Jason Wei

Melody Y. Guan, Manas Joglekar, Eric Wallace et al.

As large-scale language models increasingly impact safety-critical domains, ensuring their reliable adherence to well-defined principles remains a fundamental challenge. We introduce Deliberative Alignment, a new paradigm that directly teaches the model safety specifications and trains it to explicitly recall and accurately reason over the specifications before answering. We used this approach to align OpenAI's o-series models, and achieved highly precise adherence to OpenAI's safety policies, without requiring human-written chain-of-thoughts or answers. Deliberative Alignment pushes the Pareto frontier by simultaneously increasing robustness to jailbreaks while decreasing overrefusal rates, and also improves out-of-distribution generalization. We demonstrate that reasoning over explicitly specified policies enables more scalable, trustworthy, and interpretable alignment.

Steven Y. Feng, Varun Gangal, Jason Wei et al.

Data augmentation has recently seen increased interest in NLP due to more work in low-resource domains, new tasks, and the popularity of large-scale neural networks that require large amounts of training data. Despite this recent upsurge, this area is still relatively underexplored, perhaps due to the challenges posed by the discrete nature of language data. In this paper, we present a comprehensive and unifying survey of data augmentation for NLP by summarizing the literature in a structured manner. We first introduce and motivate data augmentation for NLP, and then discuss major methodologically representative approaches. Next, we highlight techniques that are used for popular NLP applications and tasks. We conclude by outlining current challenges and directions for future research. Overall, our paper aims to clarify the landscape of existing literature in data augmentation for NLP and motivate additional work in this area. We also present a GitHub repository with a paper list that will be continuously updated at https://github.com/styfeng/DataAug4NLP

Jerry Wei, Chengyu Huang, Soroush Vosoughi et al.

We present COVID-Q, a set of 1,690 questions about COVID-19 from 13 sources, which we annotate into 15 question categories and 207 question clusters. The most common questions in our dataset asked about transmission, prevention, and societal effects of COVID, and we found that many questions that appeared in multiple sources were not answered by any FAQ websites of reputable organizations such as the CDC and FDA. We post our dataset publicly at https://github.com/JerryWeiAI/COVID-Q. For classifying questions into 15 categories, a BERT baseline scored 58.1% accuracy when trained on 20 examples per category, and for a question clustering task, a BERT + triplet loss baseline achieved 49.5% accuracy. We hope COVID-Q can help either for direct use in developing applied systems or as a domain-specific resource for model evaluation.

Jason W. Wei, Laura J. Tafe, Yevgeniy A. Linnik et al.

Classification of histologic patterns in lung adenocarcinoma is critical for determining tumor grade and treatment for patients. However, this task is often challenging due to the heterogeneous nature of lung adenocarcinoma and the subjective criteria for evaluation. In this study, we propose a deep learning model that automatically classifies the histologic patterns of lung adenocarcinoma on surgical resection slides. Our model uses a convolutional neural network to identify regions of neoplastic cells, then aggregates those classifications to infer predominant and minor histologic patterns for any given whole-slide image. We evaluated our model on an independent set of 143 whole-slide images. It achieved a kappa score of 0.525 and an agreement of 66.6% with three pathologists for classifying the predominant patterns, slightly higher than the inter-pathologist kappa score of 0.485 and agreement of 62.7% on this test set. All evaluation metrics for our model and the three pathologists were within 95% confidence intervals of agreement. If confirmed in clinical practice, our model can assist pathologists in improving classification of lung adenocarcinoma patterns by automatically pre-screening and highlighting cancerous regions prior to review. Our approach can be generalized to any whole-slide image classification task, and code is made publicly available at https://github.com/BMIRDS/deepslide.

Jason Wei, Dan Garrette, Tal Linzen et al.

Pre-trained language models perform well on a variety of linguistic tasks that require symbolic reasoning, raising the question of whether such models implicitly represent abstract symbols and rules. We investigate this question using the case study of BERT's performance on English subject-verb agreement. Unlike prior work, we train multiple instances of BERT from scratch, allowing us to perform a series of controlled interventions at pre-training time. We show that BERT often generalizes well to subject-verb pairs that never occurred in training, suggesting a degree of rule-governed behavior. We also find, however, that performance is heavily influenced by word frequency, with experiments showing that both the absolute frequency of a verb form, as well as the frequency relative to the alternate inflection, are causally implicated in the predictions BERT makes at inference time. Closer analysis of these frequency effects reveals that BERT's behavior is consistent with a system that correctly applies the SVA rule in general but struggles to overcome strong training priors and to estimate agreement features (singular vs. plural) on infrequent lexical items.

Jason Wei

This paper asks whether extrapolating the hidden space distribution of text examples from one class onto another is a valid inductive bias for data augmentation. To operationalize this question, I propose a simple data augmentation protocol called "good-enough example extrapolation" (GE3). GE3 is lightweight and has no hyperparameters. Applied to three text classification datasets for various data imbalance scenarios, GE3 improves performance more than upsampling and other hidden-space data augmentation methods.

Jason Wei, Maarten Bosma, Vincent Y. Zhao et al.

This paper explores a simple method for improving the zero-shot learning abilities of language models. We show that instruction tuning -- finetuning language models on a collection of tasks described via instructions -- substantially improves zero-shot performance on unseen tasks. We take a 137B parameter pretrained language model and instruction-tune it on over 60 NLP tasks verbalized via natural language instruction templates. We evaluate this instruction-tuned model, which we call FLAN, on unseen task types. FLAN substantially improves the performance of its unmodified counterpart and surpasses zero-shot 175B GPT-3 on 20 of 25 tasks that we evaluate. FLAN even outperforms few-shot GPT-3 by a large margin on ANLI, RTE, BoolQ, AI2-ARC, OpenbookQA, and StoryCloze. Ablation studies reveal that number of finetuning datasets, model scale, and natural language instructions are key to the success of instruction tuning.

Ruibo Liu, Jason Wei, Soroush Vosoughi

Although automated metrics are commonly used to evaluate NLG systems, they often correlate poorly with human judgements. Newer metrics such as BERTScore have addressed many weaknesses in prior metrics such as BLEU and ROUGE, which rely on n-gram matching. These newer methods, however, are still limited in that they do not consider the generation context, so they cannot properly reward generated text that is correct but deviates from the given reference. In this paper, we propose Language Model Augmented Relevance Score (MARS), a new context-aware metric for NLG evaluation. MARS leverages off-the-shelf language models, guided by reinforcement learning, to create augmented references that consider both the generation context and available human references, which are then used as additional references to score generated text. Compared with seven existing metrics in three common NLG tasks, MARS not only achieves higher correlation with human reference judgements, but also differentiates well-formed candidates from adversarial samples to a larger degree.

Thibault Sellam, Steve Yadlowsky, Jason Wei et al.

Experiments with pre-trained models such as BERT are often based on a single checkpoint. While the conclusions drawn apply to the artifact tested in the experiment (i.e., the particular instance of the model), it is not always clear whether they hold for the more general procedure which includes the architecture, training data, initialization scheme, and loss function. Recent work has shown that repeating the pre-training process can lead to substantially different performance, suggesting that an alternate strategy is needed to make principled statements about procedures. To enable researchers to draw more robust conclusions, we introduce the MultiBERTs, a set of 25 BERT-Base checkpoints, trained with similar hyper-parameters as the original BERT model but differing in random weight initialization and shuffling of training data. We also define the Multi-Bootstrap, a non-parametric bootstrap method for statistical inference designed for settings where there are multiple pre-trained models and limited test data. To illustrate our approach, we present a case study of gender bias in coreference resolution, in which the Multi-Bootstrap lets us measure effects that may not be detected with a single checkpoint. We release our models and statistical library along with an additional set of 140 intermediate checkpoints captured during pre-training to facilitate research on learning dynamics.

Jason Wei, Clara Meister, Ryan Cotterell

The uniform information density (UID) hypothesis, which posits that speakers behaving optimally tend to distribute information uniformly across a linguistic signal, has gained traction in psycholinguistics as an explanation for certain syntactic, morphological, and prosodic choices. In this work, we explore whether the UID hypothesis can be operationalized as an inductive bias for statistical language modeling. Specifically, we augment the canonical MLE objective for training language models with a regularizer that encodes UID. In experiments on ten languages spanning five language families, we find that using UID regularization consistently improves perplexity in language models, having a larger effect when training data is limited. Moreover, via an analysis of generated sequences, we find that UID-regularized language models have other desirable properties, e.g., they generate text that is more lexically diverse. Our results not only suggest that UID is a reasonable inductive bias for language modeling, but also provide an alternative validation of the UID hypothesis using modern-day NLP tools.

Ruibo Liu, Chenyan Jia, Jason Wei et al.

Current large-scale language models can be politically biased as a result of the data they are trained on, potentially causing serious problems when they are deployed in real-world settings. In this paper, we describe metrics for measuring political bias in GPT-2 generation and propose a reinforcement learning (RL) framework for mitigating political biases in generated text. By using rewards from word embeddings or a classifier, our RL framework guides debiased generation without having access to the training data or requiring the model to be retrained. In empirical experiments on three attributes sensitive to political bias (gender, location, and topic), our methods reduced bias according to both our metrics and human evaluation, while maintaining readability and semantic coherence.

Jerry Wei, Arief Suriawinata, Bing Ren et al.

With the rise of deep learning, there has been increased interest in using neural networks for histopathology image analysis, a field that investigates the properties of biopsy or resected specimens traditionally manually examined under a microscope by pathologists. However, challenges such as limited data, costly annotation, and processing high-resolution and variable-size images make it difficult to quickly iterate over model designs. Throughout scientific history, many significant research directions have leveraged small-scale experimental setups as petri dishes to efficiently evaluate exploratory ideas. In this paper, we introduce a minimalist histopathology image analysis dataset (MHIST), an analogous petri dish for histopathology image analysis. MHIST is a binary classification dataset of 3,152 fixed-size images of colorectal polyps, each with a gold-standard label determined by the majority vote of seven board-certified gastrointestinal pathologists and annotator agreement level. MHIST occupies less than 400 MB of disk space, and a ResNet-18 baseline can be trained to convergence on MHIST in just 6 minutes using 3.5 GB of memory on a NVIDIA RTX 3090. As example use cases, we use MHIST to study natural questions such as how dataset size, network depth, transfer learning, and high-disagreement examples affect model performance. By introducing MHIST, we hope to not only help facilitate the work of current histopathology imaging researchers, but also make the field more-accessible to the general community. Our dataset is available at https://bmirds.github.io/MHIST.

Jason Wei, Chengyu Huang, Shiqi Xu et al.

Traditional data augmentation aims to increase the coverage of the input distribution by generating augmented examples that strongly resemble original samples in an online fashion where augmented examples dominate training. In this paper, we propose an alternative perspective -- a multi-task view (MTV) of data augmentation -- in which the primary task trains on original examples and the auxiliary task trains on augmented examples. In MTV data augmentation, both original and augmented samples are weighted substantively during training, relaxing the constraint that augmented examples must resemble original data and thereby allowing us to apply stronger levels of augmentation. In empirical experiments using four common data augmentation techniques on three benchmark text classification datasets, we find that the MTV leads to higher and more robust performance improvements than traditional augmentation.

Jerry Wei, Arief Suriawinata, Bing Ren et al.

Applying curriculum learning requires both a range of difficulty in data and a method for determining the difficulty of examples. In many tasks, however, satisfying these requirements can be a formidable challenge. In this paper, we contend that histopathology image classification is a compelling use case for curriculum learning. Based on the nature of histopathology images, a range of difficulty inherently exists among examples, and, since medical datasets are often labeled by multiple annotators, annotator agreement can be used as a natural proxy for the difficulty of a given example. Hence, we propose a simple curriculum learning method that trains on progressively-harder images as determined by annotator agreement. We evaluate our hypothesis on the challenging and clinically-important task of colorectal polyp classification. Whereas vanilla training achieves an AUC of 83.7% for this task, a model trained with our proposed curriculum learning approach achieves an AUC of 88.2%, an improvement of 4.5%. Our work aims to inspire researchers to think more creatively and rigorously when choosing contexts for applying curriculum learning.

Jerry Wei, Arief Suriawinata, Xiaoying Liu et al.

The unique nature of histopathology images opens the door to domain-specific formulations of image translation models. We propose a difficulty translation model that modifies colorectal histopathology images to be more challenging to classify. Our model comprises a scorer, which provides an output confidence to measure the difficulty of images, and an image translator, which learns to translate images from easy-to-classify to hard-to-classify using a training set defined by the scorer. We present three findings. First, generated images were indeed harder to classify for both human pathologists and machine learning classifiers than their corresponding source images. Second, image classifiers trained with generated images as augmented data performed better on both easy and hard images from an independent test set. Finally, human annotator agreement and our model's measure of difficulty correlated strongly, implying that for future work requiring human annotator agreement, the confidence score of a machine learning classifier could be used as a proxy.

Lia X. Harrington, Jason W. Wei, Arief A. Suriawinata et al.

Identifying patient characteristics that influence the rate of colorectal polyp recurrence can provide important insights into which patients are at higher risk for recurrence. We used natural language processing to extract polyp morphological characteristics from 953 polyp-presenting patients' electronic medical records. We used subsequent colonoscopy reports to examine how the time to polyp recurrence (731 patients experienced recurrence) is influenced by these characteristics as well as anthropometric features using Kaplan-Meier curves, Cox proportional hazards modeling, and random survival forest models. We found that the rate of recurrence differed significantly by polyp size, number, and location and patient smoking status. Additionally, right-sided colon polyps increased recurrence risk by 30% compared to left-sided polyps. History of tobacco use increased polyp recurrence risk by 20% compared to never-users. A random survival forest model showed an AUC of 0.65 and identified several other predictive variables, which can inform development of personalized polyp surveillance plans.

Jerry Wei, Arief Suriawinata, Louis Vaickus et al.

We present an image translation approach to generate augmented data for mitigating data imbalances in a dataset of histopathology images of colorectal polyps, adenomatous tumors that can lead to colorectal cancer if left untreated. By applying cycle-consistent generative adversarial networks (CycleGANs) to a source domain of normal colonic mucosa images, we generate synthetic colorectal polyp images that belong to diagnostically less common polyp classes. Generated images maintain the general structure of their source image but exhibit adenomatous features that can be enhanced with our proposed filtration module, called Path-Rank-Filter. We evaluate the quality of generated images through Turing tests with four gastrointestinal pathologists, finding that at least two of the four pathologists could not identify generated images at a statistically significant level. Finally, we demonstrate that using CycleGAN-generated images to augment training data improves the AUC of a convolutional neural network for detecting sessile serrated adenomas by over 10%, suggesting that our approach might warrant further research for other histopathology image classification tasks.

Jason W. Wei, Arief A. Suriawinata, Louis J. Vaickus et al.

Histological classification of colorectal polyps plays a critical role in both screening for colorectal cancer and care of affected patients. An accurate and automated algorithm for the classification of colorectal polyps on digitized histopathology slides could benefit clinicians and patients. Evaluate the performance and assess the generalizability of a deep neural network for colorectal polyp classification on histopathology slide images using a multi-institutional dataset. In this study, we developed a deep neural network for classification of four major colorectal polyp types, tubular adenoma, tubulovillous/villous adenoma, hyperplastic polyp, and sessile serrated adenoma, based on digitized histopathology slides from our institution, Dartmouth-Hitchcock Medical Center (DHMC), in New Hampshire. We evaluated the deep neural network on an internal dataset of 157 histopathology slide images from DHMC, as well as on an external dataset of 238 histopathology slide images from 24 different institutions spanning 13 states in the United States. We measured accuracy, sensitivity, and specificity of our model in this evaluation and compared its performance to local pathologists' diagnoses at the point-of-care retrieved from corresponding pathology laboratories. For the internal evaluation, the deep neural network had a mean accuracy of 93.5% (95% CI 89.6%-97.4%), compared with local pathologists' accuracy of 91.4% (95% CI 87.0%-95.8%). On the external test set, the deep neural network achieved an accuracy of 87.0% (95% CI 82.7%-91.3%), comparable with local pathologists' accuracy of 86.6% (95% CI 82.3%-90.9%). If confirmed in clinical settings, our model could assist pathologists by improving the diagnostic efficiency, reproducibility, and accuracy of colorectal cancer screenings.

Jason W. Wei, Jerry W. Wei, Christopher R. Jackson et al.

Celiac disease prevalence and diagnosis have increased substantially in recent years. The current gold standard for celiac disease confirmation is visual examination of duodenal mucosal biopsies. An accurate computer-aided biopsy analysis system using deep learning can help pathologists diagnose celiac disease more efficiently. In this study, we trained a deep learning model to detect celiac disease on duodenal biopsy images. Our model uses a state-of-the-art residual convolutional neural network to evaluate patches of duodenal tissue and then aggregates those predictions for whole-slide classification. We tested the model on an independent set of 212 images and evaluated its classification results against reference standards established by pathologists. Our model identified celiac disease, normal tissue, and nonspecific duodenitis with accuracies of 95.3%, 91.0%, and 89.2%, respectively. The area under the receiver operating characteristic curve was greater than 0.95 for all classes. We have developed an automated biopsy analysis system that achieves high performance in detecting celiac disease on biopsy slides. Our system can highlight areas of interest and provide preliminary classification of duodenal biopsies prior to review by pathologists. This technology has great potential for improving the accuracy and efficiency of celiac disease diagnosis.

Jason Wei, Kai Zou

We present EDA: easy data augmentation techniques for boosting performance on text classification tasks. EDA consists of four simple but powerful operations: synonym replacement, random insertion, random swap, and random deletion. On five text classification tasks, we show that EDA improves performance for both convolutional and recurrent neural networks. EDA demonstrates particularly strong results for smaller datasets; on average, across five datasets, training with EDA while using only 50% of the available training set achieved the same accuracy as normal training with all available data. We also performed extensive ablation studies and suggest parameters for practical use.