Christopher Ré

Neel Guha, Julian Nyarko, Daniel E. Ho et al.

The advent of large language models (LLMs) and their adoption by the legal community has given rise to the question: what types of legal reasoning can LLMs perform? To enable greater study of this question, we present LegalBench: a collaboratively constructed legal reasoning benchmark consisting of 162 tasks covering six different types of legal reasoning. LegalBench was built through an interdisciplinary process, in which we collected tasks designed and hand-crafted by legal professionals. Because these subject matter experts took a leading role in construction, tasks either measure legal reasoning capabilities that are practically useful, or measure reasoning skills that lawyers find interesting. To enable cross-disciplinary conversations about LLMs in the law, we additionally show how popular legal frameworks for describing legal reasoning -- which distinguish between its many forms -- correspond to LegalBench tasks, thus giving lawyers and LLM developers a common vocabulary. This paper describes LegalBench, presents an empirical evaluation of 20 open-source and commercial LLMs, and illustrates the types of research explorations LegalBench enables.

Sabri Eyuboglu, Maya Varma, Khaled Saab et al. · stanford

Machine learning models that achieve high overall accuracy often make systematic errors on important subsets (or slices) of data. Identifying underperforming slices is particularly challenging when working with high-dimensional inputs (e.g. images, audio), where important slices are often unlabeled. In order to address this issue, recent studies have proposed automated slice discovery methods (SDMs), which leverage learned model representations to mine input data for slices on which a model performs poorly. To be useful to a practitioner, these methods must identify slices that are both underperforming and coherent (i.e. united by a human-understandable concept). However, no quantitative evaluation framework currently exists for rigorously assessing SDMs with respect to these criteria. Additionally, prior qualitative evaluations have shown that SDMs often identify slices that are incoherent. In this work, we address these challenges by first designing a principled evaluation framework that enables a quantitative comparison of SDMs across 1,235 slice discovery settings in three input domains (natural images, medical images, and time-series data). Then, motivated by the recent development of powerful cross-modal representation learning approaches, we present Domino, an SDM that leverages cross-modal embeddings and a novel error-aware mixture model to discover and describe coherent slices. We find that Domino accurately identifies 36% of the 1,235 slices in our framework - a 12 percentage point improvement over prior methods. Further, Domino is the first SDM that can provide natural language descriptions of identified slices, correctly generating the exact name of the slice in 35% of settings.

Khaled Saab, Siyi Tang, Mohamed Taha et al. · stanford

A major barrier to deploying healthcare AI models is their trustworthiness. One form of trustworthiness is a model's robustness across different subgroups: while existing models may exhibit expert-level performance on aggregate metrics, they often rely on non-causal features, leading to errors in hidden subgroups. To take a step closer towards trustworthy seizure onset detection from EEG, we propose to leverage annotations that are produced by healthcare personnel in routine clinical workflows -- which we refer to as workflow notes -- that include multiple event descriptions beyond seizures. Using workflow notes, we first show that by scaling training data to an unprecedented level of 68,920 EEG hours, seizure onset detection performance significantly improves (+12.3 AUROC points) compared to relying on smaller training sets with expensive manual gold-standard labels. Second, we reveal that our binary seizure onset detection model underperforms on clinically relevant subgroups (e.g., up to a margin of 6.5 AUROC points between pediatrics and adults), while having significantly higher false positives on EEG clips showing non-epileptiform abnormalities compared to any EEG clip (+19 FPR points). To improve model robustness to hidden subgroups, we train a multilabel model that classifies 26 attributes other than seizures, such as spikes, slowing, and movement artifacts. We find that our multilabel model significantly improves overall seizure onset detection performance (+5.9 AUROC points) while greatly improving performance among subgroups (up to +8.3 AUROC points), and decreases false positives on non-epileptiform abnormalities by 8 FPR points. Finally, we propose a clinical utility metric based on false positives per 24 EEG hours and find that our multilabel model improves this clinical utility metric by a factor of 2x across different clinical settings.

Armin W. Thomas, Christopher Ré, Russell A. Poldrack

Deep learning (DL) models find increasing application in mental state decoding, where researchers seek to understand the mapping between mental states (e.g., perceiving fear or joy) and brain activity by identifying those brain regions (and networks) whose activity allows to accurately identify (i.e., decode) these states. Once a DL model has been trained to accurately decode a set of mental states, neuroimaging researchers often make use of interpretation methods from explainable artificial intelligence research to understand the model's learned mappings between mental states and brain activity. Here, we compare the explanation performance of prominent interpretation methods in a mental state decoding analysis of three functional Magnetic Resonance Imaging (fMRI) datasets. Our findings demonstrate a gradient between two key characteristics of an explanation in mental state decoding, namely, its biological plausibility and faithfulness: interpretation methods with high explanation faithfulness, which capture the model's decision process well, generally provide explanations that are biologically less plausible than the explanations of interpretation methods with less explanation faithfulness. Based on this finding, we provide specific recommendations for the application of interpretation methods in mental state decoding.

Jupinder Parmar, Khaled Saab, Brian Pogatchnik et al.

Domain generalization in medical image classification is an important problem for trustworthy machine learning to be deployed in healthcare. We find that existing approaches for domain generalization which utilize ground-truth abnormality segmentations to control feature attributions have poor out-of-distribution (OOD) performance relative to the standard baseline of empirical risk minimization (ERM). We investigate what regions of an image are important for medical image classification and show that parts of the background, that which is not contained in the abnormality segmentation, provides helpful signal. We then develop a new task-specific mask which covers all relevant regions. Utilizing this new segmentation mask significantly improves the performance of the existing methods on the OOD test sets. To obtain better generalization results than ERM, we find it necessary to scale up the training data size in addition to the usage of these task-specific masks.

Maurice Weber, Daniel Fu, Quentin Anthony et al.

Large language models are increasingly becoming a cornerstone technology in artificial intelligence, the sciences, and society as a whole, yet the optimal strategies for dataset composition and filtering remain largely elusive. Many of the top-performing models lack transparency in their dataset curation and model development processes, posing an obstacle to the development of fully open language models. In this paper, we identify three core data-related challenges that must be addressed to advance open-source language models. These include (1) transparency in model development, including the data curation process, (2) access to large quantities of high-quality data, and (3) availability of artifacts and metadata for dataset curation and analysis. To address these challenges, we release RedPajama-V1, an open reproduction of the LLaMA training dataset. In addition, we release RedPajama-V2, a massive web-only dataset consisting of raw, unfiltered text data together with quality signals and metadata. Together, the RedPajama datasets comprise over 100 trillion tokens spanning multiple domains and with their quality signals facilitate the filtering of data, aiming to inspire the development of numerous new datasets. To date, these datasets have already been used in the training of strong language models used in production, such as Snowflake Arctic, Salesforce's XGen and AI2's OLMo. To provide insight into the quality of RedPajama, we present a series of analyses and ablation studies with decoder-only language models with up to 1.6B parameters. Our findings demonstrate how quality signals for web data can be effectively leveraged to curate high-quality subsets of the dataset, underscoring the potential of RedPajama to advance the development of transparent and high-performing language models at scale.

Michael Wornow, Avanika Narayan, Krista Opsahl-Ong et al.

Automating enterprise workflows could unlock $4 trillion/year in productivity gains. Despite being of interest to the data management community for decades, the ultimate vision of end-to-end workflow automation has remained elusive. Current solutions rely on process mining and robotic process automation (RPA), in which a bot is hard-coded to follow a set of predefined rules for completing a workflow. Through case studies of a hospital and large B2B enterprise, we find that the adoption of RPA has been inhibited by high set-up costs (12-18 months), unreliable execution (60% initial accuracy), and burdensome maintenance (requiring multiple FTEs). Multimodal foundation models (FMs) such as GPT-4 offer a promising new approach for end-to-end workflow automation given their generalized reasoning and planning abilities. To study these capabilities we propose ECLAIR, a system to automate enterprise workflows with minimal human supervision. We conduct initial experiments showing that multimodal FMs can address the limitations of traditional RPA with (1) near-human-level understanding of workflows (93% accuracy on a workflow understanding task) and (2) instant set-up with minimal technical barrier (based solely on a natural language description of a workflow, ECLAIR achieves end-to-end completion rates of 40%). We identify human-AI collaboration, validation, and self-improvement as open challenges, and suggest ways they can be solved with data management techniques. Code is available at: https://github.com/HazyResearch/eclair-agents

Arjun D Desai, Batu M Ozturkler, Christopher M Sandino et al.

Deep learning (DL) has shown promise for faster, high quality accelerated MRI reconstruction. However, supervised DL methods depend on extensive amounts of fully-sampled (labeled) data and are sensitive to out-of-distribution (OOD) shifts, particularly low signal-to-noise ratio (SNR) acquisitions. To alleviate this challenge, we propose Noise2Recon, a model-agnostic, consistency training method for joint MRI reconstruction and denoising that can use both fully-sampled (labeled) and undersampled (unlabeled) scans in semi-supervised and self-supervised settings. With limited or no labeled training data, Noise2Recon outperforms compressed sensing and deep learning baselines, including supervised networks, augmentation-based training, fine-tuned denoisers, and self-supervised methods, and matches performance of supervised models, which were trained with 14x more fully-sampled scans. Noise2Recon also outperforms all baselines, including state-of-the-art fine-tuning and augmentation techniques, among low-SNR scans and when generalizing to other OOD factors, such as changes in acceleration factors and different datasets. Augmentation extent and loss weighting hyperparameters had negligible impact on Noise2Recon compared to supervised methods, which may indicate increased training stability. Our code is available at https://github.com/ad12/meddlr.

Neel Guha, Mayee F. Chen, Trevor Chow et al.

Large language models (LLMs) are increasingly used in applications where LLM inputs may span many different tasks. Recent work has found that the choice of LLM is consequential, and different LLMs may be good for different input samples. Prior approaches have thus explored how engineers might select an LLM to use for each sample (i.e. routing). While existing routing methods mostly require training auxiliary models on human-annotated data, our work explores whether it is possible to perform unsupervised routing. We propose Smoothie, a weak supervision-inspired routing approach that requires no labeled data. Given a set of outputs from different LLMs, Smoothie constructs a latent variable graphical model over embedding representations of observable LLM outputs and unknown "true" outputs. Using this graphical model, we estimate sample-dependent quality scores for each LLM, and route each sample to the LLM with the highest corresponding score. We find that Smoothie's LLM quality-scores correlate with ground-truth model quality (correctly identifying the optimal model on 9/14 tasks), and that Smoothie outperforms baselines for routing by up to 10 points accuracy.

Mayee F. Chen, Michael Y. Hu, Nicholas Lourie et al.

Language model performance depends on identifying the optimal mixture of data groups to train on (e.g., law, code, math). Prior work has proposed a diverse set of methods to efficiently learn mixture proportions, ranging from fitting regression models over training runs to dynamically updating proportions throughout training. Surprisingly, we find that no existing method consistently outperforms a simple stratified sampling baseline in terms of average test perplexity. To understand this inconsistency, we unify existing methods into a standard framework, showing they are equivalent to solving a common optimization problem: minimize average loss subject to a method-specific mixing law -- an implicit assumption on the relationship between loss and mixture proportions. This framework suggests that measuring the fidelity of a method's mixing law can offer insights into its performance. Empirically, we find that existing methods set their mixing law parameters inaccurately, resulting in the inconsistent mixing performance we observe. Using this insight, we derive a new online method named Aioli, which directly estimates the mixing law parameters throughout training and uses them to dynamically adjust proportions. Aioli outperforms stratified sampling on 6 out of 6 datasets by an average of 0.27 test perplexity points, whereas existing methods fail to consistently beat stratified sampling, doing up to 6.9 points worse. Moreover, in a practical setting where proportions are learned on shorter runs due to computational constraints, Aioli can dynamically adjust these proportions over the full training run, consistently improving performance over existing methods by up to 12.012 test perplexity points.

Gautam Machiraju, Alexander Derry, Arjun Desai et al.

Feature attribution, the ability to localize regions of the input data that are relevant for classification, is an important capability for ML models in scientific and biomedical domains. Current methods for feature attribution, which rely on "explaining" the predictions of end-to-end classifiers, suffer from imprecise feature localization and are inadequate for use with small sample sizes and high-dimensional datasets due to computational challenges. We introduce prospector heads, an efficient and interpretable alternative to explanation-based attribution methods that can be applied to any encoder and any data modality. Prospector heads generalize across modalities through experiments on sequences (text), images (pathology), and graphs (protein structures), outperforming baseline attribution methods by up to 26.3 points in mean localization AUPRC. We also demonstrate how prospector heads enable improved interpretation and discovery of class-specific patterns in input data. Through their high performance, flexibility, and generalizability, prospectors provide a framework for improving trust and transparency for ML models in complex domains.

Albert Gu, Isys Johnson, Karan Goel et al.

Recurrent neural networks (RNNs), temporal convolutions, and neural differential equations (NDEs) are popular families of deep learning models for time-series data, each with unique strengths and tradeoffs in modeling power and computational efficiency. We introduce a simple sequence model inspired by control systems that generalizes these approaches while addressing their shortcomings. The Linear State-Space Layer (LSSL) maps a sequence $u \mapsto y$ by simply simulating a linear continuous-time state-space representation $\dot{x} = Ax + Bu, y = Cx + Du$. Theoretically, we show that LSSL models are closely related to the three aforementioned families of models and inherit their strengths. For example, they generalize convolutions to continuous-time, explain common RNN heuristics, and share features of NDEs such as time-scale adaptation. We then incorporate and generalize recent theory on continuous-time memorization to introduce a trainable subset of structured matrices $A$ that endow LSSLs with long-range memory. Empirically, stacking LSSL layers into a simple deep neural network obtains state-of-the-art results across time series benchmarks for long dependencies in sequential image classification, real-world healthcare regression tasks, and speech. On a difficult speech classification task with length-16000 sequences, LSSL outperforms prior approaches by 24 accuracy points, and even outperforms baselines that use hand-crafted features on 100x shorter sequences.

Sahaana Suri, Ihab F. Ilyas, Christopher Ré et al.

Structured data, or data that adheres to a pre-defined schema, can suffer from fragmented context: information describing a single entity can be scattered across multiple datasets or tables tailored for specific business needs, with no explicit linking keys (e.g., primary key-foreign key relationships or heuristic functions). Context enrichment, or rebuilding fragmented context, using keyless joins is an implicit or explicit step in machine learning (ML) pipelines over structured data sources. This process is tedious, domain-specific, and lacks support in now-prevalent no-code ML systems that let users create ML pipelines using just input data and high-level configuration files. In response, we propose Ember, a system that abstracts and automates keyless joins to generalize context enrichment. Our key insight is that Ember can enable a general keyless join operator by constructing an index populated with task-specific embeddings. Ember learns these embeddings by leveraging Transformer-based representation learning techniques. We describe our core architectural principles and operators when developing Ember, and empirically demonstrate that Ember allows users to develop no-code pipelines for five domains, including search, recommendation and question answering, and can exceed alternatives by up to 39% recall, with as little as a single line configuration change.

Tri Dao, Nimit S. Sohoni, Albert Gu et al.

Modern neural network architectures use structured linear transformations, such as low-rank matrices, sparse matrices, permutations, and the Fourier transform, to improve inference speed and reduce memory usage compared to general linear maps. However, choosing which of the myriad structured transformations to use (and its associated parameterization) is a laborious task that requires trading off speed, space, and accuracy. We consider a different approach: we introduce a family of matrices called kaleidoscope matrices (K-matrices) that provably capture any structured matrix with near-optimal space (parameter) and time (arithmetic operation) complexity. We empirically validate that K-matrices can be automatically learned within end-to-end pipelines to replace hand-crafted procedures, in order to improve model quality. For example, replacing channel shuffles in ShuffleNet improves classification accuracy on ImageNet by up to 5%. K-matrices can also simplify hand-engineered pipelines -- we replace filter bank feature computation in speech data preprocessing with a learnable kaleidoscope layer, resulting in only 0.4% loss in accuracy on the TIMIT speech recognition task. In addition, K-matrices can capture latent structure in models: for a challenging permuted image classification task, a K-matrix based representation of permutations is able to learn the right latent structure and improves accuracy of a downstream convolutional model by over 9%. We provide a practically efficient implementation of our approach, and use K-matrices in a Transformer network to attain 36% faster end-to-end inference speed on a language translation task.

Nimit S. Sohoni, Jared A. Dunnmon, Geoffrey Angus et al.

In real-world classification tasks, each class often comprises multiple finer-grained "subclasses." As the subclass labels are frequently unavailable, models trained using only the coarser-grained class labels often exhibit highly variable performance across different subclasses. This phenomenon, known as hidden stratification, has important consequences for models deployed in safety-critical applications such as medicine. We propose GEORGE, a method to both measure and mitigate hidden stratification even when subclass labels are unknown. We first observe that unlabeled subclasses are often separable in the feature space of deep neural networks, and exploit this fact to estimate subclass labels for the training data via clustering techniques. We then use these approximate subclass labels as a form of noisy supervision in a distributionally robust optimization objective. We theoretically characterize the performance of GEORGE in terms of the worst-case generalization error across any subclass. We empirically validate GEORGE on a mix of real-world and benchmark image classification datasets, and show that our approach boosts worst-case subclass accuracy by up to 22 percentage points compared to standard training techniques, without requiring any prior information about the subclasses.

Ines Chami, Albert Gu, Vaggos Chatziafratis et al.

Similarity-based Hierarchical Clustering (HC) is a classical unsupervised machine learning algorithm that has traditionally been solved with heuristic algorithms like Average-Linkage. Recently, Dasgupta reframed HC as a discrete optimization problem by introducing a global cost function measuring the quality of a given tree. In this work, we provide the first continuous relaxation of Dasgupta's discrete optimization problem with provable quality guarantees. The key idea of our method, HypHC, is showing a direct correspondence from discrete trees to continuous representations (via the hyperbolic embeddings of their leaf nodes) and back (via a decoding algorithm that maps leaf embeddings to a dendrogram), allowing us to search the space of discrete binary trees with continuous optimization. Building on analogies between trees and hyperbolic space, we derive a continuous analogue for the notion of lowest common ancestor, which leads to a continuous relaxation of Dasgupta's discrete objective. We can show that after decoding, the global minimizer of our continuous relaxation yields a discrete tree with a (1 + epsilon)-factor approximation for Dasgupta's optimal tree, where epsilon can be made arbitrarily small and controls optimization challenges. We experimentally evaluate HypHC on a variety of HC benchmarks and find that even approximate solutions found with gradient descent have superior clustering quality than agglomerative heuristics or other gradient based algorithms. Finally, we highlight the flexibility of HypHC using end-to-end training in a downstream classification task.

Sen Wu, Hongyang R. Zhang, Gregory Valiant et al.

Data augmentation is a powerful technique to improve performance in applications such as image and text classification tasks. Yet, there is little rigorous understanding of why and how various augmentations work. In this work, we consider a family of linear transformations and study their effects on the ridge estimator in an over-parametrized linear regression setting. First, we show that transformations that preserve the labels of the data can improve estimation by enlarging the span of the training data. Second, we show that transformations that mix data can improve estimation by playing a regularization effect. Finally, we validate our theoretical insights on MNIST. Based on the insights, we propose an augmentation scheme that searches over the space of transformations by how uncertain the model is about the transformed data. We validate our proposed scheme on image and text datasets. For example, our method outperforms random sampling methods by 1.24% on CIFAR-100 using Wide-ResNet-28-10. Furthermore, we achieve comparable accuracy to the SoTA Adversarial AutoAugment on CIFAR-10, CIFAR-100, SVHN, and ImageNet datasets.

Zhaobin Kuang, Frederic Sala, Nimit Sohoni et al.

A popular way to estimate the causal effect of a variable x on y from observational data is to use an instrumental variable (IV): a third variable z that affects y only through x. The more strongly z is associated with x, the more reliable the estimate is, but such strong IVs are difficult to find. Instead, practitioners combine more commonly available IV candidates---which are not necessarily strong, or even valid, IVs---into a single "summary" that is plugged into causal effect estimators in place of an IV. In genetic epidemiology, such approaches are known as allele scores. Allele scores require strong assumptions---independence and validity of all IV candidates---for the resulting estimate to be reliable. To relax these assumptions, we propose Ivy, a new method to combine IV candidates that can handle correlated and invalid IV candidates in a robust manner. Theoretically, we characterize this robustness, its limits, and its impact on the resulting causal estimates. Empirically, Ivy can correctly identify the directionality of known relationships and is robust against false discovery (median effect size <= 0.025) on three real-world datasets with no causal effects, while allele scores return more biased estimates (median effect size >= 0.118).

Sarah M. Hooper, Jared A. Dunnmon, Matthew P. Lungren et al.

Automated medical image classification with convolutional neural networks (CNNs) has great potential to impact healthcare, particularly in resource-constrained healthcare systems where fewer trained radiologists are available. However, little is known about how well a trained CNN can perform on images with the increased noise levels, different acquisition protocols, or additional artifacts that may arise when using low-cost scanners, which can be underrepresented in datasets collected from well-funded hospitals. In this work, we investigate how a model trained to triage head computed tomography (CT) scans performs on images acquired with reduced x-ray tube current, fewer projections per gantry rotation, and limited angle scans. These changes can reduce the cost of the scanner and demands on electrical power but come at the expense of increased image noise and artifacts. We first develop a model to triage head CTs and report an area under the receiver operating characteristic curve (AUROC) of 0.77. We then show that the trained model is robust to reduced tube current and fewer projections, with the AUROC dropping only 0.65% for images acquired with a 16x reduction in tube current and 0.22% for images acquired with 8x fewer projections. Finally, for significantly degraded images acquired by a limited angle scan, we show that a model trained specifically to classify such images can overcome the technological limitations to reconstruction and maintain an AUROC within 0.09% of the original model.

Daniel Y. Fu, Mayee F. Chen, Frederic Sala et al.

Weak supervision is a popular method for building machine learning models without relying on ground truth annotations. Instead, it generates probabilistic training labels by estimating the accuracies of multiple noisy labeling sources (e.g., heuristics, crowd workers). Existing approaches use latent variable estimation to model the noisy sources, but these methods can be computationally expensive, scaling superlinearly in the data. In this work, we show that, for a class of latent variable models highly applicable to weak supervision, we can find a closed-form solution to model parameters, obviating the need for iterative solutions like stochastic gradient descent (SGD). We use this insight to build FlyingSquid, a weak supervision framework that runs orders of magnitude faster than previous weak supervision approaches and requires fewer assumptions. In particular, we prove bounds on generalization error without assuming that the latent variable model can exactly parameterize the underlying data distribution. Empirically, we validate FlyingSquid on benchmark weak supervision datasets and find that it achieves the same or higher quality compared to previous approaches without the need to tune an SGD procedure, recovers model parameters 170 times faster on average, and enables new video analysis and online learning applications.

Ines Chami, Rex Ying, Christopher Ré et al.

Graph convolutional neural networks (GCNs) embed nodes in a graph into Euclidean space, which has been shown to incur a large distortion when embedding real-world graphs with scale-free or hierarchical structure. Hyperbolic geometry offers an exciting alternative, as it enables embeddings with much smaller distortion. However, extending GCNs to hyperbolic geometry presents several unique challenges because it is not clear how to define neural network operations, such as feature transformation and aggregation, in hyperbolic space. Furthermore, since input features are often Euclidean, it is unclear how to transform the features into hyperbolic embeddings with the right amount of curvature. Here we propose Hyperbolic Graph Convolutional Neural Network (HGCN), the first inductive hyperbolic GCN that leverages both the expressiveness of GCNs and hyperbolic geometry to learn inductive node representations for hierarchical and scale-free graphs. We derive GCN operations in the hyperboloid model of hyperbolic space and map Euclidean input features to embeddings in hyperbolic spaces with different trainable curvature at each layer. Experiments demonstrate that HGCN learns embeddings that preserve hierarchical structure, and leads to improved performance when compared to Euclidean analogs, even with very low dimensional embeddings: compared to state-of-the-art GCNs, HGCN achieves an error reduction of up to 63.1% in ROC AUC for link prediction and of up to 47.5% in F1 score for node classification, also improving state-of-the art on the Pubmed dataset.

Luke Oakden-Rayner, Jared Dunnmon, Gustavo Carneiro et al.

Machine learning models for medical image analysis often suffer from poor performance on important subsets of a population that are not identified during training or testing. For example, overall performance of a cancer detection model may be high, but the model still consistently misses a rare but aggressive cancer subtype. We refer to this problem as hidden stratification, and observe that it results from incompletely describing the meaningful variation in a dataset. While hidden stratification can substantially reduce the clinical efficacy of machine learning models, its effects remain difficult to measure. In this work, we assess the utility of several possible techniques for measuring and describing hidden stratification effects, and characterize these effects on multiple medical imaging datasets. We find evidence that hidden stratification can occur in unidentified imaging subsets with low prevalence, low label quality, subtle distinguishing features, or spurious correlates, and that it can result in relative performance differences of over 20% on clinically important subsets. Finally, we explore the clinical implications of our findings, and suggest that evaluation of hidden stratification should be a critical component of any machine learning deployment in medical imaging.

Jared Dunnmon, Alexander Ratner, Nishith Khandwala et al.

Labeling training datasets has become a key barrier to building medical machine learning models. One strategy is to generate training labels programmatically, for example by applying natural language processing pipelines to text reports associated with imaging studies. We propose cross-modal data programming, which generalizes this intuitive strategy in a theoretically-grounded way that enables simpler, clinician-driven input, reduces required labeling time, and improves with additional unlabeled data. In this approach, clinicians generate training labels for models defined over a target modality (e.g. images or time series) by writing rules over an auxiliary modality (e.g. text reports). The resulting technical challenge consists of estimating the accuracies and correlations of these rules; we extend a recent unsupervised generative modeling technique to handle this cross-modal setting in a provably consistent way. Across four applications in radiography, computed tomography, and electroencephalography, and using only several hours of clinician time, our approach matches or exceeds the efficacy of physician-months of hand-labeling with statistical significance, demonstrating a fundamentally faster and more flexible way of building machine learning models in medicine.

Alexander Ratner, Braden Hancock, Jared Dunnmon et al.

As machine learning models continue to increase in complexity, collecting large hand-labeled training sets has become one of the biggest roadblocks in practice. Instead, weaker forms of supervision that provide noisier but cheaper labels are often used. However, these weak supervision sources have diverse and unknown accuracies, may output correlated labels, and may label different tasks or apply at different levels of granularity. We propose a framework for integrating and modeling such weak supervision sources by viewing them as labeling different related sub-tasks of a problem, which we refer to as the multi-task weak supervision setting. We show that by solving a matrix completion-style problem, we can recover the accuracies of these multi-task sources given their dependency structure, but without any labeled data, leading to higher-quality supervision for training an end model. Theoretically, we show that the generalization error of models trained with this approach improves with the number of unlabeled data points, and characterize the scaling with respect to the task and dependency structures. On three fine-grained classification problems, we show that our approach leads to average gains of 20.2 points in accuracy over a traditional supervised approach, 6.8 points over a majority vote baseline, and 4.1 points over a previously proposed weak supervision method that models tasks separately.

Alexander J. Ratner, Henry R. Ehrenberg, Zeshan Hussain et al.

Data augmentation is a ubiquitous technique for increasing the size of labeled training sets by leveraging task-specific data transformations that preserve class labels. While it is often easy for domain experts to specify individual transformations, constructing and tuning the more sophisticated compositions typically needed to achieve state-of-the-art results is a time-consuming manual task in practice. We propose a method for automating this process by learning a generative sequence model over user-specified transformation functions using a generative adversarial approach. Our method can make use of arbitrary, non-deterministic transformation functions, is robust to misspecified user input, and is trained on unlabeled data. The learned transformation model can then be used to perform data augmentation for any end discriminative model. In our experiments, we show the efficacy of our approach on both image and text datasets, achieving improvements of 4.0 accuracy points on CIFAR-10, 1.4 F1 points on the ACE relation extraction task, and 3.4 accuracy points when using domain-specific transformation operations on a medical imaging dataset as compared to standard heuristic augmentation approaches.

Christopher De Sa, Ce Zhang, Kunle Olukotun et al.

Gibbs sampling on factor graphs is a widely used inference technique, which often produces good empirical results. Theoretical guarantees for its performance are weak: even for tree structured graphs, the mixing time of Gibbs may be exponential in the number of variables. To help understand the behavior of Gibbs sampling, we introduce a new (hyper)graph property, called hierarchy width. We show that under suitable conditions on the weights, bounded hierarchy width ensures polynomial mixing time. Our study of hierarchy width is in part motivated by a class of factor graph templates, hierarchical templates, which have bounded hierarchy width---regardless of the data used to instantiate them. We demonstrate a rich application from natural language processing in which Gibbs sampling provably mixes rapidly and achieves accuracy that exceeds human volunteers.

Christopher Ré, Amir Abbas Sadeghian, Zifei Shan et al.

Knowledge base construction (KBC) is the process of populating a knowledge base, i.e., a relational database together with inference rules, with information extracted from documents and structured sources. KBC blurs the distinction between two traditional database problems, information extraction and information integration. For the last several years, our group has been building knowledge bases with scientific collaborators. Using our approach, we have built knowledge bases that have comparable and sometimes better quality than those constructed by human volunteers. In contrast to these knowledge bases, which took experts a decade or more human years to construct, many of our projects are constructed by a single graduate student. Our approach to KBC is based on joint probabilistic inference and learning, but we do not see inference as either a panacea or a magic bullet: inference is a tool that allows us to be systematic in how we construct, debug, and improve the quality of such systems. In addition, inference allows us to construct these systems in a more loosely coupled way than traditional approaches. To support this idea, we have built the DeepDive system, which has the design goal of letting the user "think about features---not algorithms." We think of DeepDive as declarative in that one specifies what they want but not how to get it. We describe our approach with a focus on feature engineering, which we argue is an understudied problem relative to its importance to end-to-end quality.