HAPI: A Large-scale Longitudinal Dataset of Commercial ML API Predictions

Commercial ML APIs offered by providers such as Google, Amazon and Microsoft have dramatically simplified ML adoption in many applications. Numerous companies and academics pay to use ML APIs for tasks such as object detection, OCR and sentiment analysis. Different ML APIs tackling the same task can have very heterogeneous performance. Moreover, the ML models underlying the APIs also evolve over time. As ML APIs rapidly become a valuable marketplace and a widespread way to consume machine learning, it is critical to systematically study and compare different APIs with each other and to characterize how APIs change over time. However, this topic is currently underexplored due to the lack of data. In this paper, we present HAPI (History of APIs), a longitudinal dataset of 1,761,417 instances of commercial ML API applications (involving APIs from Amazon, Google, IBM, Microsoft and other providers) across diverse tasks including image tagging, speech recognition and text mining from 2020 to 2022. Each instance consists of a query input for an API (e.g., an image or text) along with the API's output prediction/annotation and confidence scores. HAPI is the first large-scale dataset of ML API usages and is a unique resource for studying ML-as-a-service (MLaaS). As examples of the types of analyses that HAPI enables, we show that ML APIs' performance change substantially over time--several APIs' accuracies dropped on specific benchmark datasets. Even when the API's aggregate performance stays steady, its error modes can shift across different subtypes of data between 2020 and 2022. Such changes can substantially impact the entire analytics pipelines that use some ML API as a component. We further use HAPI to study commercial APIs' performance disparities across demographic subgroups over time. HAPI can stimulate more research in the growing field of MLaaS.

LegalBench: Prototyping a Collaborative Benchmark for Legal Reasoning

Can foundation models be guided to execute tasks involving legal reasoning? We believe that building a benchmark to answer this question will require sustained collaborative efforts between the computer science and legal communities. To that end, this short paper serves three purposes. First, we describe how IRAC-a framework legal scholars use to distinguish different types of legal reasoning-can guide the construction of a Foundation Model oriented benchmark. Second, we present a seed set of 44 tasks built according to this framework. We discuss initial findings, and highlight directions for new tasks. Finally-inspired by the Open Science movement-we make a call for the legal and computer science communities to join our efforts by contributing new tasks. This work is ongoing, and our progress can be tracked here: https://github.com/HazyResearch/legalbench.

Contrastive Adapters for Foundation Model Group Robustness

While large pretrained foundation models (FMs) have shown remarkable zero-shot classification robustness to dataset-level distribution shifts, their robustness to subpopulation or group shifts is relatively underexplored. We study this problem, and find that FMs such as CLIP may not be robust to various group shifts. Across 9 robustness benchmarks, zero-shot classification with their embeddings results in gaps of up to 80.7 percentage points (pp) between average and worst-group accuracy. Unfortunately, existing methods to improve robustness require retraining, which can be prohibitively expensive on large foundation models. We also find that efficient ways to improve model inference (e.g., via adapters, lightweight networks with FM embeddings as inputs) do not consistently improve and can sometimes hurt group robustness compared to zero-shot (e.g., increasing the accuracy gap by 50.1 pp on CelebA). We thus develop an adapter training strategy to effectively and efficiently improve FM group robustness. Our motivating observation is that while poor robustness results from groups in the same class being embedded far apart in the foundation model "embedding space," standard adapter training may not bring these points closer together. We thus propose contrastive adapting, which trains adapters with contrastive learning to bring sample embeddings close to both their ground-truth class embeddings and other sample embeddings in the same class. Across the 9 benchmarks, our approach consistently improves group robustness, raising worst-group accuracy by 8.5 to 56.0 pp over zero-shot. Our approach is also efficient, doing so without any FM finetuning and only a fixed set of frozen FM embeddings. On benchmarks such as Waterbirds and CelebA, this leads to worst-group accuracy comparable to state-of-the-art methods that retrain entire models, while only training $\leq$1% of the model parameters.

How to Train Your HiPPO: State Space Models with Generalized Orthogonal Basis Projections

Linear time-invariant state space models (SSM) are a classical model from engineering and statistics, that have recently been shown to be very promising in machine learning through the Structured State Space sequence model (S4). A core component of S4 involves initializing the SSM state matrix to a particular matrix called a HiPPO matrix, which was empirically important for S4's ability to handle long sequences. However, the specific matrix that S4 uses was actually derived in previous work for a particular time-varying dynamical system, and the use of this matrix as a time-invariant SSM had no known mathematical interpretation. Consequently, the theoretical mechanism by which S4 models long-range dependencies actually remains unexplained. We derive a more general and intuitive formulation of the HiPPO framework, which provides a simple mathematical interpretation of S4 as a decomposition onto exponentially-warped Legendre polynomials, explaining its ability to capture long dependencies. Our generalization introduces a theoretically rich class of SSMs that also lets us derive more intuitive S4 variants for other bases such as the Fourier basis, and explains other aspects of training S4, such as how to initialize the important timescale parameter. These insights improve S4's performance to 86% on the Long Range Arena benchmark, with 96% on the most difficult Path-X task.

On the Parameterization and Initialization of Diagonal State Space Models

State space models (SSM) have recently been shown to be very effective as a deep learning layer as a promising alternative to sequence models such as RNNs, CNNs, or Transformers. The first version to show this potential was the S4 model, which is particularly effective on tasks involving long-range dependencies by using a prescribed state matrix called the HiPPO matrix. While this has an interpretable mathematical mechanism for modeling long dependencies, it introduces a custom representation and algorithm that can be difficult to implement. On the other hand, a recent variant of S4 called DSS showed that restricting the state matrix to be fully diagonal can still preserve the performance of the original model when using a specific initialization based on approximating S4's matrix. This work seeks to systematically understand how to parameterize and initialize such diagonal state space models. While it follows from classical results that almost all SSMs have an equivalent diagonal form, we show that the initialization is critical for performance. We explain why DSS works mathematically, by showing that the diagonal restriction of S4's matrix surprisingly recovers the same kernel in the limit of infinite state dimension. We also systematically describe various design choices in parameterizing and computing diagonal SSMs, and perform a controlled empirical study ablating the effects of these choices. Our final model S4D is a simple diagonal version of S4 whose kernel computation requires just 2 lines of code and performs comparably to S4 in almost all settings, with state-of-the-art results for image, audio, and medical time-series domains, and averaging 85\% on the Long Range Arena benchmark.

The Importance of Background Information for Out of Distribution Generalization

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.

Comparing interpretation methods in mental state decoding analyses with deep learning models

Deep learning (DL) methods find increasing application in mental state decoding, where researchers seek to understand the mapping between mental states (such as accepting or rejecting a gamble) and brain activity, by identifying those brain regions (and networks) whose activity allows to accurately identify (i.e., decode) these states. Once DL models have 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 their learned mappings between mental states and brain activity. Here, we compare the explanations of prominent interpretation methods for the mental state decoding decisions of DL models trained on three functional Magnetic Resonance Imaging (fMRI) datasets. We find that interpretation methods that capture the model's decision process well, by producing faithful explanations, generally produce explanations that are less in line with the results of standard analyses of the fMRI data, when compared to the explanations of interpretation methods with less explanation faithfulness. Specifically, we find that interpretation methods that focus on how sensitively a model's decoding decision changes with the values of the input produce explanations that better match with the results of a standard general linear model analysis of the fMRI data, while interpretation methods that focus on identifying the specific contribution of an input feature's value to the decoding decision produce overall more faithful explanations that align less well with the results of standard analyses of the fMRI data.

FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness

Transformers are slow and memory-hungry on long sequences, since the time and memory complexity of self-attention are quadratic in sequence length. Approximate attention methods have attempted to address this problem by trading off model quality to reduce the compute complexity, but often do not achieve wall-clock speedup. We argue that a missing principle is making attention algorithms IO-aware -- accounting for reads and writes between levels of GPU memory. We propose FlashAttention, an IO-aware exact attention algorithm that uses tiling to reduce the number of memory reads/writes between GPU high bandwidth memory (HBM) and GPU on-chip SRAM. We analyze the IO complexity of FlashAttention, showing that it requires fewer HBM accesses than standard attention, and is optimal for a range of SRAM sizes. We also extend FlashAttention to block-sparse attention, yielding an approximate attention algorithm that is faster than any existing approximate attention method. FlashAttention trains Transformers faster than existing baselines: 15% end-to-end wall-clock speedup on BERT-large (seq. length 512) compared to the MLPerf 1.1 training speed record, 3$\times$ speedup on GPT-2 (seq. length 1K), and 2.4$\times$ speedup on long-range arena (seq. length 1K-4K). FlashAttention and block-sparse FlashAttention enable longer context in Transformers, yielding higher quality models (0.7 better perplexity on GPT-2 and 6.4 points of lift on long-document classification) and entirely new capabilities: the first Transformers to achieve better-than-chance performance on the Path-X challenge (seq. length 16K, 61.4% accuracy) and Path-256 (seq. length 64K, 63.1% accuracy).

Can Foundation Models Help Us Achieve Perfect Secrecy?

A key promise of machine learning is the ability to assist users with personal tasks. Because the personal context required to make accurate predictions is often sensitive, we require systems that protect privacy. A gold standard privacy-preserving system will satisfy perfect secrecy, meaning that interactions with the system provably reveal no additional private information to adversaries. This guarantee should hold even as we perform multiple personal tasks over the same underlying data. However, privacy and quality appear to be in tension in existing systems for personal tasks. Neural models typically require lots of training to perform well, while individual users typically hold a limited scale of data, so the systems propose to learn from the aggregate data of multiple users. This violates perfect secrecy and instead, in the last few years, academics have defended these solutions using statistical notions of privacy -- i.e., the probability of learning private information about a user should be reasonably low. Given the vulnerabilities of these solutions, we explore whether the strong perfect secrecy guarantee can be achieved using recent zero-to-few sample adaptation techniques enabled by foundation models. In response, we propose FOCUS, a framework for personal tasks. Evaluating on popular privacy benchmarks, we find the approach, satisfying perfect secrecy, competes with strong collaborative learning baselines on 6 of 7 tasks. We empirically analyze the proposal, highlighting the opportunities and limitations across task types, and model inductive biases and sizes.

Can Foundation Models Wrangle Your Data?

Foundation Models (FMs) are models trained on large corpora of data that, at very large scale, can generalize to new tasks without any task-specific finetuning. As these models continue to grow in size, innovations continue to push the boundaries of what these models can do on language and image tasks. This paper aims to understand an underexplored area of FMs: classical data tasks like cleaning and integration. As a proof-of-concept, we cast three data cleaning and integration tasks as prompting tasks and evaluate the performance of FMs on these tasks. We find that large FMs generalize and achieve SoTA performance on data cleaning and integration tasks, even though they are not trained for these data tasks. We identify specific research challenges and opportunities that these models present, including challenges with private and temporal data, and opportunities to make data driven systems more accessible to non-experts. We make our code and experiments publicly available at: https://github.com/HazyResearch/fm_data_tasks.