would like to thank Yi-Ting Chen, Kaifeng Chen, Ye Xia, Zhen Li, Chao Jia, Yinfei Yang, Zhengdong Zhang, Wei Han, Yuan Cao, Tao Zhu, Futang Peng, Soham Ghosh, Zihang Dai, Xin Li, Anelia Angelova, Jason Baldridge, Izhak Shafran, Shengyang Dai, Abhijit Ogale, Zhifeng Chen, Claire Cui, Paul Natsev, Tom Duerig for helpful discussions, Andrew Dai for help with contrastive models, Christopher Fifty and Bowen Zhang for help with video models, Yuanzhong Xu for help with model scaling, Lucas Beyer for help with data preparation, Andy Zeng for help with MSR-VTT evaluation, Hieu Pham and Simon Kornblith for help with zero-shot evaluations, Erica Moreira and Victor Gomes for help with resource coordination, Liangliang Cao for proofreading, Tom Small for creating the animations used in this blogpost, and others in the Google Brain team for support throughout this project.
In recent years, natural language processing models have dramatically improved their ability to learn general-purpose representations, which has resulted in significant performance gains for a wide range of natural language generation and natural language understanding tasks. In large part, this has been accomplished through pre-training language models on extensive unlabeled text corpora.
This pre-training formulation does not make assumptions about input signal modality, which can be language, vision, or audio, among others. Several recent papers have exploited this formulation to dramatically improve image generation results through pre-quantizing images into discrete integer codes (represented as natural numbers), and modeling them autoregressively (i.e., predicting sequences one token at a time). In these approaches, a convolutional neural network (CNN) is trained to encode an image into discrete tokens, each corresponding to a small patch of the image. A second stage CNN or Transformer is then trained to model the distribution of encoded latent variables. The second stage can also be applied to autoregressively generate an image after the training. But while such models have achieved strong performance for image generation, few studies have evaluated the learned representation for downstream discriminative tasks (such as image classification).
In “Vector-Quantized Image Modeling with Improved VQGAN”, we propose a two-stage model that reconceives traditional image quantization techniques to yield improved performance on image generation and image understanding tasks. In the first stage, an image quantization model, called VQGAN, encodes an image into lower-dimensional discrete latent codes. Then a Transformer model is trained to model the quantized latent codes of an image. This approach, which we call Vector-quantized Image Modeling (VIM), can be used for both image generation and unsupervised image representation learning. We describe multiple improvements to the image quantizer and show that training a stronger image quantizer is a key component for improving both image generation and image understanding.
Vector-Quantized Image Modeling with ViT-VQGAN One recent, commonly used model that quantizes images into integer tokens is the Vector-quantized Variational AutoEncoder (VQVAE), a CNN-based auto-encoder whose latent space is a matrix of discrete learnable variables, trained end-to-end. VQGAN is an improved version of this that introduces an adversarial loss to promote high quality reconstruction. VQGAN uses transformer-like elements in the form of non-local attention blocks, which allows it to capture distant interactions using fewer layers.
In our work, we propose taking this approach one step further by replacing both the CNN encoder and decoder with ViT. In addition, we introduce a linear projection from the output of the encoder to a low-dimensional latent variable space for lookup of the integer tokens. Specifically, we reduced the encoder output from a 768-dimension vector to a 32- or 8-dimension vector per code, which we found encourages the decoder to better utilize the token outputs, improving model capacity and efficiency.
With our trained ViT-VQGAN, images are encoded into discrete tokens represented by integers, each of which encompasses an 8x8 patch of the input image. Using these tokens, we train a decoder-only Transformer to predict a sequence of image tokens autoregressively. This two-stage model, VIM, is able to perform unconditioned image generation by simply sampling token-by-token from the output softmax distribution of the Transformer model.
VIM is also capable of performing class-conditioned generation, such as synthesizing a specific image of a given class (e.g., a dog or a cat). We extend the unconditional generation to class-conditioned generation by prepending a class-ID token before the image tokens during both training and sampling.
To test the image understanding capabilities of VIM, we also fine-tune a linear projection layer to perform ImageNet classification, a standard benchmark for measuring image understanding abilities. Similar to ImageGPT, we take a layer output at a specific block, average over the sequence of token features (frozen) and insert a softmax layer (learnable) projecting averaged features to class logits. This allows us to capture intermediate features that provide more information useful for representation learning.
Experimental Results We train all ViT-VQGAN models with a training batch size of 256 distributed across 128 CloudTPUv4 cores. All models are trained with an input image resolution of 256x256. On top of the pre-learned ViT-VQGAN image quantizer, we train Transformer models for unconditional and class-conditioned image synthesis and compare with previous work.
We measure the performance of our proposed methods for class-conditioned image synthesis and unsupervised representation learning on the widely used ImageNet benchmark. In the table below we demonstrate the class-conditioned image synthesis performance measured by the Fréchet Inception Distance (FID). Compared to prior work, VIM improves the FID to 3.07 (lower is better), a relative improvement of 58.6% over the VQGAN model (FID 7.35). VIM also improves the capacity for image understanding, as indicated by the Inception Score (IS), which goes from 188.6 to 227.4, a 20.6% improvement relative to VQGAN.
After training a generative model, we test the learned image representations by fine-tuning a linear layer to perform ImageNet classification, a standard benchmark for measuring image understanding abilities. Our model outperforms previous generative models on the image understanding task, improving classification accuracy through linear probing (i.e., training a single linear classification layer, while keeping the rest of the model frozen) from 60.3% (iGPT-L) to 73.2%. These results showcase VIM’s strong generation results as well as image representation learning abilities.
Conclusion We propose Vector-quantized Image Modeling (VIM), which pretrains a Transformer to predict image tokens autoregressively, where discrete image tokens are produced from improved ViT-VQGAN image quantizers. With our proposed improvements on image quantization, we demonstrate superior results on both image generation and understanding. We hope our results can inspire future work towards more unified approaches for image generation and understanding.
Acknowledgements We would like to thank Xin Li, Han Zhang, Ruoming Pang, James Qin, Alexander Ku, Yuanzhong Xu, Jason Baldridge, Yonghui Wu for the preparation of the VIM paper. We thank Wei Han, Yuan Cao, Jiquan Ngiam, Vijay Vasudevan, Zhifeng Chen and Claire Cui for helpful discussions and feedback, and others on the Google Research and Brain Team for support throughout this project.
When people converse with one another, context and references play a critical role in driving their conversation more efficiently. For instance, if one asks the question “Who wrote Romeo and Juliet?” and, after receiving an answer, asks “Where was he born?”, it is clear that ‘he’ is referring to William Shakespeare without the need to explicitly mention him. Or if someone mentions “python” in a sentence, one can use the context from the conversation to determine whether they are referring to a type of snake or a computer language. If a virtual assistant cannot robustly handle context and references, users would be required to adapt to the limitation of the technology by repeating previously shared contextual information in their follow-up queries to ensure that the assistant understands their requests and can provide relevant answers.
In this post, we present a technology currently deployed on Google Assistant that allows users to speak in a natural manner when referencing context that was defined in previous queries and answers. The technology, based on the latest machine learning (ML) advances, rephrases a user’s follow-up query to explicitly mention the missing contextual information, thus enabling it to be answered as a stand-alone query. While Assistant considers many types of context for interpreting the user input, in this post we are focusing on short-term conversation history.
Context Handling by Rephrasing One of the approaches taken by Assistant to understand contextual queries is to detect if an input utterance is referring to previous context and then rephrase it internally to explicitly include the missing information. Following on from the previous example in which the user asked who wrote Romeo and Juliet, one may ask follow-up questions like “When?”. Assistant recognizes that this question is referring to both the subject (Romeo and Juliet) and answer from the previous query (William Shakespeare) and can rephrase “When?” to “When did William Shakespeare write Romeo and Juliet?”
While there are other ways to handle context, for instance, by applying rules directly to symbolic representations of the meaning of queries, like intents and arguments, the advantage of the rephrasing approach is that it operates horizontally at the string level across any query answering, parsing, or action fulfillment module.
A Wide Variety of Contextual Queries The natural language processing field, traditionally, has not put much emphasis on a general approach to context, focusing on the understanding of stand-alone queries that are fully specified. Accurately incorporating context is a challenging problem, especially when considering the large variety of contextual query types. The table below contains example conversations that illustrate query variability and some of the many contextual challenges that Assistant’s rephrasing method can resolve (e.g., differentiating between referential and non-referential cases or identifying what context a query is referencing). We demonstrate how Assistant is now able to rephrase follow-up queries, adding contextual information before providing an answer.
System Architecture At a high level, the rephrasing system generates rephrasing candidates by using different types of candidate generators. Each rephrasing candidate is then scored based on a number of signals, and the one with the highest score is selected.
Candidate Generation To generate rephrasing candidates we use a hybrid approach that applies different techniques, which we classify into three categories:
Candidate Scoring We extract a number of signals for each rephrasing candidate and use an ML model to select the most promising candidate. Some of the signals depend only on the current query and its context. For example, is the topic of the current query similar to the topic of the previous query? Or, is the current query a good stand-alone query or does it look incomplete? Other signals depend on the candidate itself: How much of the information of the context does the candidate preserve? Is the candidate well-formed from a linguistic point of view? Etc.
Recently, new signals generated by BERT and MUM models have significantly improved the performance of the ranker, fixing about one-third of the recall headroom while minimizing false positives on query sequences that are not contextual (and therefore do not require a rephrasing).
Conclusion The solution described here attempts to resolve contextual queries by rephrasing them in order to make them fully answerable in a stand-alone manner, i.e., without having to relate to other information during the fulfillment phase. The benefit of this approach is that it is agnostic to the mechanisms that would fulfill the query, thus making it usable as a horizontal layer to be deployed before any further processing.
Given the variety of contexts naturally used in human languages, we adopted a hybrid approach that combines linguistic rules, large amounts of historic data through logs, and ML models based on state-of-the-art Transformer approaches. By generating a number of rephrasing candidates for each query and its context, and then scoring and ranking them using a variety of signals, Assistant can rephrase and thus correctly interpret most contextual queries. As Assistant can handle most types of linguistic references, we are empowering users to have more natural conversations. To make such multi-turn conversations even less cumbersome, Assistant users can turn on Continued Conversation mode to enable asking follow-up queries without the need to repeat "Hey Google" between each query. We are also using this technology in other virtual assistant settings, for instance, interpreting context from something shown on a screen or playing on a speaker.
Acknowledgements This post reflects the combined work of Aliaksei Severyn, André Farias, Cheng-Chun Lee, Florian Thöle, Gabriel Carvajal, Gyorgy Gyepesi, Julien Cretin, Liana Marinescu, Martin Bölle, Patrick Siegler, Sebastian Krause, Victor Ähdel, Victoria Fossum, Vincent Zhao. We also thank Amar Subramanya, Dave Orr, Yury Pinsky for helpful discussions and support.
Economics, combinatorics, physics, and signal processing conspire to make it difficult to design, build, and operate high-quality, cost-effective wireless networks. The radio transceivers that communicate with our mobile phones, the equipment that supports them (such as power and wired networking), and the physical space they occupy are all expensive, so it’s important to be judicious in choosing sites for new transceivers. Even when the set of available sites is limited, there are exponentially many possible networks that can be built. For example, given only 50 sites, there are 250 (over a million billion) possibilities!
Further complicating things, for every location where service is needed, one must know which transceiver provides the strongest signal and how strong it is. However, the physical characteristics of radio propagation in an environment containing buildings, hills, foliage, and other clutter are incredibly complex, so accurate predictions require sophisticated, computationally-intensive models. Building all possible sites would yield the best coverage and capacity, but even if this were not prohibitively expensive, it would create unacceptable interference among nearby transceivers. Balancing these trade-offs is a core mathematical difficulty.
The goal of wireless network planning is to decide where to place new transceivers to maximize coverage and capacity while minimizing cost and interference. Building an automatic network planning system (a.k.a., auto-planner) that quickly solves national-scale problems at fine-grained resolution without compromising solution quality has been among the most important and difficult open challenges in telecom research for decades.
To address these issues, we are piloting network planning tools built using detailed geometric models derived from high-resolution geographic data, that feed into radio propagation models powered by distributed computing. This system provides fast, high-accuracy predictions of signal strength. Our optimization algorithms then intelligently sift through the exponential space of possible networks to output a small menu of candidate networks that each achieve different desirable trade-offs among cost, coverage, and interference, while ensuring enough capacity to meet demand.
Radio Propagation The propagation of radio waves near Earth's surface is complicated. Like ripples in a pond, they decay with distance traveled, but they can also penetrate, bounce off, or bend around obstacles, further weakening the signal. Computing radio wave attenuation across a real-world landscape (called path loss) is a hybrid process combining traditional physics-based calculations with learned corrections accounting for obstruction, diffraction, reflection, and scattering of the signal by clutter (e.g., trees and buildings).
We have developed a radio propagation modeling engine that leverages the same high-res geodata that powers Google Earth, Maps and Street View to map the 3D distribution of vegetation and buildings. While accounting for signal origin, frequency, broadcast strength, etc., we train signal correction models using extensive real-world measurements, which account for diverse propagation environments — from flat to hilly terrain and from dense urban to sparse rural areas.
While such hybrid approaches are common, using detailed geodata enables accurate path loss predictions below one-meter resolution. Our propagation engine provides fast point-to-point path loss calculations and scales massively via distributed computation. For instance, computing coverage for 25,000 transceivers scattered across the continental United States can be done at 4 meter resolution in only 1.5 hours, using 1000 CPU cores.
Auto-Planning Inputs Once accurate coverage estimates are available, we can use them to optimize network planning, for example, deciding where to place hundreds of new sites to maximize network quality. The auto-planning solver addresses large-scale combinatorial optimization problems such as these, using a fast, robust, scalable approach.
Formally, an auto-planning input instance contains a set of demand points (usually a square grid) where service is to be provided, a set of candidate transceiver sites, predicted signal strengths from candidate sites to demand points (supplied by the propagation model), and a cost budget. Each demand point includes a demand quantity (e.g., estimated from the population of wireless users), and each site includes a cost and capacity. Signal strengths below some threshold are omitted. Finally, the input may include an overall cost budget.
Data Summarization for Large Instances Auto-planning inputs can be huge, not just because of the number of candidate sites (tens of thousands), and demand points (billions), but also because it requires signal strengths to all demand points from all nearby candidate sites. Simple downsampling is insufficient because population density may vary widely over a given region. Therefore, we apply methods like priority sampling to shrink the data. This technique produces a low-variance, unbiased estimate of the original data, preserving an accurate view of the network traffic and interference statistics, and shrinking the input data enough that a city-size instance fits into memory on one machine.
Multi-objective Optimization via Local Search Combinatorial optimization remains a difficult task, so we created a domain-specific local search algorithm to optimize network quality. The local search algorithmic paradigm is widely applied to address computationally-hard optimization problems. Such algorithms move from one solution to another through a search space of candidate solutions by applying small local changes, stopping at a time limit or when the solution is locally optimal. To evaluate the quality of a candidate network, we combine the different objective functions into a single one, as described in the following section.
The number of local steps to reach a local optimum, number of candidate moves we evaluate per step, and time to evaluate each candidate can all be large when dealing with realistic networks. To achieve a high-quality algorithm that finishes within hours (rather than days), we must address each of these components. Fast candidate evaluation benefits greatly from dynamic data structures that maintain the mapping between each demand point and the site in the candidate solution that provides the strongest signal to it. We update this “strongest-signal” map efficiently as the candidate solution evolves during local search. The following observations help limit both the number of steps to convergence and evaluations per step.
Selecting two nearby sites is usually not ideal because they interfere. Our algorithm explicitly forbids such pairs of sites, thereby steering the search toward better solutions while greatly reducing the number of moves considered per step. We identify pairs of forbidden sites based on the demand points they cover, as measured by the weighted Jaccard index. This captures their functional proximity much better than simple geographic distance does, especially in urban or hilly areas where radio propagation is highly non-isotropic.
Breaking the local search into epochs also helps. The first epoch mostly adds sites to increase the coverage area while avoiding forbidden pairs. As we approach the cost budget, we begin a second epoch that includes swap moves between forbidden pairs to fine-tune the interference. This restriction limits the number of candidate moves per step, while focusing on those that improve interference with less change to coverage.
Outputting a Diverse Set of Good Solutions As mentioned before, auto-planning must balance three competing objectives: maximizing coverage, while minimizing interference and capacity violations, subject to a cost budget. There is no single correct tradeoff, so our algorithm delegates the final decision to the user by providing a small menu of candidate networks with different emphases. We apply a multiplier to each objective and optimize the sum. Raising the multiplier for a component guides the algorithm to emphasize it. We perform grid search over multipliers and budgets, generating a large number of solutions, filter out any that are worse than another solution along all four components (including cost), and finally select a small subset that represent different tradeoffs.
Conclusion We described our efforts to address the most vexing challenges facing telecom network operators. Using combinatorial optimization in concert with geospatial and radio propagation modeling, we built a scalable auto-planner for wireless telecommunication networks. We are actively exploring how to expand these capabilities to best meet the needs of our customers. Stay tuned!
For questions and other inquiries, please reach out to wireless-network-interest@google.com.
Acknowledgements These technological advances were enabled by the tireless work of our collaborators: Aaron Archer, Serge Barbosa Da Torre, Imad Fattouch, Danny Liberty, Pishoy Maksy, Zifei Tong, and Mat Varghese. Special thanks to Corinna Cortes, Mazin Gilbert, Rob Katcher, Michael Purdy, Bea Sebastian, Dave Vadasz, Josh Williams, and Aaron Yonas, along with Serge and especially Aaron Archer for their assistance with this blog post.
In recent years, scaling up the size of language models has been shown to be a reliable way to improve performance on a range of natural language processing (NLP) tasks. Today’s language models at the scale of 100B or more parameters achieve strong performance on tasks like sentiment analysis and machine translation, even with little or no training examples. Even the largest language models, however, can still struggle with certain multi-step reasoning tasks, such as math word problems and commonsense reasoning. How might we enable language models to perform such reasoning tasks?
In “Chain of Thought Prompting Elicits Reasoning in Large Language Models,” we explore a prompting method for improving the reasoning abilities of language models. Called chain of thought prompting, this method enables models to decompose multi-step problems into intermediate steps. With chain of thought prompting, language models of sufficient scale (~100B parameters) can solve complex reasoning problems that are not solvable with standard prompting methods.
Comparison to Standard Prompting With standard prompting (popularized by GPT-3) the model is given examples of input–output pairs (formatted as questions and answers) before being asked to predict the answer for a test-time example (shown below on the left). In chain of thought prompting (below, right), the model is prompted to produce intermediate reasoning steps before giving the final answer to a multi-step problem. The idea is that a model-generated chain of thought would mimic an intuitive thought process when working through a multi-step reasoning problem. While producing a thought process has been previously accomplished via fine-tuning, we show that such thought processes can be elicited by including a few examples of chain of thought via prompting only, which does not require a large training dataset or modifying the language model’s weights.
Chain of thought reasoning allows models to decompose complex problems into intermediate steps that are solved individually. Moreover, the language-based nature of chain of thought makes it applicable to any task that a person could solve via language. We find through empirical experiments that chain of thought prompting can improve performance on various reasoning tasks, and that successful chain of thought reasoning is an emergent property of model scale — that is, the benefits of chain of thought prompting only materialize with a sufficient number of model parameters (around 100B).
Arithmetic Reasoning One class of tasks where language models typically struggle is arithmetic reasoning (i.e., solving math word problems). Two benchmarks in arithmetic reasoning are MultiArith and GSM8K, which test the ability of language models to solve multi-step math problems similar to the one shown in the figure above. We evaluate both the LaMDA collection of language models ranging from 422M to 137B parameters, as well as the PaLM collection of language models ranging from 8B to 540B parameters. We manually compose chains of thought to include in the examples for chain of thought prompting.
For these two benchmarks, using standard prompting leads to relatively flat scaling curves: increasing the scale of the model does not substantially improve performance (shown below). However, we find that when using chain of thought prompting, increasing model scale leads to improved performance that substantially outperforms standard prompting for large model sizes.
On the GSM8K dataset of math word problems, PaLM shows remarkable performance when scaled to 540B parameters. As shown in the table below, combining chain of thought prompting with the 540B parameter PaLM model leads to new state-of-the-art performance of 58%, surpassing the prior state of the art of 55% achieved by fine-tuning GPT-3 175B on a large training set and then ranking potential solutions via a specially trained verifier. Moreover, follow-up work on self-consistency shows that the performance of chain of thought prompting can be improved further by taking the majority vote of a broad set of generated reasoning processes, which results in 74% accuracy on GSM8K.
Commonsense Reasoning In addition to arithmetic reasoning, we consider whether the language-based nature of chain of thought prompting also makes it applicable to commonsense reasoning, which involves reasoning about physical and human interactions under the presumption of general background knowledge. For these evaluations, we use the CommonsenseQA and StrategyQA benchmarks, as well as two domain-specific tasks from BIG-Bench collaboration regarding date understanding and sports understanding. Example questions are below:
As shown below, for CommonsenseQA, StrategyQA, and Date Understanding, performance improved with model scale, and employing chain of thought prompting led to additional small improvements. Chain of thought prompting had the biggest improvement on sports understanding, for which PaLM 540B’s chain of thought performance surpassed that of an unaided sports enthusiast (95% vs 84%).
Conclusions Chain of thought prompting is a simple and broadly applicable method for improving the ability of language models to perform various reasoning tasks. Through experiments on arithmetic and commonsense reasoning, we find that chain of thought prompting is an emergent property of model scale. Broadening the range of reasoning tasks that language models can perform will hopefully inspire further work on language-based approaches to reasoning.
Acknowledgements It was an honor and privilege to work with Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Quoc Le on this project.
Machine translation (MT) technology has made significant advances in recent years, as deep learning has been integrated with natural language processing (NLP). Performance on research benchmarks like WMT have soared, and translation services have improved in quality and expanded to include new languages. Nevertheless, while existing translation services cover languages spoken by the majority of people world wide, they only include around 100 languages in total, just over 1% of those actively spoken globally. Moreover, the languages that are currently represented are overwhelmingly European, largely overlooking regions of high linguistic diversity, like Africa and the Americas.
There are two key bottlenecks towards building functioning translation models for the long tail of languages. The first arises from data scarcity; digitized data for many languages is limited and can be difficult to find on the web due to quality issues with Language Identification (LangID) models. The second challenge arises from modeling limitations. MT models usually train on large amounts of parallel (translated) text, but without such data, models must learn to translate from limited amounts of monolingual text, which is a novel area of research. Both of these challenges need to be addressed for translation models to reach sufficient quality.
In “Building Machine Translation Systems for the Next Thousand Languages”, we describe how to build high-quality monolingual datasets for over a thousand languages that do not have translation datasets available and demonstrate how one can use monolingual data alone to train MT models. As part of this effort, we are expanding Google Translate to include 24 under-resourced languages. For these languages, we created monolingual datasets by developing and using specialized neural language identification models combined with novel filtering approaches. The techniques we introduce supplement massively multilingual models with a self supervised task to enable zero-resource translation. Finally, we highlight how native speakers have helped us realize this accomplishment.
Meet the Data Automatically gathering usable textual data for under-resourced languages is much more difficult than it may seem. Tasks like LangID, which work well for high-resource languages, are unsuccessful for under-resourced languages, and many publicly available datasets crawled from the web often contain more noise than usable data for the languages they attempt to support. In our early attempts to identify under-resourced languages on the web by training a standard Compact Language Detector v3 (CLD3) LangID model, we too found that the dataset was too noisy to be usable.
As an alternative, we trained a Transformer-based, semi-supervised LangID model on over 1000 languages. This model supplements the LangID task with the MAsked Sequence-to-Sequence (MASS) task to better generalize over noisy web data. MASS simply garbles the input by randomly removing sequences of tokens from it, and trains the model to predict these sequences. We applied the Transformer-based model to a dataset that had been filtered with a CLD3 model and trained to recognize clusters of similar languages.
We then applied the open sourced Term Frequency-Inverse Internet Frequency (TF-IIF) filtering to the resulting dataset to find and discard sentences that were actually in related high-resource languages, and developed a variety of language-specific filters to eliminate specific pathologies. The result of this effort was a dataset with monolingual text in over 1000 languages, of which 400 had over 100,000 sentences. We performed human evaluations on samples of 68 of these languages and found that the majority (>70%) reflected high-quality, in-language content.
Meet the Models Once we had a dataset of monolingual text in over 1000 languages, we then developed a simple yet practical approach for zero-resource translation, i.e., translation for languages with no in-language parallel text and no language-specific translation examples. Rather than limiting our model to an artificial scenario with only monolingual text, we also include all available parallel text data with millions of examples for higher resource languages to enable the model to learn the translation task. Simultaneously, we train the model to learn representations of under-resourced languages directly from monolingual text using the MASS task. In order to solve this task, the model is forced to develop a sophisticated representation of the language in question, developing a complex understanding of how words relate to other words in a sentence.
Relying on the benefits of transfer learning in massively multilingual models, we train a single giant translation model on all available data for over 1000 languages. The model trains on monolingual text for all 1138 languages and on parallel text for a subset of 112 of the higher-resourced languages.
At training time, any input the model sees has a special token indicating which language the output should be in, exactly like the standard formulation for multilingual translation. Our additional innovation is to use the same special tokens for both the monolingual MASS task and the translation task. Therefore, the token translate_to_french may indicate that the source is in English and needs to be translated to French (the translation task), or it may mean that the source is in garbled French and needs to be translated to fluent French (the MASS task). By using the same tags for both tasks, a translate_to_french tag takes on the meaning, “Produce a fluent output in French that is semantically close to the input,” regardless of whether the input is garbled in the same language or in another language entirely. From the model’s perspective, there is not much difference between the two.
Surprisingly, this simple procedure produces high quality zero-shot translations. The BLEU and ChrF scores for the resulting model are in the 10–40 and 20–60 ranges respectively, indicating mid- to high-quality translation. We observed meaningful translations even for highly inflected languages like Quechua and Kalaallisut, despite these languages being linguistically dissimilar to all other languages in the model. However, we only computed these metrics on the small subset of languages with human-translated evaluation sets. In order to understand the quality of translation for the remaining languages, we developed an evaluation metric based on round-trip translation, which allowed us to see that several hundred languages are reaching high translation quality.
To further improve quality, we use the model to generate large amounts of synthetic parallel data, filter the data based on round-trip translation (comparing a sentence translated into another language and back again), and continue training the model on this filtered synthetic data via back-translation and self-training. Finally, we fine-tune the model on a smaller subset of 30 languages and distill it into a model small enough to be served.
Contributions from Native Speakers Regular communication with native speakers of these languages was critical for our research. We collaborated with over 100 people at Google and other institutions who spoke these languages. Some volunteers helped develop specialized filters to remove out-of-language content overlooked by automatic methods, for instance Hindi mixed with Sanskrit. Others helped with transliterating between different scripts used by the languages, for instance between Meetei Mayek and Bengali, for which sufficient tools didn’t exist; and yet others helped with a gamut of tasks related to evaluation. Native speakers were also key for advising in matters of political sensitivity, like the appropriate name for the language, and the appropriate writing system to use for it. And only native speakers could answer the ultimate question: given the current quality of translation, would it be valuable to the community for Google Translate to support this language?
Closing Notes This advance is an exciting first step toward supporting more language technologies in under-resourced languages. Most importantly, we want to stress that the quality of translations produced by these models still lags far behind that of the higher-resource languages supported by Google Translate. These models are certainly a useful first tool for understanding content in under-resourced languages, but they will make mistakes and exhibit their own biases. As with any ML-driven tool, one should consider the output carefully.
The complete list of new languages added to Google Translate in this update:
Acknowledgements We would like to thank Julia Kreutzer, Orhan Firat, Daan van Esch, Aditya Siddhant, Mengmeng Niu, Pallavi Baljekar, Xavier Garcia, Wolfgang Macherey, Theresa Breiner, Vera Axelrod, Jason Riesa, Yuan Cao, Mia Xu Chen, Klaus Macherey, Maxim Krikun, Pidong Wang, Alexander Gutkin, Apurva Shah, Yanping Huang, Zhifeng Chen, Yonghui Wu, and Macduff Hughes for their contributions to the research, engineering, and leadership of this project.
We would also like to extend our deepest gratitude to the following native speakers and members of affected communities, who helped us in a wide variety of ways: Yasser Salah Eddine Bouchareb (Algerian Arabic); Mfoniso Ukwak (Anaang); Bhaskar Borthakur, Kishor Barman, Rasika Saikia, Suraj Bharech (Assamese); Ruben Hilare Quispe (Aymara); Devina Suyanto (Balinese); Allahserix Auguste Tapo, Bakary Diarrassouba, Maimouna Siby (Bambara); Mohammad Jahangir (Baluchi); Subhajit Naskar (Bengali); Animesh Pathak, Ankur Bapna, Anup Mohan, Chaitanya Joshi, Chandan Dubey, Kapil Kumar, Manish Katiyar, Mayank Srivastava, Neeharika, Saumya Pathak, Tanya Sinha, Vikas Singh (Bhojpuri); Bowen Liang, Ellie Chio, Eric Dong, Frank Tang, Jeff Pitman, John Wong, Kenneth Chang, Manish Goregaokar, Mingfei Lau, Ryan Li, Yiwen Luo (Cantonese); Monang Setyawan (Caribbean Javanese); Craig Cornelius (Cherokee); Anton Prokopyev (Chuvash); Rajat Dogra, Sid Dogra (Dogri); Mohamed Kamagate (Dyula); Chris Assigbe, Dan Ameme, Emeafa Doe, Irene Nyavor, Thierry Gnanih, Yvonne Dumor (Ewe); Abdoulaye Barry, Adama Diallo, Fauzia van der Leeuw, Ibrahima Barry (Fulfulde); Isabel Papadimitriou (Greek); Alex Rudnick (Guarani); Mohammad Khdeir (Gulf Arabic); Paul Remollata (Hiligaynon); Ankur Bapna (Hindi); Mfoniso Ukwak (Ibibio); Nze Lawson (Igbo); D.J. Abuy, Miami Cabansay (Ilocano); Archana Koul, Shashwat Razdan, Sujeet Akula (Kashmiri); Jatin Kulkarni, Salil Rajadhyaksha, Sanjeet Hegde Desai, Sharayu Shenoy, Shashank Shanbhag, Shashi Shenoy (Konkani); Ryan Michael, Terrence Taylor (Krio); Bokan Jaff, Medya Ghazizadeh, Roshna Omer Abdulrahman, Saman Vaisipour, Sarchia Khursheed (Kurdish (Sorani));Suphian Tweel (Libyan Arabic); Doudou Kisabaka (Lingala); Colleen Mallahan, John Quinn (Luganda); Cynthia Mboli (Luyia); Abhishek Kumar, Neeraj Mishra, Priyaranjan Jha, Saket Kumar, Snehal Bhilare (Maithili); Lisa Wang (Mandarin Chinese); Cibu Johny (Malayalam); Viresh Ratnakar (Marathi); Abhi Sanoujam, Gautam Thockchom, Pritam Pebam, Sam Chaomai, Shangkar Mayanglambam, Thangjam Hindustani Devi (Meiteilon (Manipuri)); Hala Ajil (Mesopotamian Arabic); Hamdanil Rasyid (Minangkabau); Elizabeth John, Remi Ralte, S Lallienkawl Gangte,Vaiphei Thatsing, Vanlalzami Vanlalzami (Mizo); George Ouais (MSA); Ahmed Kachkach, Hanaa El Azizi (Morrocan Arabic); Ujjwal Rajbhandari (Newari); Ebuka Ufere, Gabriel Fynecontry, Onome Ofoman, Titi Akinsanmi (Nigerian Pidgin); Marwa Khost Jarkas (North Levantine Arabic); Abduselam Shaltu, Ace Patterson, Adel Kassem, Mo Ali, Yonas Hambissa (Oromo); Helvia Taina, Marisol Necochea (Quechua); AbdelKarim Mardini (Saidi Arabic); Ishank Saxena, Manasa Harish, Manish Godara, Mayank Agrawal, Nitin Kashyap, Ranjani Padmanabhan, Ruchi Lohani, Shilpa Jindal, Shreevatsa Rajagopalan, Vaibhav Agarwal, Vinod Krishnan (Sanskrit); Nabil Shahid (Saraiki); Ayanda Mnyakeni (Sesotho, Sepedi); Landis Baker (Seychellois Creole); Taps Matangira (Shona); Ashraf Elsharif (Sudanese Arabic); Sakhile Dlamini (Swati); Hakim Sidahmed (Tamazight); Melvin Johnson (Tamil); Sneha Kudugunta (Telugu); Alexander Tekle, Bserat Ghebremicael, Nami Russom, Naud Ghebre (Tigrinya); Abigail Annkah, Diana Akron, Maame Ofori, Monica Opoku-Geren, Seth Duodu-baah, Yvonne Dumor (Twi); Ousmane Loum (Wolof); and Daniel Virtheim (Yiddish).
The promise of deep reinforcement learning (RL) in solving complex, high-dimensional problems autonomously has attracted much interest in areas such as robotics, game playing, and self-driving cars. However, effectively training an RL policy requires exploring a large set of robot states and actions, including many that are not safe for the robot. This is a considerable risk, for example, when training a legged robot. Because such robots are inherently unstable, there is a high likelihood of the robot falling during learning, which could cause damage.
The risk of damage can be mitigated to some extent by learning the control policy in computer simulation and then deploying it in the real world. However, this approach usually requires addressing the difficult sim-to-real gap, i.e., the policy trained in simulation can not be readily deployed in the real world for various reasons, such as sensor noise in deployment or the simulator not being realistic enough during training. Another approach to solve this issue is to directly learn or fine-tune a control policy in the real world. But again, the main challenge is to assure safety during learning.
In “Safe Reinforcement Learning for Legged Locomotion”, we introduce a safe RL framework for learning legged locomotion while satisfying safety constraints during training. Our goal is to learn locomotion skills autonomously in the real world without the robot falling during the entire learning process. Our learning framework adopts a two-policy safe RL framework: a “safe recovery policy” that recovers robots from near-unsafe states, and a “learner policy” that is optimized to perform the desired control task. The safe learning framework switches between the safe recovery policy and the learner policy to enable robots to safely acquire novel and agile motor skills.
The Proposed Framework Our goal is to ensure that during the entire learning process, the robot never falls, regardless of the learner policy being used. Similar to how a child learns to ride a bike, our approach teaches an agent a policy while using "training wheels", i.e., a safe recovery policy. We first define a set of states, which we call a “safety trigger set”, where the robot is close to violating safety constraints but can still be saved by a safe recovery policy. For example, the safety trigger set can be defined as a set of states with the height of the robots being below a certain threshold and the roll, pitch, yaw angles being too large, which is an indication of falls. When the learner policy results in the robot being within the safety trigger set (i.e., where it is likely to fall), we switch to the safe recovery policy, which drives the robot back to a safe state. We determine when to switch back to the learner policy by leveraging an approximate dynamics model of the robot to predict the future robot trajectory. For example, based on the position of the robot’s legs and the current angle of the robot based on sensors for roll, pitch, and yaw, is it likely to fall in the future? If the predicted future states are all safe, we hand the control back to the learner policy, otherwise, we keep using the safe recovery policy.
This approach ensures safety in complex systems without resorting to opaque neural networks that may be sensitive to distribution shifts in application. In addition, the learner policy is able to explore states that are near safety violations, which is useful for learning a robust policy.
Because we use “approximated” dynamics to predict the future trajectory, we also examine how much safer a robot would be if we used a much more accurate model for its dynamics. We provide a theoretical analysis of this problem and show that our approach can achieve minimal safety performance loss compared to one with a full knowledge about the system dynamics.
Legged Locomotion Tasks To demonstrate the effectiveness of the algorithm, we consider learning three different legged locomotion skills:
Implementation Details We use a hierarchical policy framework that combines RL and a traditional control approach for the learner and safe recovery policies. This framework consists of a high-level RL policy, which produces gait parameters (e.g., stepping frequency) and feet placements, and pairs it with a low-level process controller called model predictive control (MPC) that takes in these parameters and computes the desired torque for each motor in the robot. Because we do not directly command the motors’ angles, this approach provides more stable operation, streamlines the policy training due to a smaller action space, and results in a more robust policy. The input of the RL policy network includes the previous gait parameters, the height of the robot, base orientation, linear, angular velocities, and feedback to indicate whether the robot is approaching the safety trigger set. We use the same setup for each task.
We train a safe recovery policy with a reward for reaching stability as soon as possible. Furthermore, we design the safety trigger set with inspiration from capturability theory. In particular, the initial safety trigger set is defined to ensure that the robot’s feet can not fall outside of the positions from which the robot can safely recover using the safe recovery policy. We then fine-tune this set on the real robot with a random policy to prevent the robot from falling.
Real-World Experiment Results We report the real-world experimental results showing the reward learning curves and the percentage of safe recovery policy activations on the efficient gait, catwalk, and two-leg balance tasks. To ensure that the robot can learn to be safe, we add a penalty when triggering the safe recovery policy. Here, all the policies are trained from scratch, except for the two-leg balance task, which was pre-trained in simulation because it requires more training steps.
Overall, we see that on these tasks, the reward increases, and the percentage of uses of the safe recovery policy decreases over policy updates. For instance, the percentage of uses of the safe recovery policy decreases from 20% to near 0% in the efficient gait task. For the two-leg balance task, the percentage drops from near 82.5% to 67.5%, suggesting that the two-leg balance is substantially harder than the previous two tasks. Still, the policy does improve the reward. This observation implies that the learner can gradually learn the task while avoiding the need to trigger the safe recovery policy. In addition, this suggests that it is possible to design a safe trigger set and a safe recovery policy that does not impede the exploration of the policy as the performance increases.
In addition, the following video shows the learning process for the two-leg balance task, including the interplay between the learner policy and the safe recovery policy, and the reset to the initial position when an episode ends. We can see that the robot tries to catch itself when falling by putting down the lifted legs (front left and rear right) outward, creating a support polygon. After the learning episode ends, the robot walks back to the reset position automatically. This allows us to train policy autonomously and safely without human supervision.
Finally, we show the clips of learned policies. First, in the catwalk task, the distance between two sides of the legs is 0.09m, which is 40.9% smaller than the nominal distance. Second, in the two-leg balance task, the robot can maintain balance by jumping up to four times via two legs, compared to one jump from the policy pre-trained from simulation.
Conclusion We presented a safe RL framework and demonstrated how it can be used to train a robotic policy with no falls and without the need for a manual reset during the entire learning process for the efficient gait and catwalk tasks. This approach even enables training of a two-leg balance task with only four falls. The safe recovery policy is triggered only when needed, allowing the robot to more fully explore the environment. Our results suggest that learning legged locomotion skills autonomously and safely is possible in the real world, which could unlock new opportunities including offline dataset collection for robot learning.
No model is without limitation. We currently ignore the model uncertainty from the environment and non-linear dynamics in our theoretical analysis. Including these would further improve the generality of our approach. In addition, some hyper-parameters of the switching criteria are currently being heuristically tuned. It would be more efficient to automatically determine when to switch based on the learning progress. Furthermore, it would be interesting to extend this safe RL framework to other robot applications, such as robot manipulation. Finally, designing an appropriate reward when incorporating the safe recovery policy can impact learning performance. We use a penalty-based approach that obtained reasonable results in these experiments, but we plan to investigate this in future work to make further performance improvements.
Acknowledgements We would like to thank our paper co-authors: Tingnan Zhang, Linda Luu, Sehoon Ha, Jie Tan, and Wenhao Yu. We would also like to thank the team members of Robotics at Google for discussions and feedback.
