USC at ICLR 2025

Researchers from the USC Viterbi School of Engineering are presenting 24 papers at the 2025 International Conference on Learning Representations (ICLR), Apr. 24-28, one of the premier global conferences advancing representation learning—commonly known as deep learning. This year’s work includes contributions from faculty and students across the School of Advanced Computing, a unit within Viterbi, notably the Thomas Lord Department of Computer Science and the Ming Hsieh Department of Electrical and Computer Engineering.

Computer Science Assistant Professor Yue Wang and his students are presenting four papers, including a prestigious oral presentation for PhysBench: Benchmarking and Enhancing Vision-Language Models for Physical World Understanding, underscoring USC’s growing leadership in foundational AI research. Additionally, USC is well-represented with several spotlight papers, showcasing key breakthroughs in areas ranging from deep learning to social aspects and applications in health, computer vision, and neuroscience, reflecting the schools’ cross-cutting strengths and commitment to real-world impact.

RESEARCH

ORAL papers were accepted at a rate of just under 2% while SPOTLIGHT papers were in the top 3% of accepted papers in 2025

Applied AI

Computer Vision

LoRA3D: Low-Rank Self-Calibration of 3D Geometric Foundation Models (SPOTLIGHT)

Ziqi Lu, Heng Yang, Danfei Xu, Boyi Li, Boris Ivanovic, Marco Pavone, Yue Wang

Emerging 3D geometric foundation models, such as DUSt3R, offer a promising approach for in-the-wild 3D vision tasks. However, due to the high-dimensional nature of the problem space and scarcity of high-quality 3D data, these pre-trained models still struggle to generalize to many challenging circumstances, such as limited view overlap or low lighting. To address this, we propose LoRA3D, an efficient self-calibration pipeline to specialize the pre-trained models to target scenes using their own multi-view predictions. Taking sparse RGB images as input, we leverage robust optimization techniques to refine multi-view predictions and align them into a global coordinate frame. In particular, we incorporate prediction confidence into the geometric optimization process, automatically re-weighting the confidence to better reflect point estimation accuracy. We use the calibrated confidence to generate high-quality pseudo labels for the calibrating views and fine-tune the models using low-rank adaptation (LoRA) on the pseudo-labeled data. Our method does not require any external priors or manual labels. It completes the self-calibration process on a single standard GPU within just 5 minutes. Each low-rank adapter requires only 18MB of storage. We evaluated our method on more than 160 scenes from the Replica, TUM and Waymo Open datasets, achieving up to 88% performance improvement on 3D reconstruction, multi-view pose estimation and novel-view rendering.

OmniRe: Omni Urban Scene Reconstruction (SPOTLIGHT)

Ziyu Chen, Jiawei Yang, Jiahui Huang, Riccardo de Lutio, Janick Martinez Esturo, Boris Ivanovic, Or Litany, Zan Gojcic, Sanja Fidler, Marco Pavone, Li Song, Yue Wang

We introduce OmniRe, a comprehensive system for efficiently creating high-fidelity digital twins of dynamic real-world scenes from on-device logs. Recent methods using neural fields or Gaussian Splatting primarily focus on vehicles, hindering a holistic framework for all dynamic foregrounds demanded by downstream applications, e.g., the simulation of human behavior. OmniRe extends beyond vehicle modeling to enable accurate, full-length reconstruction of diverse dynamic objects in urban scenes. Our approach builds scene graphs on 3DGS and constructs multiple Gaussian representations in canonical spaces that model various dynamic actors, including vehicles, pedestrians, cyclists, and others. OmniRe allows holistically reconstructing any dynamic object in the scene, enabling advanced simulations (~60 Hz) that include human-participated scenarios, such as pedestrian behavior simulation and human-vehicle interaction. This comprehensive simulation capability is unmatched by existing methods. Extensive evaluations on the Waymo dataset show that our approach outperforms prior state-of-the-art methods quantitatively and qualitatively by a large margin. We further extend our results to 5 additional popular driving datasets to demonstrate its generalizability on common urban scenes.

SC-OmniGS: Self-Calibrating Omnidirectional Gaussian Splatting

Huajian Huang, Yingshu Chen, Longwei Li, Hui Cheng, Tristan Braud, Yajie Zhao, Sai-Kit Yeung

360-degree cameras streamline data collection for radiance field 3D reconstruction by capturing comprehensive scene data. However, traditional radiance field methods do not address the specific challenges inherent to 360-degree images. We present SC-OmniGS, a novel self-calibrating omnidirectional Gaussian splatting system for fast and accurate omnidirectional radiance field reconstruction using 360-degree images. Rather than converting 360-degree images to cube maps and performing perspective image calibration, we treat 360-degree images as a whole sphere and derive a mathematical framework that enables direct omnidirectional camera pose calibration accompanied by 3D Gaussians optimization. Furthermore, we introduce a differentiable omnidirectional camera model in order to rectify the distortion of real-world data for performance enhancement. Overall, the omnidirectional camera intrinsic model, extrinsic poses, and 3D Gaussians are jointly optimized by minimizing weighted spherical photometric loss. Extensive experiments have demonstrated that our proposed SC-OmniGS is able to recover a high-quality radiance field from noisy camera poses or even no pose prior in challenging scenarios characterized by wide baselines and non-object-centric configurations. The noticeable performance gain in the real-world dataset captured by consumer-grade omnidirectional cameras verifies the effectiveness of our general omnidirectional camera model in reducing the distortion of 360-degree images.

Neuroscience/Cognitive Science

Probabilistic Geometric Principal Component Analysis (SPOTLIGHT)

Han-Lin Hsieh, Maryam Shanechi 

Dimensionality reduction is critical across various domains of science including neuroscience. Probabilistic Principal Component Analysis (PPCA) is a prominent dimensionality reduction method that provides a probabilistic approach unlike the deterministic approach of PCA and serves as a connection between PCA and Factor Analysis (FA). Despite their power, PPCA and its extensions are mainly based on linear models and can only describe the data in a Euclidean coordinate system around the mean of data. However, in many neuroscience applications, data may be distributed around a nonlinear geometry (i.e., manifold) rather than lying in the Euclidean space around the mean. We develop Probabilistic Geometric Principal Component Analysis (PGPCA) for such datasets as a new dimensionality reduction algorithm that can explicitly incorporate knowledge about a given nonlinear manifold that is first fitted from these data. Further, we show how in addition to the Euclidean coordinate system, a geometric coordinate system can be derived for the manifold to capture the deviations of data from the manifold and noise. We also derive a data-driven EM algorithm for learning the PGPCA model parameters. As such, PGPCA generalizes PPCA to better describe data distributions by incorporating a nonlinear manifold geometry. In simulations and brain data analyses, we show that PGPCA can effectively model the data distribution around various given manifolds and outperforms PPCA for such data. Moreover, PGPCA provides the capability to test whether the new geometric coordinate system better describes the data than the Euclidean one. Finally, PGPCA can perform dimensionality reduction and learn the data distribution both around and on the manifold. These capabilities make PGPCA valuable for enhancing the efficacy of dimensionality reduction for analysis of high-dimensional data that exhibit noise and are distributed around a nonlinear manifold, especially for neural data.

Health

Scaling Wearable Foundation Models

Girish Narayanswamy, Xin Liu, Kumar Ayush, Yuzhe Yang, Xuhai Xu, Shun Liao, Jake Garrison, Shyam Tailor, Jacob Sunshine, Yun Liu, Tim Althoff, Shrikanth Narayanan, Pushmeet Kohli, Jiening Zhan, Mark Malhotra, Shwetak Patel, Samy Abdel-Ghaffar, Daniel McDuff

Wearable sensors have become ubiquitous thanks to a variety of health tracking features. The resulting continuous and longitudinal measurements from everyday life generate large volumes of data. However, making sense of these observations for scientific and actionable insights is non-trivial. Inspired by the empirical success of generative modeling, where large neural networks learn powerful representations from vast amounts of text, image, video, or audio data, we investigate the scaling properties of wearable sensor foundation models across compute, data, and model size. Using a dataset of up to 40 million hours of in-situ heart rate, heart rate variability, accelerometer, electrodermal activity, skin temperature, and altimeter per-minute data from over 165,000 people, we create LSM, a multimodal foundation model built on the largest wearable-signals dataset with the most extensive range of sensor modalities to date. Our results establish the scaling laws of LSM for tasks such as imputation, interpolation and extrapolation across both time and sensor modalities. Moreover, we highlight how LSM enables sample-efficient downstream learning for tasks including exercise and activity recognition.

MediConfusion: Can you trust your AI radiologist? Probing the reliability of multimodal medical foundation models

Mohammad Shahab Sepehri, Zalan Fabian, Maryam Soltanolkotabi, Mahdi Soltanolkotabi

Multimodal Large Language Models (MLLMs) have tremendous potential to improve the accuracy, availability, and cost-effectiveness of healthcare by providing automated solutions or serving as aids to medical professionals. Despite promising first steps in developing medical MLLMs in the past few years, their capabilities and limitations are not well understood. Recently, many benchmark datasets have been proposed that test the general medical knowledge of such models across a variety of medical areas. However, the systematic failure modes and vulnerabilities of such models are severely underexplored with most medical benchmarks failing to expose the shortcomings of existing models in this safety-critical domain. In this paper, we introduce MediConfusion, a challenging medical Visual Question Answering (VQA) benchmark dataset, that probes the failure modes of medical MLLMs from a vision perspective. We reveal that state-of-the-art models are easily confused by image pairs that are otherwise visually dissimilar and clearly distinct for medical experts. Strikingly, all available models (open-source or proprietary) achieve performance below random guessing on MediConfusion, raising serious concerns about the reliability of existing medical MLLMs for healthcare deployment. We also extract common patterns of model failure that may help the design of a new generation of more trustworthy and reliable MLLMs in healthcare.

Robotics, Autonomy, Planning

SRSA: Skill Retrieval and Adaptation for Robotic Assembly Tasks (SPOTLIGHT)

Yijie Guo, Bingjie Tang, Iretiayo Akinola, Dieter Fox, Abhishek Gupta, Yashraj Narang

Enabling robots to learn novel tasks in a data-efficient manner is a long-standing challenge. Common strategies involve carefully leveraging prior experiences, especially transition data collected on related tasks. Although much progress has been made for general pick-and-place manipulation, far fewer studies have investigated contact-rich assembly tasks, where precise control is essential. We introduce SRSA (Skill Retrieval and Skill Adaptation), a novel framework designed to address this problem by utilizing a pre-existing skill library containing policies for diverse assembly tasks. The challenge lies in identifying which skill from the library is most relevant for fine-tuning on a new task. Our key hypothesis is that skills showing higher zero-shot success rates on a new task are better suited for rapid and effective fine-tuning on that task. To this end, we propose to predict the transfer success for all skills in the skill library on a novel task, and then use this prediction to guide the skill retrieval process. We establish a framework that jointly captures features of object geometry, physical dynamics, and expert actions to represent the tasks, allowing us to efficiently learn the transfer success predictor. Extensive experiments demonstrate that SRSA significantly outperforms the leading baseline. When retrieving and fine-tuning skills on unseen tasks, SRSA achieves a 19% relative improvement in success rate, exhibits 2.6x lower standard deviation across random seeds, and requires 2.4x fewer transition samples to reach a satisfactory success rate, compared to the baseline. In a continual learning setup, SRSA efficiently learns policies for new tasks and incorporates them into the skill library, enhancing future policy learning. Furthermore, policies trained with SRSA in simulation achieve a 90% mean success rate when deployed in the real world.

Deep Learning

Generative Models and Autoencoders

DreamDistribution: Prompt Distribution Learning for Text-to-Image Diffusion Models

Brian Nlong Zhao, Yuhang Xiao, Jiashu Xu, Xinyang Jiang , Yifan Yang , Dongsheng Li, Laurent Itti, Vibhav Vineet , Yunhao Ge

The popularization of Text-to-Image (T2I) diffusion models enables the generation of high-quality images from text descriptions. However, generating diverse customized images with reference visual attributes remains challenging. This work focuses on personalizing T2I diffusion models at a more abstract concept or category level, adapting commonalities from a set of reference images while creating new instances with sufficient variations. We introduce a solution that allows a pretrained T2I diffusion model to learn a set of soft prompts, enabling the generation of novel images by sampling prompts from the learned distribution. These prompts offer text-guided editing capabilities and additional flexibility in controlling variation and mixing between multiple distributions. We also show the adaptability of the learned prompt distribution to other tasks, such as text-to-3D. Finally we demonstrate effectiveness of our approach through quantitative analysis including automatic evaluation and human assessment.

MLLMs Know Where to Look: Training-free Perception of Small Visual Details with Multimodal LLMs

Jiarui Zhang, Mahyar Khayatkhoei, Prateek Chhikara, Filip Ilievski

Multimodal Large Language Models (MLLMs) have experienced rapid progress in visual recognition tasks in recent years. Given their potential integration into many critical applications, it is important to understand the limitations of their visual perception. In this work, we study whether MLLMs can perceive small visual details as effectively as large ones when answering questions about images. We observe that their performance is very sensitive to the size of the visual subject of the question, and further show that this effect is in fact causal by conducting an intervention study. Next, we study the attention patterns of MLLMs when answering visual questions, and intriguingly find that they consistently know where to look, even when they provide the wrong answer. Based on these findings, we then propose training-free visual intervention methods that leverage the internal knowledge of any MLLM itself, in the form of attention and gradient maps, to enhance its perception of small visual details. We evaluate our proposed methods on two widely-used MLLMs and seven visual question answering benchmarks and show that they can significantly improve MLLMs’ accuracy without requiring any training. Our results elucidate the risk of applying MLLMs to visual recognition tasks concerning small details and indicate that visual intervention using the model’s internal state is a promising direction to mitigate this risk.

Revisiting Energy Based Models as Policies: Ranking Noise Contrastive Estimation and Interpolating Energy Models

Sumeet Singh, Vikas Sindhwani, Stephen Tu

A crucial design decision for any robot learning pipeline is the choice of policy representation: what type of model should be used to generate the next set of robot actions? Owing to the inherent multi-modal nature of many robotic tasks, combined with the recent successes in generative modeling, researchers have turned to state-of-the-art probabilistic models such as diffusion models for policy representation. In this work, we revisit the choice of energy-based models (EBM) as a policy class.

We show that the prevailing folklore—that energy models in high dimensional continuous spaces are impractical to train—is false. We develop a practical training objective and algorithm for energy models which combines several key ingredients: (i) ranking noise contrastive estimation (R-NCE), (ii) learnable negative samplers, and (iii) non-adversarial joint training. We prove that our proposed objective function is asymptotically consistent and quantify its limiting variance. On the other hand, we show that the Implicit Behavior Cloning (IBC) objective is actually biased even at the population level, providing a mathematical explanation for the poor performance of IBC trained energy policies in several independent follow-up works. We further extend our algorithm to learn a continuous stochastic process that bridges noise and data, modeling this process with a family of EBMs indexed by scale variable. In doing so, we demonstrate that the core idea behind recent progress in generative modeling is actually compatible with EBMs. Altogether, our proposed training algorithms enable us to train energy-based models as policies which compete with—and even outperform—diffusion models and other state-of-the-art approaches in several challenging multi-modal benchmarks: obstacle avoidance path planning and contact-rich block pushing.

Shallow diffusion networks provably learn hidden low-dimensional structure

Nicholas Boffi, Arthur Jacot, Stephen Tu, Ingvar Ziemann

Diffusion-based generative models provide a powerful framework for learning to sample from a complex target distribution. The remarkable empirical success of these models applied to high-dimensional signals, including images and video, stands in stark contrast to classical results highlighting the curse of dimensionality for distribution recovery. In this work, we take a step towards understanding this gap through a careful analysis of learning diffusion models over the Barron space of single hidden layer neural networks. In particular, we show that these shallow models provably adapt to simple forms of low-dimensional structure, such as an unknown linear subspace or hidden independence, thereby avoiding the curse of dimensionality. We combine our results with recent analyses of sampling with diffusions to provide an end-to-end sample complexity bound for learning to sample from structured distributions. Importantly, our results do not require specialized architectures tailored to particular latent structures, and instead rely on the low-index structure of the Barron space to adapt to the underlying distribution.

Robustness

Targeted Attack Improves Protection against Unauthorized Diffusion Customization (SPOTLIGHT)

Boyang Zheng, Chumeng Liang, Xiaoyu Wu

Diffusion models build a new milestone for image generation yet raising public concerns, for they can be fine-tuned on unauthorized images for customization. Protection based on adversarial attacks rises to encounter this unauthorized diffusion customization, by adding protective watermarks to images and poisoning diffusion models. However, current protection, leveraging untargeted attacks, does not appear to be effective enough. In this paper, we propose a simple yet effective improvement for the protection against unauthorized diffusion customization by introducing targeted attacks. We show that by carefully selecting the target, targeted attacks significantly outperform untargeted attacks in poisoning diffusion models and degrading the customization image quality. Extensive experiments validate the superiority of our method on two mainstream customization methods of diffusion models, compared to existing protections. To explain the surprising success of targeted attacks, we delve into the mechanism of attack-based protections and propose a hypothesis based on our observation, which enhances the comprehension of attack-based protections. To the best of our knowledge, we are the first to both reveal the vulnerability of diffusion models to targeted attacks and leverage targeted attacks to enhance protection against unauthorized diffusion customization.

Large Language Models

LiveBench: A Challenging, Contamination-Free LLM Benchmark (SPOTLIGHT)

Colin White, Samuel Dooley, Manley Roberts, Arka Pal, Benjamin Feuer, Siddhartha Jain, Ravid Shwartz-Ziv, Neel Jain, Khalid Saifullah, Siddartha Naidu, Chinmay Hegde, Yann LeCun, Tom Goldstein, Willie Neiswanger, Micah Goldblum

Test set contamination, wherein test data from a benchmark ends up in a newer model’s training set, is a well-documented obstacle for fair LLM evaluation and can quickly render benchmarks obsolete. To mitigate this, many recent benchmarks crowdsource new prompts and evaluations from human or LLM judges; however, these can introduce significant biases, and break down when scoring hard questions. In this work, we introduce a new benchmark for LLMs designed to be immune to both test set contamination and the pitfalls of LLM judging and human crowdsourcing. We release LiveBench, the first benchmark that (1) contains frequently-updated questions from recent information sources, (2) scores answers automatically according to objective ground-truth values, and (3) contains a wide variety of challenging tasks, spanning math, coding, reasoning, language, instruction following, and data analysis. To achieve this, LiveBench contains questions that are based on recently-released math competitions, arXiv papers, news articles, and datasets, and it contains harder, contamination-free versions of tasks from previous benchmarks such as Big-Bench Hard, AMPS, and IFEval. We evaluate many prominent closed-source models, as well as dozens of open-source models ranging from 0.5B to 110B in size. LiveBench is difficult, with top models achieving below 65% accuracy. We release all questions, code, and model answers. Questions will be added and updated on a monthly basis, and we will release new tasks and harder versions of tasks over time so that LiveBench can distinguish between the capabilities of LLMs as they improve in the future. We welcome community engagement and collaboration for expanding the benchmark tasks and models.

The Semantic Hub Hypothesis: Language Models Share Semantic Representations Across Languages and Modalities

Zhaofeng Wu, Xinyan Velocity Yu, Dani Yogatama, Jiasen Lu, Yoon Kim

Modern language models can process inputs across diverse languages and modalities. We hypothesize that models acquire this capability through learning a shared representation space across heterogeneous data types (e.g., different languages and modalities), which places semantically similar inputs near one another, even if they are from different modalities/languages. We term this the semantic hub hypothesis, following the hub-and-spoke model from neuroscience (Patterson et al., 2007) which posits that semantic knowledge in the human brain is organized through a transmodal semantic “hub” which integrates information from various modality-specific “spokes” regions. We first show that model representations for semantically equivalent inputs in different languages are similar in the intermediate layers, and that this space can be interpreted using the model’s dominant pretraining language via the logit lens. This tendency extends to other data types, including arithmetic expressions, code, and visual/audio inputs. Interventions in the shared representation space in one data type also predictably affect model outputs in other data types, suggesting that this shared representations space is not simply a vestigial byproduct of large-scale training on broad data, but something that is actively utilized by the model during input processing.

 

DSBench: How Far Are Data Science Agents from Becoming Data Science Experts?

Liqiang Jing, Zhehui Huang, Xiaoyang Wang, Wenlin Yao, Wenhao Yu, Kaixin Ma, Hongming Zhangm Xinya Du, Dong Yu

Large Language Models (LLMs) and Large Vision-Language Models (LVLMs) have demonstrated impressive language/vision reasoning abilities, igniting the recent trend of building agents for targeted applications such as shopping assistants or AI software engineers. Recently, many data science benchmarks have been proposed to investigate their performance in the data science domain. However, existing data science benchmarks still fall short when compared to real-world data science applications due to their simplified settings. To bridge this gap, we introduce DSBench, a comprehensive benchmark designed to evaluate data science agents with realistic tasks. This benchmark includes 466 data analysis tasks and 74 data modeling tasks, sourced from Eloquence and Kaggle competitions. DSBench offers a realistic setting by encompassing long contexts, multimodal task backgrounds, reasoning with large data files and multi-table structures, and performing end-to-end data modeling tasks. Our evaluation of state-of-the-art LLMs, LVLMs, and agents shows that they struggle with most tasks, with the best agent solving only 34.12% of data analysis tasks and achieving a 34.74% Relative Performance Gap (RPG). These findings underscore the need for further advancements in developing more practical, intelligent, and autonomous data science agents.

 

DeLLMa: Decision Making Under Uncertainty with Large Language Models (SPOTLIGHT)

Ollie Liu, Deqing Fu, Dani Yogatama, Willie Neiswanger

The potential of large language models (LLMs) as decision support tools is increasingly being explored in fields such as business, engineering, and medicine, which often face challenging tasks of decision-making under uncertainty. In this paper, we show that directly prompting LLMs on these types of decision-making problems can yield poor results, especially as the problem complexity increases. To aid in these tasks, we propose DeLLMa (Decision-making Large Language Model assistant), a framework designed to enhance decision-making accuracy in uncertain environments. DeLLMa involves a multi-step reasoning procedure that integrates recent best practices in scaling inference-time reasoning, drawing upon principles from decision theory and utility theory, to provide an accurate and human-auditable decision-making process. We validate our procedure on multiple realistic decision-making environments, demonstrating that DeLLMa can consistently enhance the decision-making performance of leading language models, and achieve up to a 40% increase in accuracy over competing methods. Additionally, we show how performance improves when scaling compute at test time, and carry out human evaluations to benchmark components of DeLLMa.

Neuron-based Multifractal Analysis of Neuron Interaction Dynamics in Large Models

Xiongye Xiao, Heng Ping, Chenyu Zhou, Defu Cao, Yaxing Li, Yi-Zhuo Zhou, Shixuan Li, Nikos Kanakaris, Paul Bogdan

In recent years, there has been increasing attention on the capabilities of large-scale models, particularly in handling complex tasks that small-scale models are unable to perform. Notably, large language models (LLMs) have demonstrated “intelligent” abilities such as complex reasoning and abstract language comprehension, reflecting cognitive-like behaviors. However, current research on emergent abilities in large models predominantly focuses on the relationship between model performance and size, leaving a significant gap in the systematic quantitative analysis of the internal structures and mechanisms driving these emergent abilities. Drawing inspiration from neuroscience research on brain network structure and self-organization, we propose (i) a general network representation of large models, (ii) a new analytical framework — Neuron-based Multifractal Analysis (NeuroMFA) – for structural analysis, and (iii) a novel structure-based metric as a proxy for emergent abilities of large models. By linking structural features to the capabilities of large models, NeuroMFA provides a quantitative framework for analyzing emergent phenomena in large models. Our experiments show that the proposed method yields a comprehensive measure of the network’s evolving heterogeneity and organization, offering theoretical foundations and a new perspective for investigating emergence in large models.

Representation Learning

Multiview Equivariance Improves 3D Correspondence Understanding with Minimal Feature Finetuning

Yang You, Yixin Li, Congyue Deng, Yue Wang, Leonidas Guibas

Vision foundation models, particularly the ViT family, have revolutionized image understanding by providing rich semantic features. However, despite their success in 2D comprehension, their abilities on grasping 3D spatial relationships are still unclear. In this work, we evaluate and enhance the 3D awareness of ViT-based models. We begin by systematically assessing their ability to learn 3D equivariant features, specifically examining the consistency of semantic embeddings across different viewpoints. Our findings indicate that improved 3D equivariance leads to better performance on various downstream tasks, including pose estimation, tracking, and semantic transfer. Building on this insight, we propose a simple yet effective finetuning strategy based on 3D correspondences, which significantly enhances the 3D understanding of existing vision models. Remarkably, even finetuning on a single object for just one iteration results in substantial performance gains.

Transfer, Multitask and Meta-learning

MetaOOD: Automatic Selection of OOD Detection Models

Yuehan Qin,Yichi Zhang, Yi Nian, Xueying Ding, Yue Zhao

How can we automatically select an out-of-distribution (OOD) detection model for various underlying tasks? This is crucial for maintaining the reliability of open-world applications by identifying data distribution shifts, particularly in critical domains such as online transactions, autonomous driving, and real-time patient diagnosis. Despite the availability of numerous OOD detection methods, the challenge of selecting an optimal model for diverse tasks remains largely underexplored, especially in scenarios lacking ground truth labels. In this work, we introduce MetaOOD, the first zero-shot, unsupervised framework that utilizes meta-learning to select an OOD detection model automatically. As a meta-learning approach, MetaOOD leverages historical performance data of existing methods across various benchmark OOD detection datasets, enabling the effective selection of a suitable model for new datasets without the need for labeled data at the test time. To quantify task similarities more accurately, we introduce language model-based embeddings that capture the distinctive OOD characteristics of both datasets and detection models. Through extensive experimentation with 24 unique test dataset pairs to choose from among 11 OOD detection models, we demonstrate that MetaOOD significantly outperforms existing methods and only brings marginal time overhead. Our results, validated by Wilcoxon statistical tests, show that MetaOOD surpasses a diverse group of 11 baselines, including established OOD detectors and advanced unsupervised selection methods.

Interpretability and Explainable AI

Transformers Learn Low Sensitivity Functions: Investigations and Implications

Bhavya Vasudeva, Deqing Fu, Tianyi Zhou, Elliott Kau, Youqi Huang, Vatsal Sharan

Transformers achieve state-of-the-art accuracy and robustness across many tasks, but an understanding of their inductive biases and how those biases differ from other neural network architectures remains elusive. In this work, we identify the sensitivity of the model to token-wise random perturbations in the input as a unified metric which explains the inductive bias of transformers across different data modalities and distinguishes them from other architectures. We show that transformers have lower sensitivity1 than MLPs, CNNs, ConvMixers and LSTMs, across both vision and language tasks. We also show that this low-sensitivity bias has important implications: i) lower sensitivity correlates with improved robustness; it can also be used as an efficient intervention to further improve the robustness of transformers; ii) it corresponds to flatter minima in the loss landscape; and iii) it can serve as a progress measure for grokking. We support these findings with theoretical results showing (weak) spectral bias of transformers in the NTK regime, and improved robustness due to the lower sensitivity.

Attributing Culture-Conditioned Generations to Pretraining Corpora

Huihan Li, Arnav Goel, Keyu He, Xiang Ren

In open-ended generative tasks like narrative writing or dialogue, large language models often exhibit cultural biases, showing limited knowledge and generating templated outputs for less prevalent cultures. Recent works show that these biases may stem from uneven cultural representation in pretraining corpora. This work investigates how pretraining leads to biased culture-conditioned generations by analyzing how models associate entities with cultures based on pretraining data patterns. We propose the MEMOED framework (MEMOrization from prEtraining Document) to determine whether a generation for a culture arises from memorization. Using MEMOED on culture-conditioned generations about food and clothing for 110 cultures, we find that high-frequency cultures in pretraining data yield more generations with memorized symbols, while some low-frequency cultures produce none. Additionally, the model favors generating entities with extraordinarily high frequency regardless of the conditioned culture, reflecting biases toward frequent pretraining terms irrespective of relevance. We hope that the MEMOED framework and our insights will inspire more works on attributing model performance on pretraining data.

Physics-Informed Machine Learning

PhysBench: Benchmarking and Enhancing Vision-Language Models for Physical World Understanding (ORAL)

Wei Chow, Jiageng Mao, Boyi Li, Daniel Seita, Vitor Campagnolo Guizilini, Yue Wang

Understanding the physical world is a fundamental challenge in embodied AI, critical for enabling agents to perform complex tasks and operate safely in real-world environments. While Vision-Language Models (VLMs) have shown great promise in reasoning and task planning for embodied agents, their ability to comprehend physical phenomena remains extremely limited. To close this gap, we introduce PhysBench, a comprehensive benchmark designed to evaluate VLMs’ physical world understanding capability across a diverse set of tasks. PhysBench contains 10,002 entries of interleaved video-image-text data, categorized into four major domains: physical object properties, physical object relationships, physical scene understanding, and physics-based dynamics, further divided into 19 subclasses and 8 distinct capability dimensions. Our extensive experiments, conducted on 75 representative VLMs, reveal that while these models excel in common-sense reasoning, they struggle with understanding the physical world—likely due to the absence of physical knowledge in their training data and the lack of embedded physical priors. To tackle the shortfall, we introduce PhysAgent, a novel framework that combines the generalization strengths of VLMs with the specialized expertise of vision models, significantly enhancing VLMs’ physical understanding across a variety of tasks, including an 18.4% improvement on GPT-4o. Furthermore, our results demonstrate that enhancing VLMs’ physical world understanding capabilities can help embodied agents such as MOKA. We believe that PhysBench and PhysAgent offer valuable insights and contribute to bridging the gap between VLMs and physical world understanding

Privacy-Preserving Statistics and Machine Learning

DiSK: Differentially Private Optimizer with Simplified Kalman Filter for Noise Reduction

Xinwei Zhang, Zhiqi Bu, Borja Balle, Mingyi Hong, Meisam Razaviyayn, Vahab Mirrokni

Differential privacy (DP) offers a robust framework for safeguarding individual data privacy. To utilize DP in training modern machine learning models, differentially private optimizers have been widely used in recent years. A popular approach to privatize an optimizer is to clip the individual gradients and add sufficiently large noise to the clipped gradient. This approach led to the development of DP optimizers that have comparable performance with their non-private counterparts in fine-tuning tasks or in tasks with a small number of training parameters. However, a significant performance drop is observed when these optimizers are applied to large-scale training. This degradation stems from the substantial noise injection required to maintain DP, which disrupts the optimizer’s dynamics. This paper introduces DiSK, a novel framework designed to significantly enhance the performance of DP optimizers. DiSK employs Kalman filtering, a technique drawn from control and signal processing, to effectively denoise privatized gradients and generate progressively refined gradient estimations. To ensure practicality for large-scale training, we simplify the Kalman filtering process, minimizing its memory and computational demands. We establish theoretical privacy-utility trade-off guarantees for DiSK, and demonstrate provable improvements over standard DP optimizers like DPSGD in terms of iteration complexity upper-bound. Extensive experiments across diverse tasks, including vision tasks such as CIFAR-100 and ImageNet-1k and language fine-tuning tasks such as GLUE, E2E, and DART, validate the effectiveness of DiSK. The results showcase its ability to significantly improve the performance of DP optimizers, surpassing state-of-the-art results under the same privacy constraints on several benchmarks.

Social Aspects

Trustworthy Machine Learning

Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback

Stephen Casper, Xander Davies, Claudia Shi, Thomas Krendl Gilbert, Jérémy Scheurer, Javier Rando, Rachel Freedman, Tomasz Korbak, David Lindner, Pedro Freire, Tony Wang, Samuel Marks, Charbel-Raphaël Segerie, Micah Carroll, Andi Peng, Phillip Christoffersen, Mehul Damani, Stewart Slocum, Usman Anwar, Anand Siththaranjan, Max Nadeau, Eric J. Michaud, Jacob Pfau, Dmitrii Krasheninnikov, Xin Chen, Lauro Langosco, Peter Hase, Erdem Bıyık, Anca Dragan, David Krueger, Dorsa Sadigh, Dylan Hadfield-Menell

Reinforcement learning from human feedback (RLHF) is a technique for training AI systems to align with human goals. RLHF has emerged as the central method used to finetune state-of-the-art large language models (LLMs). Despite this popularity, there has been relatively little public work systematizing its flaws. In this paper, we (1) survey open problems and fundamental limitations of RLHF and related methods; (2) overview techniques to understand, improve, and complement RLHF in practice; and (3) propose auditing and disclosure standards to improve societal oversight of RLHF systems. Our work emphasizes the limitations of RLHF and highlights the importance of a multi-layered approach to the development of safer AI systems.

Theory

Game theory

Strategic Classification with Externalities

Safwan Hossain, Evi Micha, Yiling Chen, Ariel Procaccia

We propose a new variant of the strategic classification problem: a principal reveals a classifier, and n agents report their (possibly manipulated) features to be classified. Motivated by real-world applications, our model crucially allows the manipulation of one agent to affect another; that is, it explicitly captures inter-agent externalities. The principal-agent interactions are formally modeled as a Stackelberg game, with the resulting agent manipulation dynamics captured as a simultaneous game. We show that under certain assumptions, the pure Nash Equilibrium of this agent manipulation game is unique and can be efficiently computed. Leveraging this result, PAC learning guarantees are established for the learner: informally, we show that it is possible to learn classifiers that minimize loss on the distribution, even when a random number of agents are manipulating their way to a pure Nash Equilibrium. We also comment on the optimization of such classifiers through gradient-based approaches. This work sets the theoretical foundations for a more realistic analysis of classifiers that are robust against multiple strategic actors interacting in a common environment.

Last-Iterate Convergence Properties of Regret-Matching Algorithms in Games

Yang Cai, Gabriele Farina, Julien Grand-Clément, Christian Kroer, Chung-Wei Lee, Haipeng Luo, Weiqiang Zheng

Algorithms based on regret matching, specifically regret matching+ (RM+), and its variants are the most popular approaches for solving large-scale two-player zero-sum games in practice. Unlike algorithms such as optimistic gradient descent ascent, which have strong last-iterate and ergodic convergence properties for zero-sum games, virtually nothing is known about the last-iterate properties of regret-matching algorithms. Since last-iterate convergence is an attractive property both for numerical optimization reasons and because no-regret learning is viewed as a plausible method of real-world learning in games. In this paper, we study the last-iterate convergence properties of various popular variants of RM+. First, we show numerically that several practical variants such as simultaneous RM+, alternating RM+, and simultaneous predictive RM+, all lack last-iterate convergence guarantees even on a simple 3×3 game. Then, we go on to show that recent variants of these algorithms based on a smoothing technique do enjoy last-iterate convergence: we prove that extragradient RM+ and smooth PRM+ enjoy asymptotic last-iterate convergence (without a rate) and 1/t best-iterate convergence. Finally, we introduce restarted variants of these algorithms, and show that in both cases they enjoy linear-rate last-iterate convergence.