Xuezhi Wang

Chengrun Yang, Xuezhi Wang, Yifeng Lu et al. · deepmind

Optimization is ubiquitous. While derivative-based algorithms have been powerful tools for various problems, the absence of gradient imposes challenges on many real-world applications. In this work, we propose Optimization by PROmpting (OPRO), a simple and effective approach to leverage large language models (LLMs) as optimizers, where the optimization task is described in natural language. In each optimization step, the LLM generates new solutions from the prompt that contains previously generated solutions with their values, then the new solutions are evaluated and added to the prompt for the next optimization step. We first showcase OPRO on linear regression and traveling salesman problems, then move on to our main application in prompt optimization, where the goal is to find instructions that maximize the task accuracy. With a variety of LLMs, we demonstrate that the best prompts optimized by OPRO outperform human-designed prompts by up to 8% on GSM8K, and by up to 50% on Big-Bench Hard tasks. Code at https://github.com/google-deepmind/opro.

Freda Shi, Mirac Suzgun, Markus Freitag et al. · deepmind

We evaluate the reasoning abilities of large language models in multilingual settings. We introduce the Multilingual Grade School Math (MGSM) benchmark, by manually translating 250 grade-school math problems from the GSM8K dataset (Cobbe et al., 2021) into ten typologically diverse languages. We find that the ability to solve MGSM problems via chain-of-thought prompting emerges with increasing model scale, and that models have strikingly strong multilingual reasoning abilities, even in underrepresented languages such as Bengali and Swahili. Finally, we show that the multilingual reasoning abilities of language models extend to other tasks such as commonsense reasoning and word-in-context semantic judgment. The MGSM benchmark is publicly available at https://github.com/google-research/url-nlp.

Zhiqing Sun, Xuezhi Wang, Yi Tay et al. · cmu

We propose a new paradigm to help Large Language Models (LLMs) generate more accurate factual knowledge without retrieving from an external corpus, called RECITation-augmented gEneration (RECITE). Different from retrieval-augmented language models that retrieve relevant documents before generating the outputs, given an input, RECITE first recites one or several relevant passages from LLMs' own memory via sampling, and then produces the final answers. We show that RECITE is a powerful paradigm for knowledge-intensive NLP tasks. Specifically, we show that by utilizing recitation as the intermediate step, a recite-and-answer scheme can achieve new state-of-the-art performance in various closed-book question answering (CBQA) tasks. In experiments, we verify the effectiveness of \method~on four pre-trained models (PaLM, UL2, OPT, and Codex) and three CBQA tasks (Natural Questions, TriviaQA, and HotpotQA). Our code is available at "https://github.com/Edward-Sun/RECITE".

Hyung Won Chung, Le Hou, Shayne Longpre et al. · cmu, deepmind

Finetuning language models on a collection of datasets phrased as instructions has been shown to improve model performance and generalization to unseen tasks. In this paper we explore instruction finetuning with a particular focus on (1) scaling the number of tasks, (2) scaling the model size, and (3) finetuning on chain-of-thought data. We find that instruction finetuning with the above aspects dramatically improves performance on a variety of model classes (PaLM, T5, U-PaLM), prompting setups (zero-shot, few-shot, CoT), and evaluation benchmarks (MMLU, BBH, TyDiQA, MGSM, open-ended generation). For instance, Flan-PaLM 540B instruction-finetuned on 1.8K tasks outperforms PALM 540B by a large margin (+9.4% on average). Flan-PaLM 540B achieves state-of-the-art performance on several benchmarks, such as 75.2% on five-shot MMLU. We also publicly release Flan-T5 checkpoints, which achieve strong few-shot performance even compared to much larger models, such as PaLM 62B. Overall, instruction finetuning is a general method for improving the performance and usability of pretrained language models.

Aakanksha Chowdhery, Sharan Narang, Jacob Devlin et al. · deepmind, stanford

Large language models have been shown to achieve remarkable performance across a variety of natural language tasks using few-shot learning, which drastically reduces the number of task-specific training examples needed to adapt the model to a particular application. To further our understanding of the impact of scale on few-shot learning, we trained a 540-billion parameter, densely activated, Transformer language model, which we call Pathways Language Model PaLM. We trained PaLM on 6144 TPU v4 chips using Pathways, a new ML system which enables highly efficient training across multiple TPU Pods. We demonstrate continued benefits of scaling by achieving state-of-the-art few-shot learning results on hundreds of language understanding and generation benchmarks. On a number of these tasks, PaLM 540B achieves breakthrough performance, outperforming the finetuned state-of-the-art on a suite of multi-step reasoning tasks, and outperforming average human performance on the recently released BIG-bench benchmark. A significant number of BIG-bench tasks showed discontinuous improvements from model scale, meaning that performance steeply increased as we scaled to our largest model. PaLM also has strong capabilities in multilingual tasks and source code generation, which we demonstrate on a wide array of benchmarks. We additionally provide a comprehensive analysis on bias and toxicity, and study the extent of training data memorization with respect to model scale. Finally, we discuss the ethical considerations related to large language models and discuss potential mitigation strategies.

Yi Tay, Mostafa Dehghani, Vinh Q. Tran et al. · mit

Existing pre-trained models are generally geared towards a particular class of problems. To date, there seems to be still no consensus on what the right architecture and pre-training setup should be. This paper presents a unified framework for pre-training models that are universally effective across datasets and setups. We begin by disentangling architectural archetypes with pre-training objectives -- two concepts that are commonly conflated. Next, we present a generalized & unified perspective for self-supervision in NLP and show how different pre-training objectives can be cast as one another and how interpolating between different objectives can be effective. We then propose Mixture-of-Denoisers (MoD), a pre-training objective that combines diverse pre-training paradigms together. We furthermore introduce a notion of mode switching, wherein downstream fine-tuning is associated with specific pre-training schemes. We conduct extensive ablative experiments to compare multiple pre-training objectives and find that our method pushes the Pareto-frontier by outperforming T5 & GPT-like models across multiple diverse setups. By scaling our model up to 20B parameters, we achieve SOTA performance on 50 well-established supervised finetuning based NLP tasks. Our model also achieve strong results at in-context learning, outperforming 175B GPT-3 on zero-shot SuperGLUE and tripling the performance of T5-XXL on one-shot summarization. On 0-shot MMLU, UL2 20B outperforms T0 and T5 models. UL2 20B also works well with chain-of-thought prompting and reasoning, making it an appealing choice for research into reasoning at a small to medium scale of 20B parameters. Finally, we apply FLAN instruction tuning to the UL2 20B model, achieving MMLU and Big-Bench scores competitive to FLAN-PaLM 62B. We release Flax-based T5X checkpoints for the UL2 20B & Flan-UL2 20B.

Yanzhe Zhang, Xuezhi Wang, Diyi Yang · gatech

Continual learning is essential for real-world deployment when there is a need to quickly adapt the model to new tasks without forgetting knowledge of old tasks. Existing work on continual sequence generation either always reuses existing parameters to learn new tasks, which is vulnerable to catastrophic forgetting on dissimilar tasks, or blindly adds new parameters for every new task, which could prevent knowledge sharing between similar tasks. To get the best of both worlds, in this work, we propose continual sequence generation with adaptive compositional modules to adaptively add modules in transformer architectures and compose both old and new modules for new tasks. We also incorporate pseudo experience replay to facilitate knowledge transfer in those shared modules. Experiment results on various sequences of generation tasks show that our framework can adaptively add modules or reuse modules based on task similarity, outperforming state-of-the-art baselines in terms of both performance and parameter efficiency. We make our code public at https://github.com/GT-SALT/Adaptive-Compositional-Modules.

Albert Lu, Hongxin Zhang, Yanzhe Zhang et al. · gatech

The limits of open-ended generative models are unclear, yet increasingly important. What causes them to succeed and what causes them to fail? In this paper, we take a prompt-centric approach to analyzing and bounding the abilities of open-ended generative models. We present a generic methodology of analysis with two challenging prompt constraint types: structural and stylistic. These constraint types are categorized into a set of well-defined constraints that are analyzable by a single prompt. We then systematically create a diverse set of simple, natural, and useful prompts to robustly analyze each individual constraint. Using the GPT-3 text-davinci-002 model as a case study, we generate outputs from our collection of prompts and analyze the model's generative failures. We also show the generalizability of our proposed method on other large models like BLOOM and OPT. Our results and our in-context mitigation strategies reveal open challenges for future research. We have publicly released our code at https://github.com/SALT-NLP/Bound-Cap-LLM.

Tianjun Zhang, Xuezhi Wang, Denny Zhou et al. · berkeley

Careful prompt design is critical to the use of large language models in zero-shot or few-shot learning. As a consequence, there is a growing interest in automated methods to design optimal prompts. In this work, we propose Test-time Prompt Editing using Reinforcement learning (TEMPERA). In contrast to prior prompt generation methods, TEMPERA can efficiently leverage prior knowledge, is adaptive to different queries and provides an interpretable prompt for every query. To achieve this, we design a novel action space that allows flexible editing of the initial prompts covering a wide set of commonly-used components like instructions, few-shot exemplars, and verbalizers. The proposed method achieves significant gains compared with recent SoTA approaches like prompt tuning, AutoPrompt, and RLPrompt, across a variety of tasks including sentiment analysis, topic classification, natural language inference, and reading comprehension. Our method achieves 5.33x on average improvement in sample efficiency when compared to the traditional fine-tuning methods.

Jindong Gu, Ahmad Beirami, Xuezhi Wang et al. · deepmind, oxford

With the advent of vision-language models (VLMs) that can perform in-context and prompt-based learning, how can we design prompting approaches that robustly generalize to distribution shift and can be used on novel classes outside the support set of the prompts? In this work, we first define two types of robustness to distribution shift on VLMs, namely, robustness on base classes (the classes included in the support set of prompts) and robustness on novel classes. Then, we study the robustness of existing in-context learning and prompt learning approaches, where we find that prompt learning performs robustly on test images from base classes, while it does not generalize well on images from novel classes. We propose robust prompt learning by integrating multiple-scale image features into the prompt, which improves both types of robustness. Comprehensive experiments are conducted to study the defined robustness on six benchmarks and show the effectiveness of our proposal.

Denny Zhou, Nathanael Schärli, Le Hou et al.

Chain-of-thought prompting has demonstrated remarkable performance on various natural language reasoning tasks. However, it tends to perform poorly on tasks which requires solving problems harder than the exemplars shown in the prompts. To overcome this challenge of easy-to-hard generalization, we propose a novel prompting strategy, least-to-most prompting. The key idea in this strategy is to break down a complex problem into a series of simpler subproblems and then solve them in sequence. Solving each subproblem is facilitated by the answers to previously solved subproblems. Our experimental results on tasks related to symbolic manipulation, compositional generalization, and math reasoning reveal that least-to-most prompting is capable of generalizing to more difficult problems than those seen in the prompts. A notable finding is that when the GPT-3 code-davinci-002 model is used with least-to-most prompting, it can solve the compositional generalization benchmark SCAN in any split (including length split) with an accuracy of at least 99% using just 14 exemplars, compared to only 16% accuracy with chain-of-thought prompting. This is particularly noteworthy because neural-symbolic models in the literature that specialize in solving SCAN are trained on the entire training set containing over 15,000 examples. We have included prompts for all the tasks in the Appendix.

Xuezhi Wang, Jason Wei, Dale Schuurmans et al.

Chain-of-thought prompting combined with pre-trained large language models has achieved encouraging results on complex reasoning tasks. In this paper, we propose a new decoding strategy, self-consistency, to replace the naive greedy decoding used in chain-of-thought prompting. It first samples a diverse set of reasoning paths instead of only taking the greedy one, and then selects the most consistent answer by marginalizing out the sampled reasoning paths. Self-consistency leverages the intuition that a complex reasoning problem typically admits multiple different ways of thinking leading to its unique correct answer. Our extensive empirical evaluation shows that self-consistency boosts the performance of chain-of-thought prompting with a striking margin on a range of popular arithmetic and commonsense reasoning benchmarks, including GSM8K (+17.9%), SVAMP (+11.0%), AQuA (+12.2%), StrategyQA (+6.4%) and ARC-challenge (+3.9%).

Xinyun Chen, Renat Aksitov, Uri Alon et al. · deepmind

Self-consistency with chain-of-thought prompting (CoT) has demonstrated remarkable performance gains on various challenging tasks, by utilizing multiple reasoning paths sampled from large language models (LLMs). However, self-consistency relies on the answer extraction process to aggregate multiple solutions, which is not applicable to free-form answers. In this work, we propose Universal Self-Consistency (USC), which leverages LLMs themselves to select the most consistent answer among multiple candidates. We evaluate USC on a variety of benchmarks, including mathematical reasoning, code generation, long-context summarization, and open-ended question answering. On open-ended generation tasks where the original self-consistency method is not applicable, USC effectively utilizes multiple samples and improves the performance. For mathematical reasoning, USC matches the standard self-consistency performance without requiring the answer formats to be similar. Finally, without access to execution results, USC also matches the execution-based voting performance on code generation.

Jiaxin Huang, Shixiang Shane Gu, Le Hou et al.

Large Language Models (LLMs) have achieved excellent performances in various tasks. However, fine-tuning an LLM requires extensive supervision. Human, on the other hand, may improve their reasoning abilities by self-thinking without external inputs. In this work, we demonstrate that an LLM is also capable of self-improving with only unlabeled datasets. We use a pre-trained LLM to generate "high-confidence" rationale-augmented answers for unlabeled questions using Chain-of-Thought prompting and self-consistency, and fine-tune the LLM using those self-generated solutions as target outputs. We show that our approach improves the general reasoning ability of a 540B-parameter LLM (74.4%->82.1% on GSM8K, 78.2%->83.0% on DROP, 90.0%->94.4% on OpenBookQA, and 63.4%->67.9% on ANLI-A3) and achieves state-of-the-art-level performance, without any ground truth label. We conduct ablation studies and show that fine-tuning on reasoning is critical for self-improvement.

Xuezhi Wang, Jason Wei, Dale Schuurmans et al.

Recent research has shown that rationales, or step-by-step chains of thought, can be used to improve performance in multi-step reasoning tasks. We reconsider rationale-augmented prompting for few-shot in-context learning, where (input -> output) prompts are expanded to (input, rationale -> output) prompts. For rationale-augmented prompting we demonstrate how existing approaches, which rely on manual prompt engineering, are subject to sub-optimal rationales that may harm performance. To mitigate this brittleness, we propose a unified framework of rationale-augmented ensembles, where we identify rationale sampling in the output space as the key component to robustly improve performance. This framework is general and can easily be extended to common natural language processing tasks, even those that do not traditionally leverage intermediate steps, such as question answering, word sense disambiguation, and sentiment analysis. We demonstrate that rationale-augmented ensembles achieve more accurate and interpretable results than existing prompting approaches--including standard prompting without rationales and rationale-based chain-of-thought prompting--while simultaneously improving interpretability of model predictions through the associated rationales.

Song Li, Yongbin You, Xuezhi Wang et al. · pku

Multilingual intelligent assistants, such as ChatGPT, have recently gained popularity. To further expand the applications of multilingual artificial intelligence assistants and facilitate international communication, it is essential to enhance the performance of multilingual speech recognition, which is a crucial component of speech interaction. In this paper, we propose two simple and parameter-efficient methods: language prompt tuning and frame-level language adapter, to respectively enhance language-configurable and language-agnostic multilingual speech recognition. Additionally, we explore the feasibility of integrating these two approaches using parameter-efficient fine-tuning methods. Our experiments demonstrate significant performance improvements across seven languages using our proposed methods.

Yao Qin, Xuezhi Wang, Balaji Lakshminarayanan et al.

A wide breadth of research has devised data augmentation approaches that can improve both accuracy and generalization performance for neural networks. However, augmented data can end up being far from the clean training data and what is the appropriate label is less clear. Despite this, most existing work simply uses one-hot labels for augmented data. In this paper, we show re-using one-hot labels for highly distorted data might run the risk of adding noise and degrading accuracy and calibration. To mitigate this, we propose a generic method AutoLabel to automatically learn the confidence in the labels for augmented data, based on the transformation distance between the clean distribution and augmented distribution. AutoLabel is built on label smoothing and is guided by the calibration-performance over a hold-out validation set. We successfully apply AutoLabel to three different data augmentation techniques: the state-of-the-art RandAug, AugMix, and adversarial training. Experiments on CIFAR-10, CIFAR-100 and ImageNet show that AutoLabel significantly improves existing data augmentation techniques over models' calibration and accuracy, especially under distributional shift.

Jieyu Zhao, Xuezhi Wang, Yao Qin et al.

Large pre-trained language models have shown remarkable performance over the past few years. These models, however, sometimes learn superficial features from the dataset and cannot generalize to the distributions that are dissimilar to the training scenario. There have been several approaches proposed to reduce model's reliance on these bias features which can improve model robustness in the out-of-distribution setting. However, existing methods usually use a fixed low-capacity model to deal with various bias features, which ignore the learnability of those features. In this paper, we analyze a set of existing bias features and demonstrate there is no single model that works best for all the cases. We further show that by choosing an appropriate bias model, we can obtain a better robustness result than baselines with a more sophisticated model design.

Xuezhi Wang, Denny Zhou

In enhancing the reasoning capabilities of large language models (LLMs), prior research primarily focuses on specific prompting techniques such as few-shot or zero-shot chain-of-thought (CoT) prompting. These methods, while effective, often involve manually intensive prompt engineering. Our study takes a novel approach by asking: Can LLMs reason effectively without prompting? Our findings reveal that, intriguingly, CoT reasoning paths can be elicited from pre-trained LLMs by simply altering the \textit{decoding} process. Rather than conventional greedy decoding, we investigate the top-$k$ alternative tokens, uncovering that CoT paths are frequently inherent in these sequences. This approach not only bypasses the confounders of prompting but also allows us to assess the LLMs' \textit{intrinsic} reasoning abilities. Moreover, we observe that the presence of a CoT in the decoding path correlates with a higher confidence in the model's decoded answer. This confidence metric effectively differentiates between CoT and non-CoT paths. Extensive empirical studies on various reasoning benchmarks show that the proposed CoT-decoding effectively elicits reasoning capabilities from language models, which were previously obscured by standard greedy decoding.

Yufan Huang, Yanzhe Zhang, Jiaao Chen et al.

Continual learning has become increasingly important as it enables NLP models to constantly learn and gain knowledge over time. Previous continual learning methods are mainly designed to preserve knowledge from previous tasks, without much emphasis on how to well generalize models to new tasks. In this work, we propose an information disentanglement based regularization method for continual learning on text classification. Our proposed method first disentangles text hidden spaces into representations that are generic to all tasks and representations specific to each individual task, and further regularizes these representations differently to better constrain the knowledge required to generalize. We also introduce two simple auxiliary tasks: next sentence prediction and task-id prediction, for learning better generic and specific representation spaces. Experiments conducted on large-scale benchmarks demonstrate the effectiveness of our method in continual text classification tasks with various sequences and lengths over state-of-the-art baselines. We have publicly released our code at https://github.com/GT-SALT/IDBR.

Yongchao Zhou, Uri Alon, Xinyun Chen et al.

Length generalization, defined as the ability to extrapolate from shorter training sequences to longer test ones, is a significant challenge for language models. This issue persists even with large-scale Transformers handling relatively straightforward tasks. In this paper, we test the Transformer's ability of length generalization using the task of addition of two integers. We show that the success of length generalization is intricately linked to the data format and the type of position encoding. Using the right combination of data format and position encodings, we show for the first time that standard Transformers can extrapolate to a sequence length that is 2.5x the input length. Nevertheless, unlike in-distribution generalization, length generalization remains fragile, significantly influenced by factors like random weight initialization and training data order, leading to large variances across different random seeds.

Xinyun Chen, Ryan A. Chi, Xuezhi Wang et al.

Large language models (LLMs) have accomplished remarkable reasoning performance in various domains. However, in the domain of reasoning tasks, we discover a frailty: LLMs are surprisingly brittle to the ordering of the premises, despite the fact that such ordering does not alter the underlying task. In particular, we observe that LLMs achieve the best performance when the premise order aligns with the context required in intermediate reasoning steps. For example, in deductive reasoning tasks, presenting the premises in the same order as the ground truth proof in the prompt (as opposed to random ordering) drastically increases the model's accuracy. We first examine the effect of premise ordering on deductive reasoning on a variety of LLMs, and our evaluation shows that permuting the premise order can cause a performance drop of over 30%. In addition, we release the benchmark R-GSM, based on GSM8K, to examine the ordering effect for mathematical problem-solving, and we again observe a significant drop in accuracy, relative to the original GSM8K benchmark.

Yuqi Li, Qingqing Long, Yihang Zhou et al.

Zero-shot hashing (ZSH) has shown excellent success owing to its efficiency and generalization in large-scale retrieval scenarios. While considerable success has been achieved, there still exist urgent limitations. Existing works ignore the locality relationships of representations and attributes, which have effective transferability between seeable classes and unseeable classes. Also, the continuous-value attributes are not fully harnessed. In response, we conduct a COMprehensive Attribute Exploration for ZSH, named COMAE, which depicts the relationships from seen classes to unseen ones through three meticulously designed explorations, i.e., point-wise, pair-wise and class-wise consistency constraints. By regressing attributes from the proposed attribute prototype network, COMAE learns the local features that are relevant to the visual attributes. Then COMAE utilizes contrastive learning to comprehensively depict the context of attributes, rather than instance-independent optimization. Finally, the class-wise constraint is designed to cohesively learn the hash code, image representation, and visual attributes more effectively. Experimental results on the popular ZSH datasets demonstrate that COMAE outperforms state-of-the-art hashing techniques, especially in scenarios with a larger number of unseen label classes.

Zhijian Chen, Chuan Hu, Min Wu et al.

Emerging topics in biomedical research are continuously expanding, providing a wealth of information about genes and their function. This rapid proliferation of knowledge presents unprecedented opportunities for scientific discovery and formidable challenges for researchers striving to keep abreast of the latest advancements. One significant challenge is navigating the vast corpus of literature to extract vital gene-related information, a time-consuming and cumbersome task. To enhance the efficiency of this process, it is crucial to address several key challenges: (1) the overwhelming volume of literature, (2) the complexity of gene functions, and (3) the automated integration and generation. In response, we propose GeneSUM, a two-stage automated gene summary extractor utilizing a large language model (LLM). Our approach retrieves and eliminates redundancy of target gene literature and then fine-tunes the LLM to refine and streamline the summarization process. We conducted extensive experiments to validate the efficacy of our proposed framework. The results demonstrate that LLM significantly enhances the integration of gene-specific information, allowing more efficient decision-making in ongoing research.

Haotian Chen, Qingqing Long, Meng Xiao et al.

Scientific literature question answering is a pivotal step towards new scientific discoveries. Recently, \textit{two-stage} retrieval-augmented generated large language models (RAG-LLMs) have shown impressive advancements in this domain. Such a two-stage framework, especially the second stage (reranker), is particularly essential in the scientific domain, where subtle differences in terminology may have a greatly negative impact on the final factual-oriented or knowledge-intensive answers. Despite this significant progress, the potential and limitations of these works remain unexplored. In this work, we present a Scientific Rerank-oriented RAG Benchmark (SciRerankBench), for evaluating rerankers within RAG-LLMs systems, spanning five scientific subjects. To rigorously assess the reranker performance in terms of noise resilience, relevance disambiguation, and factual consistency, we develop three types of question-context-answer (Q-C-A) pairs, i.e., Noisy Contexts (NC), Semantically Similar but Logically Irrelevant Contexts (SSLI), and Counterfactual Contexts (CC). Through systematic evaluation of 13 widely used rerankers on five families of LLMs, we provide detailed insights into their relative strengths and limitations. To the best of our knowledge, SciRerankBench is the first benchmark specifically developed to evaluate rerankers within RAG-LLMs, which provides valuable observations and guidance for their future development.

Meng Song, Xuezhi Wang, Tanay Biradar et al.

Transformer-based methods have exhibited significant generalization ability when prompted with target-domain demonstrations or example solutions during inference. Although demonstrations, as a way of task specification, can capture rich information that may be hard to specify by language, it remains unclear what information is extracted from the demonstrations to help generalization. Moreover, assuming access to demonstrations of an unseen task is impractical or unreasonable in many real-world scenarios, especially in robotics applications. These questions motivate us to explore what the minimally sufficient prompt could be to elicit the same level of generalization ability as the demonstrations. We study this problem in the contextural RL setting which allows for quantitative measurement of generalization and is commonly adopted by meta-RL and multi-task RL benchmarks. In this setting, the training and test Markov Decision Processes (MDPs) only differ in certain properties, which we refer to as task parameters. We show that conditioning a decision transformer on these task parameters alone can enable zero-shot generalization on par with or better than its demonstration-conditioned counterpart. This suggests that task parameters are essential for the generalization and DT models are trying to recover it from the demonstration prompt. To extract the remaining generalizable information from the supervision, we introduce an additional learnable prompt which is demonstrated to further boost zero-shot generalization across a range of robotic control, manipulation, and navigation benchmark tasks.

Bailin Wang, Zi Wang, Xuezhi Wang et al.

Large language models (LLMs) can learn to perform a wide range of natural language tasks from just a handful of in-context examples. However, for generating strings from highly structured languages (e.g., semantic parsing to complex domain-specific languages), it is challenging for the LLM to generalize from just a few exemplars. We propose \emph{grammar prompting}, a simple approach to enable LLMs to use external knowledge and domain-specific constraints, expressed through a grammar in Backus--Naur Form (BNF), during in-context learning. Grammar prompting augments each demonstration example with a specialized grammar that is minimally sufficient for generating the particular output example, where the specialized grammar is a subset of the full DSL grammar. For inference, the LLM first predicts a BNF grammar given a test input, and then generates the output according to the rules of the grammar. Experiments demonstrate that grammar prompting can enable LLMs to perform competitively on a diverse set of DSL generation tasks, including semantic parsing (SMCalFlow, Overnight, GeoQuery), PDDL planning, and SMILES-based molecule generation.

Tianle Cai, Xuezhi Wang, Tengyu Ma et al.

Recent research has highlighted the potential of large language models (LLMs) to improve their problem-solving capabilities with the aid of suitable external tools. In our work, we further advance this concept by introducing a closed-loop framework, referred to as LLMs A s Tool Makers (LATM), where LLMs create their own reusable tools for problem-solving. Our approach consists of two phases: 1) tool making: an LLM acts as the tool maker that crafts tools for a set of tasks. 2) tool using: another LLM acts as the tool user, which applies the tool built by the tool maker for problem-solving. On the problem-solving server side, tool-making enables continual tool generation and caching as new requests emerge. This framework enables subsequent requests to access cached tools via their corresponding APIs, enhancing the efficiency of task resolution. Recognizing that tool-making requires more sophisticated capabilities, we assign this task to a powerful, albeit resource-intensive, model. Conversely, the simpler tool-using phase is delegated to a lightweight model. This strategic division of labor allows the once-off cost of tool-making to be spread over multiple instances of tool-using, significantly reducing average costs while maintaining strong performance. Furthermore, our method offers a functional cache through the caching and reuse of tools, which stores the functionality of a class of requests instead of the natural language responses from LLMs, thus extending the applicability of the conventional cache mechanism. We evaluate our approach across various complex reasoning tasks, including Big-Bench tasks. With GPT-4 as the tool maker and GPT-3.5 as the tool user, LATM demonstrates performance equivalent to using GPT-4 for both roles, but with a significantly reduced inference cost.

Ananth Balashankar, Xuezhi Wang, Yao Qin et al.

Counterfactual Data Augmentation (CDA) is a commonly used technique for improving robustness in natural language classifiers. However, one fundamental challenge is how to discover meaningful counterfactuals and efficiently label them, with minimal human labeling cost. Most existing methods either completely rely on human-annotated labels, an expensive process which limits the scale of counterfactual data, or implicitly assume label invariance, which may mislead the model with incorrect labels. In this paper, we present a novel framework that utilizes counterfactual generative models to generate a large number of diverse counterfactuals by actively sampling from regions of uncertainty, and then automatically label them with a learned pairwise classifier. Our key insight is that we can more correctly label the generated counterfactuals by training a pairwise classifier that interpolates the relationship between the original example and the counterfactual. We demonstrate that with a small amount of human-annotated counterfactual data (10%), we can generate a counterfactual augmentation dataset with learned labels, that provides an 18-20% improvement in robustness and a 14-21% reduction in errors on 6 out-of-domain datasets, comparable to that of a fully human-annotated counterfactual dataset for both sentiment classification and question paraphrase tasks.

Xuezhi Wang, Haohan Wang, Diyi Yang

As NLP models achieved state-of-the-art performances over benchmarks and gained wide applications, it has been increasingly important to ensure the safe deployment of these models in the real world, e.g., making sure the models are robust against unseen or challenging scenarios. Despite robustness being an increasingly studied topic, it has been separately explored in applications like vision and NLP, with various definitions, evaluation and mitigation strategies in multiple lines of research. In this paper, we aim to provide a unifying survey of how to define, measure and improve robustness in NLP. We first connect multiple definitions of robustness, then unify various lines of work on identifying robustness failures and evaluating models' robustness. Correspondingly, we present mitigation strategies that are data-driven, model-driven, and inductive-prior-based, with a more systematic view of how to effectively improve robustness in NLP models. Finally, we conclude by outlining open challenges and future directions to motivate further research in this area.

Yao Qin, Chiyuan Zhang, Ting Chen et al.

We investigate the robustness of vision transformers (ViTs) through the lens of their special patch-based architectural structure, i.e., they process an image as a sequence of image patches. We find that ViTs are surprisingly insensitive to patch-based transformations, even when the transformation largely destroys the original semantics and makes the image unrecognizable by humans. This indicates that ViTs heavily use features that survived such transformations but are generally not indicative of the semantic class to humans. Further investigations show that these features are useful but non-robust, as ViTs trained on them can achieve high in-distribution accuracy, but break down under distribution shifts. From this understanding, we ask: can training the model to rely less on these features improve ViT robustness and out-of-distribution performance? We use the images transformed with our patch-based operations as negatively augmented views and offer losses to regularize the training away from using non-robust features. This is a complementary view to existing research that mostly focuses on augmenting inputs with semantic-preserving transformations to enforce models' invariance. We show that patch-based negative augmentation consistently improves robustness of ViTs across a wide set of ImageNet based robustness benchmarks. Furthermore, we find our patch-based negative augmentation are complementary to traditional (positive) data augmentation, and together boost the performance further.

Tianlu Wang, Rohit Sridhar, Diyi Yang et al.

Recently, NLP models have achieved remarkable progress across a variety of tasks; however, they have also been criticized for being not robust. Many robustness problems can be attributed to models exploiting spurious correlations, or shortcuts between the training data and the task labels. Most existing work identifies a limited set of task-specific shortcuts via human priors or error analyses, which requires extensive expertise and efforts. In this paper, we aim to automatically identify such spurious correlations in NLP models at scale. We first leverage existing interpretability methods to extract tokens that significantly affect model's decision process from the input text. We then distinguish "genuine" tokens and "spurious" tokens by analyzing model predictions across multiple corpora and further verify them through knowledge-aware perturbations. We show that our proposed method can effectively and efficiently identify a scalable set of "shortcuts", and mitigating these leads to more robust models in multiple applications.

Flavien Prost, Pranjal Awasthi, Nick Blumm et al.

In this work we study the problem of measuring the fairness of a machine learning model under noisy information. Focusing on group fairness metrics, we investigate the particular but common situation when the evaluation requires controlling for the confounding effect of covariate variables. In a practical setting, we might not be able to jointly observe the covariate and group information, and a standard workaround is to then use proxies for one or more of these variables. Prior works have demonstrated the challenges with using a proxy for sensitive attributes, and strong independence assumptions are needed to provide guarantees on the accuracy of the noisy estimates. In contrast, in this work we study using a proxy for the covariate variable and present a theoretical analysis that aims to characterize weaker conditions under which accurate fairness evaluation is possible. Furthermore, our theory identifies potential sources of errors and decouples them into two interpretable parts $γ$ and $ε$. The first part $γ$ depends solely on the performance of the proxy such as precision and recall, whereas the second part $ε$ captures correlations between all the variables of interest. We show that in many scenarios the error in the estimates is dominated by $γ$ via a linear dependence, whereas the dependence on the correlations $ε$ only constitutes a lower order term. As a result we expand the understanding of scenarios where measuring model fairness via proxies can be an effective approach. Finally, we compare, via simulations, the theoretical upper-bounds to the distribution of simulated estimation errors and show that assuming some structure on the data, even weak, is key to significantly improve both theoretical guarantees and empirical results.

Pranjal Awasthi, Alex Beutel, Matthaeus Kleindessner et al.

Training and evaluation of fair classifiers is a challenging problem. This is partly due to the fact that most fairness metrics of interest depend on both the sensitive attribute information and label information of the data points. In many scenarios it is not possible to collect large datasets with such information. An alternate approach that is commonly used is to separately train an attribute classifier on data with sensitive attribute information, and then use it later in the ML pipeline to evaluate the bias of a given classifier. While such decoupling helps alleviate the problem of demographic scarcity, it raises several natural questions such as: how should the attribute classifier be trained?, and how should one use a given attribute classifier for accurate bias estimation? In this work we study this question from both theoretical and empirical perspectives. We first experimentally demonstrate that the test accuracy of the attribute classifier is not always correlated with its effectiveness in bias estimation for a downstream model. In order to further investigate this phenomenon, we analyze an idealized theoretical model and characterize the structure of the optimal classifier. Our analysis has surprising and counter-intuitive implications where in certain regimes one might want to distribute the error of the attribute classifier as unevenly as possible among the different subgroups. Based on our analysis we develop heuristics for both training and using attribute classifiers for bias estimation in the data scarce regime. We empirically demonstrate the effectiveness of our approach on real and simulated data.

Sirui Yao, Yoni Halpern, Nithum Thain et al.

Imagine a food recommender system -- how would we check if it is \emph{causing} and fostering unhealthy eating habits or merely reflecting users' interests? How much of a user's experience over time with a recommender is caused by the recommender system's choices and biases, and how much is based on the user's preferences and biases? Popularity bias and filter bubbles are two of the most well-studied recommender system biases, but most of the prior research has focused on understanding the system behavior in a single recommendation step. How do these biases interplay with user behavior, and what types of user experiences are created from repeated interactions? In this work, we offer a simulation framework for measuring the impact of a recommender system under different types of user behavior. Using this simulation framework, we can (a) isolate the effect of the recommender system from the user preferences, and (b) examine how the system performs not just on average for an "average user" but also the extreme experiences under atypical user behavior. As part of the simulation framework, we propose a set of evaluation metrics over the simulations to understand the recommender system's behavior. Finally, we present two empirical case studies -- one on traditional collaborative filtering in MovieLens and one on a large-scale production recommender system -- to understand how popularity bias manifests over time.

Alexander D'Amour, Katherine Heller, Dan Moldovan et al.

ML models often exhibit unexpectedly poor behavior when they are deployed in real-world domains. We identify underspecification as a key reason for these failures. An ML pipeline is underspecified when it can return many predictors with equivalently strong held-out performance in the training domain. Underspecification is common in modern ML pipelines, such as those based on deep learning. Predictors returned by underspecified pipelines are often treated as equivalent based on their training domain performance, but we show here that such predictors can behave very differently in deployment domains. This ambiguity can lead to instability and poor model behavior in practice, and is a distinct failure mode from previously identified issues arising from structural mismatch between training and deployment domains. We show that this problem appears in a wide variety of practical ML pipelines, using examples from computer vision, medical imaging, natural language processing, clinical risk prediction based on electronic health records, and medical genomics. Our results show the need to explicitly account for underspecification in modeling pipelines that are intended for real-world deployment in any domain.

Kellie Webster, Xuezhi Wang, Ian Tenney et al.

Pre-trained models have revolutionized natural language understanding. However, researchers have found they can encode artifacts undesired in many applications, such as professions correlating with one gender more than another. We explore such gendered correlations as a case study for how to address unintended correlations in pre-trained models. We define metrics and reveal that it is possible for models with similar accuracy to encode correlations at very different rates. We show how measured correlations can be reduced with general-purpose techniques, and highlight the trade offs different strategies have. With these results, we make recommendations for training robust models: (1) carefully evaluate unintended correlations, (2) be mindful of seemingly innocuous configuration differences, and (3) focus on general mitigations.

Tianlu Wang, Xuezhi Wang, Yao Qin et al.

NLP models are shown to suffer from robustness issues, i.e., a model's prediction can be easily changed under small perturbations to the input. In this work, we present a Controlled Adversarial Text Generation (CAT-Gen) model that, given an input text, generates adversarial texts through controllable attributes that are known to be invariant to task labels. For example, in order to attack a model for sentiment classification over product reviews, we can use the product categories as the controllable attribute which would not change the sentiment of the reviews. Experiments on real-world NLP datasets demonstrate that our method can generate more diverse and fluent adversarial texts, compared to many existing adversarial text generation approaches. We further use our generated adversarial examples to improve models through adversarial training, and we demonstrate that our generated attacks are more robust against model re-training and different model architectures.

Yao Qin, Xuezhi Wang, Alex Beutel et al.

Neural networks lack adversarial robustness, i.e., they are vulnerable to adversarial examples that through small perturbations to inputs cause incorrect predictions. Further, trust is undermined when models give miscalibrated predictions, i.e., the predicted probability is not a good indicator of how much we should trust our model. In this paper, we study the connection between adversarial robustness and calibration and find that the inputs for which the model is sensitive to small perturbations (are easily attacked) are more likely to have poorly calibrated predictions. Based on this insight, we examine if calibration can be improved by addressing those adversarially unrobust inputs. To this end, we propose Adversarial Robustness based Adaptive Label Smoothing (AR-AdaLS) that integrates the correlations of adversarial robustness and calibration into training by adaptively softening labels for an example based on how easily it can be attacked by an adversary. We find that our method, taking the adversarial robustness of the in-distribution data into consideration, leads to better calibration over the model even under distributional shifts. In addition, AR-AdaLS can also be applied to an ensemble model to further improve model calibration.

Preethi Lahoti, Alex Beutel, Jilin Chen et al.

Much of the previous machine learning (ML) fairness literature assumes that protected features such as race and sex are present in the dataset, and relies upon them to mitigate fairness concerns. However, in practice factors like privacy and regulation often preclude the collection of protected features, or their use for training or inference, severely limiting the applicability of traditional fairness research. Therefore we ask: How can we train an ML model to improve fairness when we do not even know the protected group memberships? In this work we address this problem by proposing Adversarially Reweighted Learning (ARL). In particular, we hypothesize that non-protected features and task labels are valuable for identifying fairness issues, and can be used to co-train an adversarial reweighting approach for improving fairness. Our results show that {ARL} improves Rawlsian Max-Min fairness, with notable AUC improvements for worst-case protected groups in multiple datasets, outperforming state-of-the-art alternatives.

Xuezhi Wang, Nithum Thain, Anu Sinha et al.

How can we build recommender systems to take into account fairness? Real-world recommender systems are often composed of multiple models, built by multiple teams. However, most research on fairness focuses on improving fairness in a single model. Further, recent research on classification fairness has shown that combining multiple "fair" classifiers can still result in an "unfair" classification system. This presents a significant challenge: how do we understand and improve fairness in recommender systems composed of multiple components? In this paper, we study the compositionality of recommender fairness. We consider two recently proposed fairness ranking metrics: equality of exposure and pairwise ranking accuracy. While we show that fairness in recommendation is not guaranteed to compose, we provide theory for a set of conditions under which fairness of individual models does compose. We then present an analytical framework for both understanding whether a real system's signals can achieve compositional fairness, and improving which component would have the greatest impact on the fairness of the overall system. In addition to the theoretical results, we find on multiple datasets -- including a large-scale real-world recommender system -- that the overall system's end-to-end fairness is largely achievable by improving fairness in individual components.

Candice Schumann, Xuezhi Wang, Alex Beutel et al.

If our models are used in new or unexpected cases, do we know if they will make fair predictions? Previously, researchers developed ways to debias a model for a single problem domain. However, this is often not how models are trained and used in practice. For example, labels and demographics (sensitive attributes) are often hard to observe, resulting in auxiliary or synthetic data to be used for training, and proxies of the sensitive attribute to be used for evaluation of fairness. A model trained for one setting may be picked up and used in many others, particularly as is common with pre-training and cloud APIs. Despite the pervasiveness of these complexities, remarkably little work in the fairness literature has theoretically examined these issues. We frame all of these settings as domain adaptation problems: how can we use what we have learned in a source domain to debias in a new target domain, without directly debiasing on the target domain as if it is a completely new problem? We offer new theoretical guarantees of improving fairness across domains, and offer a modeling approach to transfer to data-sparse target domains. We give empirical results validating the theory and showing that these modeling approaches can improve fairness metrics with less data.

Saehan Jo, Immanuel Trummer, Weicheng Yu et al.

We present a novel natural language query interface, the AggChecker, aimed at text summaries of relational data sets. The tool focuses on natural language claims that translate into an SQL query and a claimed query result. Similar in spirit to a spell checker, the AggChecker marks up text passages that seem to be inconsistent with the actual data. At the heart of the system is a probabilistic model that reasons about the input document in a holistic fashion. Based on claim keywords and the document structure, it maps each text claim to a probability distribution over associated query translations. By efficiently executing tens to hundreds of thousands of candidate translations for a typical input document, the system maps text claims to correctness probabilities. This process becomes practical via a specialized processing backend, avoiding redundant work via query merging and result caching. Verification is an interactive process in which users are shown tentative results, enabling them to take corrective actions if necessary. Our system was tested on a set of 53 public articles containing 392 claims. Our test cases include articles from major newspapers, summaries of survey results, and Wikipedia articles. Our tool revealed erroneous claims in roughly a third of test cases. A detailed user study shows that users using our tool are in average six times faster at checking text summaries, compared to generic SQL interfaces. In fully automated verification, our tool achieves significantly higher recall and precision than baselines from the areas of natural language query interfaces and fact-checking.

Dekang Zhu, Dan P. Guralnik, Xuezhi Wang et al.

We derive a statistical model for estimation of a dendrogram from single linkage hierarchical clustering (SLHC) that takes account of uncertainty through noise or corruption in the measurements of separation of data. Our focus is on just the estimation of the hierarchy of partitions afforded by the dendrogram, rather than the heights in the latter. The concept of estimating this "dendrogram structure'' is introduced, and an approximate maximum likelihood estimator (MLE) for the dendrogram structure is described. These ideas are illustrated by a simple Monte Carlo simulation that, at least for small data sets, suggests the method outperforms SLHC in the presence of noise.

Dekang Zhu, Dan P. Guralnik, Xuezhi Wang et al.

Distance-based hierarchical clustering (HC) methods are widely used in unsupervised data analysis but few authors take account of uncertainty in the distance data. We incorporate a statistical model of the uncertainty through corruption or noise in the pairwise distances and investigate the problem of estimating the HC as unknown parameters from measurements. Specifically, we focus on single linkage hierarchical clustering (SLHC) and study its geometry. We prove that under fairly reasonable conditions on the probability distribution governing measurements, SLHC is equivalent to maximum partial profile likelihood estimation (MPPLE) with some of the information contained in the data ignored. At the same time, we show that direct evaluation of SLHC on maximum likelihood estimation (MLE) of pairwise distances yields a consistent estimator. Consequently, a full MLE is expected to perform better than SLHC in getting the correct HC results for the ground truth metric.