BZPEERPapers, people, evidence and research paths
Menu

Discover

SearchFind papers, people, topics, institutions and opportunities from one place.GraphSee how papers, authors, topics and evidence connect.

Research tools

HomeContinue from the papers, people and topics that matter now.CollectionsTurn saved papers into reading lists, evidence maps and shareable context.ResearchReconstruct the path that led you from search to paper to question.CollaborationMove from a paper, topic or expert request into a focused thread.PublishCreate full-text research articles with metadata, formulas and versions.

Network

ExpertsFind specialists whose work and reviews explain why they can help.CommunitiesJoin research groups organized around papers, topics and shared evidence.

Opportunities

ListingsBrowse roles, calls and collaborations connected to real research context.

Account

Log inReturn to your saved papers, graph views and active threads.Create accountStart saving papers, building a research trail and joining teams.
Log inCreate account

Source author record

Jacob Steinhardt

Jacob Steinhardt appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record
Machine LearningArtificial IntelligenceComputer VisionComputation and Languagemath.STStatistics Theorycs.CYeess.SPComputationComputer Science and Game TheoryCryptography and SecurityData Structures and AlgorithmsHuman-Computer Interaction

Catalog footprint

What is connected

20works
13topics
4close collaborators

Actions

Connect this record

Log in to claim
Open graphBrowse works

Research graph

See the researcher in context

Open full explorer

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

Published work

20 published item(s)

preprint2026arXiv

Log analysis is necessary for credible evaluation of AI agents

Agent benchmarks typically report only final outcomes: pass or fail. This threatens evaluation credibility in three ways. First, scores may be inflated or deflated by shortcuts and benchmark artifacts, misrepresenting capability. Second, benchmark performance may fail to predict real-world utility due to scaffold limitations and recurring failure modes. Finally, capability scores may conceal dangerous or catastrophic actions taken by the agent. We argue that log analysis -- the systematic tracking and analysis of the inputs, execution, and outputs of an AI agent -- is necessary to overcome these validity threats and promote credible agent evaluation. In this paper, we (1) present a taxonomy of threats to credible evaluation documented through log analysis, and (2) develop a set of guiding principles for log analysis. We illustrate these principles on tau-Bench Airline, revealing that pass^5 performance was under-elicited by nearly 50% and surfacing deployment failure modes invisible to outcome metrics. We conclude with pragmatic recommendations to increase uptake of log analysis, directed at diverse stakeholders including benchmark creators, model developers, independent evaluators, and deployers.

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

Describing Differences between Text Distributions with Natural Language

How do two distributions of texts differ? Humans are slow at answering this, since discovering patterns might require tediously reading through hundreds of samples. We propose to automatically summarize the differences by "learning a natural language hypothesis": given two distributions $D_{0}$ and $D_{1}$, we search for a description that is more often true for $D_{1}$, e.g., "is military-related." To tackle this problem, we fine-tune GPT-3 to propose descriptions with the prompt: "[samples of $D_{0}$] + [samples of $D_{1}$] + the difference between them is_____." We then re-rank the descriptions by checking how often they hold on a larger set of samples with a learned verifier. On a benchmark of 54 real-world binary classification tasks, while GPT-3 Curie (13B) only generates a description similar to human annotation 7% of the time, the performance reaches 61% with fine-tuning and re-ranking, and our best system using GPT-3 Davinci (175B) reaches 76%. We apply our system to describe distribution shifts, debug dataset shortcuts, summarize unknown tasks, and label text clusters, and present analyses based on automatically generated descriptions.

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

More Than a Toy: Random Matrix Models Predict How Real-World Neural Representations Generalize

Of theories for why large-scale machine learning models generalize despite being vastly overparameterized, which of their assumptions are needed to capture the qualitative phenomena of generalization in the real world? On one hand, we find that most theoretical analyses fall short of capturing these qualitative phenomena even for kernel regression, when applied to kernels derived from large-scale neural networks (e.g., ResNet-50) and real data (e.g., CIFAR-100). On the other hand, we find that the classical GCV estimator (Craven and Wahba, 1978) accurately predicts generalization risk even in such overparameterized settings. To bolster this empirical finding, we prove that the GCV estimator converges to the generalization risk whenever a local random matrix law holds. Finally, we apply this random matrix theory lens to explain why pretrained representations generalize better as well as what factors govern scaling laws for kernel regression. Our findings suggest that random matrix theory, rather than just being a toy model, may be central to understanding the properties of neural representations in practice.

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

PixMix: Dreamlike Pictures Comprehensively Improve Safety Measures

In real-world applications of machine learning, reliable and safe systems must consider measures of performance beyond standard test set accuracy. These other goals include out-of-distribution (OOD) robustness, prediction consistency, resilience to adversaries, calibrated uncertainty estimates, and the ability to detect anomalous inputs. However, improving performance towards these goals is often a balancing act that today's methods cannot achieve without sacrificing performance on other safety axes. For instance, adversarial training improves adversarial robustness but sharply degrades other classifier performance metrics. Similarly, strong data augmentation and regularization techniques often improve OOD robustness but harm anomaly detection, raising the question of whether a Pareto improvement on all existing safety measures is possible. To meet this challenge, we design a new data augmentation strategy utilizing the natural structural complexity of pictures such as fractals, which outperforms numerous baselines, is near Pareto-optimal, and roundly improves safety measures.

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

Predicting Out-of-Distribution Error with the Projection Norm

We propose a metric -- Projection Norm -- to predict a model's performance on out-of-distribution (OOD) data without access to ground truth labels. Projection Norm first uses model predictions to pseudo-label test samples and then trains a new model on the pseudo-labels. The more the new model's parameters differ from an in-distribution model, the greater the predicted OOD error. Empirically, our approach outperforms existing methods on both image and text classification tasks and across different network architectures. Theoretically, we connect our approach to a bound on the test error for overparameterized linear models. Furthermore, we find that Projection Norm is the only approach that achieves non-trivial detection performance on adversarial examples. Our code is available at https://github.com/yaodongyu/ProjNorm.

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

Scaling Out-of-Distribution Detection for Real-World Settings

Detecting out-of-distribution examples is important for safety-critical machine learning applications such as detecting novel biological phenomena and self-driving cars. However, existing research mainly focuses on simple small-scale settings. To set the stage for more realistic out-of-distribution detection, we depart from small-scale settings and explore large-scale multiclass and multi-label settings with high-resolution images and thousands of classes. To make future work in real-world settings possible, we create new benchmarks for three large-scale settings. To test ImageNet multiclass anomaly detectors, we introduce the Species dataset containing over 700,000 images and over a thousand anomalous species. We leverage ImageNet-21K to evaluate PASCAL VOC and COCO multilabel anomaly detectors. Third, we introduce a new benchmark for anomaly segmentation by introducing a segmentation benchmark with road anomalies. We conduct extensive experiments in these more realistic settings for out-of-distribution detection and find that a surprisingly simple detector based on the maximum logit outperforms prior methods in all the large-scale multi-class, multi-label, and segmentation tasks, establishing a simple new baseline for future work.

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Models

Reward hacking -- where RL agents exploit gaps in misspecified reward functions -- has been widely observed, but not yet systematically studied. To understand how reward hacking arises, we construct four RL environments with misspecified rewards. We investigate reward hacking as a function of agent capabilities: model capacity, action space resolution, observation space noise, and training time. More capable agents often exploit reward misspecifications, achieving higher proxy reward and lower true reward than less capable agents. Moreover, we find instances of phase transitions: capability thresholds at which the agent's behavior qualitatively shifts, leading to a sharp decrease in the true reward. Such phase transitions pose challenges to monitoring the safety of ML systems. To address this, we propose an anomaly detection task for aberrant policies and offer several baseline detectors.

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

Unsolved Problems in ML Safety

Machine learning (ML) systems are rapidly increasing in size, are acquiring new capabilities, and are increasingly deployed in high-stakes settings. As with other powerful technologies, safety for ML should be a leading research priority. In response to emerging safety challenges in ML, such as those introduced by recent large-scale models, we provide a new roadmap for ML Safety and refine the technical problems that the field needs to address. We present four problems ready for research, namely withstanding hazards ("Robustness"), identifying hazards ("Monitoring"), reducing inherent model hazards ("Alignment"), and reducing systemic hazards ("Systemic Safety"). Throughout, we clarify each problem's motivation and provide concrete research directions.

0 citations0 reviewsNo code linkedOpen access
preprint2022arXiv

What Would Jiminy Cricket Do? Towards Agents That Behave Morally

When making everyday decisions, people are guided by their conscience, an internal sense of right and wrong. By contrast, artificial agents are currently not endowed with a moral sense. As a consequence, they may learn to behave immorally when trained on environments that ignore moral concerns, such as violent video games. With the advent of generally capable agents that pretrain on many environments, it will become necessary to mitigate inherited biases from environments that teach immoral behavior. To facilitate the development of agents that avoid causing wanton harm, we introduce Jiminy Cricket, an environment suite of 25 text-based adventure games with thousands of diverse, morally salient scenarios. By annotating every possible game state, the Jiminy Cricket environments robustly evaluate whether agents can act morally while maximizing reward. Using models with commonsense moral knowledge, we create an elementary artificial conscience that assesses and guides agents. In extensive experiments, we find that the artificial conscience approach can steer agents towards moral behavior without sacrificing performance.

0 citations0 reviewsNo code linkedOpen access
preprint2021arXiv

Measuring Massive Multitask Language Understanding

We propose a new test to measure a text model's multitask accuracy. The test covers 57 tasks including elementary mathematics, US history, computer science, law, and more. To attain high accuracy on this test, models must possess extensive world knowledge and problem solving ability. We find that while most recent models have near random-chance accuracy, the very largest GPT-3 model improves over random chance by almost 20 percentage points on average. However, on every one of the 57 tasks, the best models still need substantial improvements before they can reach expert-level accuracy. Models also have lopsided performance and frequently do not know when they are wrong. Worse, they still have near-random accuracy on some socially important subjects such as morality and law. By comprehensively evaluating the breadth and depth of a model's academic and professional understanding, our test can be used to analyze models across many tasks and to identify important shortcomings.

0 citations0 reviewsNo code linkedOpen access
preprint2021arXiv

Natural Adversarial Examples

We introduce two challenging datasets that reliably cause machine learning model performance to substantially degrade. The datasets are collected with a simple adversarial filtration technique to create datasets with limited spurious cues. Our datasets' real-world, unmodified examples transfer to various unseen models reliably, demonstrating that computer vision models have shared weaknesses. The first dataset is called ImageNet-A and is like the ImageNet test set, but it is far more challenging for existing models. We also curate an adversarial out-of-distribution detection dataset called ImageNet-O, which is the first out-of-distribution detection dataset created for ImageNet models. On ImageNet-A a DenseNet-121 obtains around 2% accuracy, an accuracy drop of approximately 90%, and its out-of-distribution detection performance on ImageNet-O is near random chance levels. We find that existing data augmentation techniques hardly boost performance, and using other public training datasets provides improvements that are limited. However, we find that improvements to computer vision architectures provide a promising path towards robust models.

0 citations0 reviewsNo code linkedOpen access
preprint2020arXiv

Identifying Statistical Bias in Dataset Replication

Dataset replication is a useful tool for assessing whether improvements in test accuracy on a specific benchmark correspond to improvements in models' ability to generalize reliably. In this work, we present unintuitive yet significant ways in which standard approaches to dataset replication introduce statistical bias, skewing the resulting observations. We study ImageNet-v2, a replication of the ImageNet dataset on which models exhibit a significant (11-14%) drop in accuracy, even after controlling for a standard human-in-the-loop measure of data quality. We show that after correcting for the identified statistical bias, only an estimated $3.6\% \pm 1.5\%$ of the original $11.7\% \pm 1.0\%$ accuracy drop remains unaccounted for. We conclude with concrete recommendations for recognizing and avoiding bias in dataset replication. Code for our study is publicly available at http://github.com/MadryLab/dataset-replication-analysis .

0 citations0 reviewsNo code linkedOpen access
preprint2020arXiv

Robust estimation via generalized quasi-gradients

We explore why many recently proposed robust estimation problems are efficiently solvable, even though the underlying optimization problems are non-convex. We study the loss landscape of these robust estimation problems, and identify the existence of "generalized quasi-gradients". Whenever these quasi-gradients exist, a large family of low-regret algorithms are guaranteed to approximate the global minimum; this includes the commonly-used filtering algorithm. For robust mean estimation of distributions under bounded covariance, we show that any first-order stationary point of the associated optimization problem is an {approximate global minimum} if and only if the corruption level $ε< 1/3$. Consequently, any optimization algorithm that aproaches a stationary point yields an efficient robust estimator with breakdown point $1/3$. With careful initialization and step size, we improve this to $1/2$, which is optimal. For other tasks, including linear regression and joint mean and covariance estimation, the loss landscape is more rugged: there are stationary points arbitrarily far from the global minimum. Nevertheless, we show that generalized quasi-gradients exist and construct efficient algorithms. These algorithms are simpler than previous ones in the literature, and for linear regression we improve the estimation error from $O(\sqrtε)$ to the optimal rate of $O(ε)$ for small $ε$ assuming certified hypercontractivity. For mean estimation with near-identity covariance, we show that a simple gradient descent algorithm achieves breakdown point $1/3$ and iteration complexity $\tilde{O}(d/ε^2)$.

0 citations0 reviewsNo code linkedOpen access
preprint2020arXiv

When does the Tukey median work?

We analyze the performance of the Tukey median estimator under total variation (TV) distance corruptions. Previous results show that under Huber's additive corruption model, the breakdown point is 1/3 for high-dimensional halfspace-symmetric distributions. We show that under TV corruptions, the breakdown point reduces to 1/4 for the same set of distributions. We also show that a certain projection algorithm can attain the optimal breakdown point of 1/2. Both the Tukey median estimator and the projection algorithm achieve sample complexity linear in dimension.

0 citations0 reviewsNo code linkedOpen access
preprint2016arXiv

Avoiding Imposters and Delinquents: Adversarial Crowdsourcing and Peer Prediction

We consider a crowdsourcing model in which $n$ workers are asked to rate the quality of $n$ items previously generated by other workers. An unknown set of $αn$ workers generate reliable ratings, while the remaining workers may behave arbitrarily and possibly adversarially. The manager of the experiment can also manually evaluate the quality of a small number of items, and wishes to curate together almost all of the high-quality items with at most an $ε$ fraction of low-quality items. Perhaps surprisingly, we show that this is possible with an amount of work required of the manager, and each worker, that does not scale with $n$: the dataset can be curated with $\tilde{O}\Big(\frac{1}{βα^3ε^4}\Big)$ ratings per worker, and $\tilde{O}\Big(\frac{1}{βε^2}\Big)$ ratings by the manager, where $β$ is the fraction of high-quality items. Our results extend to the more general setting of peer prediction, including peer grading in online classrooms.

0 citations0 reviewsNo code linkedOpen access
preprint2016arXiv

Concrete Problems in AI Safety

Rapid progress in machine learning and artificial intelligence (AI) has brought increasing attention to the potential impacts of AI technologies on society. In this paper we discuss one such potential impact: the problem of accidents in machine learning systems, defined as unintended and harmful behavior that may emerge from poor design of real-world AI systems. We present a list of five practical research problems related to accident risk, categorized according to whether the problem originates from having the wrong objective function ("avoiding side effects" and "avoiding reward hacking"), an objective function that is too expensive to evaluate frequently ("scalable supervision"), or undesirable behavior during the learning process ("safe exploration" and "distributional shift"). We review previous work in these areas as well as suggesting research directions with a focus on relevance to cutting-edge AI systems. Finally, we consider the high-level question of how to think most productively about the safety of forward-looking applications of AI.

0 citations0 reviewsNo code linkedOpen access
preprint2016arXiv

Unsupervised Risk Estimation Using Only Conditional Independence Structure

We show how to estimate a model's test error from unlabeled data, on distributions very different from the training distribution, while assuming only that certain conditional independencies are preserved between train and test. We do not need to assume that the optimal predictor is the same between train and test, or that the true distribution lies in any parametric family. We can also efficiently differentiate the error estimate to perform unsupervised discriminative learning. Our technical tool is the method of moments, which allows us to exploit conditional independencies in the absence of a fully-specified model. Our framework encompasses a large family of losses including the log and exponential loss, and extends to structured output settings such as hidden Markov models.

0 citations0 reviewsNo code linkedOpen access
preprint2015arXiv

Learning Fast-Mixing Models for Structured Prediction

Markov Chain Monte Carlo (MCMC) algorithms are often used for approximate inference inside learning, but their slow mixing can be difficult to diagnose and the approximations can seriously degrade learning. To alleviate these issues, we define a new model family using strong Doeblin Markov chains, whose mixing times can be precisely controlled by a parameter. We also develop an algorithm to learn such models, which involves maximizing the data likelihood under the induced stationary distribution of these chains. We show empirical improvements on two challenging inference tasks.

0 citations0 reviewsNo code linkedOpen access
preprint2015arXiv

Reified Context Models

A classic tension exists between exact inference in a simple model and approximate inference in a complex model. The latter offers expressivity and thus accuracy, but the former provides coverage of the space, an important property for confidence estimation and learning with indirect supervision. In this work, we introduce a new approach, reified context models, to reconcile this tension. Specifically, we let the amount of context (the arity of the factors in a graphical model) be chosen "at run-time" by reifying it---that is, letting this choice itself be a random variable inside the model. Empirically, we show that our approach obtains expressivity and coverage on three natural language tasks.

0 citations0 reviewsNo code linkedOpen access
preprint2014arXiv

The Statistics of Streaming Sparse Regression

We present a sparse analogue to stochastic gradient descent that is guaranteed to perform well under similar conditions to the lasso. In the linear regression setup with irrepresentable noise features, our algorithm recovers the support set of the optimal parameter vector with high probability, and achieves a statistically quasi-optimal rate of convergence of Op(k log(d)/T), where k is the sparsity of the solution, d is the number of features, and T is the number of training examples. Meanwhile, our algorithm does not require any more computational resources than stochastic gradient descent. In our experiments, we find that our method substantially out-performs existing streaming algorithms on both real and simulated data.

0 citations0 reviewsNo code linkedOpen access