arXiv subjects

Xuezhi Wang

Publications and source records attributed to Xuezhi Wang.

At least 55 records · Page 3Linked to original sources

arXiv2023

Recitation-Augmented Language Models

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".

cs.CL↗

arXiv2023

Quantum diamond magnetometry for navigation in GNSS denied environments

In this paper, a probabilistic method for map matching localisation based on magnetometery measurement and total magnetic intensity maps is described. We show that the method is able to effectively address the challenge issues associated with map matching using geophysical maps and provides a mechanism of handling map measurement ambiguity and a way of evaluating the underlying quality. Furthermore, the effectiveness of the magnetometery map matching localisation is demonstrated using the simulation of removing position drift of an inertial navigation system, that arises in INS over a long duration, by the magnetometery aiding in the absence GNSS positioning. Simulation results using online maps verified the robustness and effectiveness of the proposed algorithm, particularly, the aiding precision will be getting better if a high sensitivity magnetometer is used.

math.OC↗

arXiv2023

Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

We explore how generating a chain of thought -- a series of intermediate reasoning steps -- significantly improves the ability of large language models to perform complex reasoning. In particular, we show how such reasoning abilities emerge naturally in sufficiently large language models via a simple method called chain of thought prompting, where a few chain of thought demonstrations are provided as exemplars in prompting. Experiments on three large language models show that chain of thought prompting improves performance on a range of arithmetic, commonsense, and symbolic reasoning tasks. The empirical gains can be striking. For instance, prompting a 540B-parameter language model with just eight chain of thought exemplars achieves state of the art accuracy on the GSM8K benchmark of math word problems, surpassing even finetuned GPT-3 with a verifier.

cs.CL↗

arXiv2022

Scaling Instruction-Finetuned Language Models

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.

cs.LG↗

arXiv2022

TEMPERA: Test-Time Prompting via Reinforcement Learning

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.

cs.CL↗

arXiv2022

Investigating Ensemble Methods for Model Robustness Improvement of Text Classifiers

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.

cs.CL↗

arXiv2022

Large Language Models Can Self-Improve

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.

cs.CL↗

arXiv2022

Language Models are Multilingual Chain-of-Thought Reasoners

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.

cs.CL↗

arXiv2022

PaLM: Scaling Language Modeling with Pathways

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.

cs.CL↗

arXiv2022

Rationale-Augmented Ensembles in Language Models

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.

cs.CL↗

arXiv2022

Identifying and Mitigating Spurious Correlations for Improving Robustness in NLP Models

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.

cs.CL↗

arXiv2022

Measure and Improve Robustness in NLP Models: A Survey

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.

cs.CL↗

arXiv2022

Gravity aided navigation using Viterbi map matching algorithm

In GNSS-denied environments, aiding a vehicle's inertial navigation system (INS) is crucial to reducing the accumulated navigation drift caused by sensor errors (e.g. bias and noise). One potential solution is to use measurements of gravity as an aiding source. The measurements are matched to a geo-referenced map of Earth's gravity in order to estimate the vehicle's position. In this paper, we propose a novel formulation of the map matching problem using a hidden Markov model (HMM). Specifically, we treat the spatial cells of the map as the hidden states of the HMM and present a Viterbi style algorithm to estimate the most likely sequence of states, i.e. most likely sequence of vehicle positions, that results in the sequence of observed gravity measurements. Using a realistic gravity map, we demonstrate the accuracy of our Viterbi map matching algorithm in a navigation scenario and illustrate its robustness compared to existing methods.

cs.RO↗

arXiv2022

Continual Sequence Generation with Adaptive Compositional Modules

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.

cs.CL↗

arXiv2022

Probabilistic Map Matching for Robust Inertial Navigation Aiding

Robust aiding of inertial navigation systems in GNSS-denied environments is critical for the removal of accumulated navigation error caused by the drift and bias inherent in inertial sensors. One way to perform such an aiding uses matching of geophysical measurements, such as gravimetry, gravity gradiometry or magnetometry, with a known geo-referenced map. Although simple in concept, this map matching procedure is challenging: the measurements themselves are noisy; their associated spatial location is uncertain; and the measurements may match multiple points within the map (i.e. non-unique solution). In this paper, we propose a probabilistic multiple hypotheses tracker to solve the map matching problem and allow robust inertial navigation aiding. Our approach addresses the problem both locally, via probabilistic data association, and temporally by incorporating the underlying platform kinematic constraints into the tracker. The map matching output is then integrated into the navigation system using an unscented Kalman filter. Additionally, we present a statistical measure of local map information density -- the map feature variability -- and use it to weight the output covariance of the proposed algorithm. The effectiveness and robustness of the proposed algorithm are demonstrated using a navigation scenario involving gravitational map matching.

cs.RO↗

arXiv2021

Improving Calibration through the Relationship with Adversarial Robustness

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.

cs.LG↗

arXiv2021

Extraction of the second and fourth radial moments of nuclear charge density from the elastic electron-nucleus scattering

The expression for the radial moments $\left\langle r^{n}\right\rangle_{c}$ of the nuclear charge density has been discussed under the plane wave Born approximation (PWBA) method recently, which is significant to investigate the nuclear surface thickness and neutron distribution radius. In this paper, we extend the studies of extracting second-order moment $\left\langle r^{2}\right\rangle_{c}$ and fourth-order moment $\left\langle r^{4}\right\rangle_{c}$ from the Coulomb form factors $|F_{C}(q)|^2$ by the distorted wave Born approximation (DWBA) at the small momentum transfer $ q $ region. Based on the relativistic mean-field (RMF) calculations, the DWBA form factors $F_{C}^{DW}(q)$ are expanded into $ q^4 $, where the corresponding charge distributions are corrected by the contributions of neutron and spin-orbit densities. In the small $ q $ region, it is found that the experimental $|F_{C}(q)|^2$ can be well reproduced by considering the contributions of the $\left\langle r^{4}\right\rangle_{c}$ at the small $ q $ region. Through further analyzing the second-order and fourth-order expansion coefficients of the $F_{C}^{DW}(q)$, the relationship between the expansion coefficients and proton number $ Z $ is obtained. By the relationship, we extract the $\left\langle r^{2}\right\rangle_{c}$ and $\left\langle r^{4}\right\rangle_{c}$ from the limited experimental data of form factors at the small $ q $ region. Within the permissible range of error, the extracted $\left\langle r^{n}\right\rangle_{c}$ are consistent with the experimental data in this paper.

nucl-th↗

arXiv2021

Nucleon momentum distribution of $^{56}\text{Fe}$ from the axially deformed relativistic mean-field model with nucleon--nucleon correlations

Nucleon momentum distribution (NMD), particularly its high-momentum components, is essential for understanding the nucleon--nucleon ($ NN $) correlations in nuclei. Herein, we develop the studies of NMD of $^{56}\text{Fe}$ from the axially deformed relativistic mean-field (RMF) model. Moreover, we introduce the effects of $ NN $ correlation into the RMF model from phenomenological models based on deuteron and nuclear matter. For the region $ k k_{\text{F}} $, the high-momentum components of the RMF model are modified by the effects of $ NN $ correlation, which agree with the experimental data. Comparing the NMD of relativistic and non-relativistic mean-field models, the relativistic effects on nuclear structures in momentum space are analyzed. Finally, by analogizing the tensor correlations in deuteron and Jastrow-type correlations in nuclear matter, the behaviors and contributions of $ NN $ correlations in $^{56}\text{Fe}$ are further analyzed, which helps clarify the effects of the tensor force on the NMD of heavy nuclei.

nucl-th↗