ICLR 2025 Papers

Non-convex Machine Learning problems typically do not adhere to the standard smoothness assumption. Based on empirical findings, Zhang et al. (2020b) proposed a more realistic generalized $(L_0,L_1)$-smoothness assumption, though it remains largely unexplored. Many existing algorithms designed for standard smooth problems need to be revised. However, in the context of Federated Learning, only a few works address this problem but rely on additional limiting assumptions. In this paper, we address this gap in the literature: we propose and analyze new methods with local steps, partial participation of clients, and Random Reshuffling without extra restrictive assumptions beyond generalized smoothness. The proposed methods are based on the proper interplay between clients' and server's stepsizes and gradient clipping. Furthermore, we perform the first analysis of these methods under the Polyak-Łojasiewicz condition. Our theory is consistent with the known results for standard smooth problems, and our experimental results support the theoretical insights.

Models of Virtual Cells and Virtual Tissues at single-cell resolution would allow us to test perturbations in silico and accelerate progress in tissue and cell engineering. However, most such models are not rooted in causal inference and as a result, could mistake correlation for causation. We introduce Celcomen, a novel generative graph neural network grounded in mathematical causality to disentangle intra- and inter-cellular gene regulation in spatial transcriptomics and single-cell data. Celcomen can also be prompted by perturbations to generate spatial counterfactuals, thus offering insights into experimentally inaccessible states, with potential applications in human health. We validate the model's disentanglement and identifiability through simulations, and demonstrate its counterfactual predictions in clinically relevant settings, including human glioblastoma and fetal spleen, recovering inflammation-related gene programs post immune system perturbation. Moreover, it supports mechanistic interpretability, as its parameters can be reverse-engineered from observed behavior, making it an accessible model for understanding both neural networks and complex biological systems.

Time-series forecasting is a challenging problem that traditionally requires specialized models custom-trained for the specific task at hand. Recently, inspired by the success of large language models, foundation models pre-trained on vast amounts of time-series data from diverse domains have emerged as a promising candidate for general-purpose time-series forecasting. The defining characteristic of these foundation models is their ability to perform zero-shot learning, that is, forecasting a new system from limited context data without explicit re-training or fine-tuning. Here, we evaluate whether the zero-shot learning paradigm extends to the challenging task of forecasting chaotic systems. Across 135 distinct chaotic dynamical systems and $10^8$ timepoints, we find that foundation models produce competitive forecasts compared to custom-trained models (including NBEATS, TiDE, etc.), particularly when training data is limited. Interestingly, even after point forecasts fail, large foundation models are able to preserve the geometric and statistical properties of the chaotic attractors. We attribute this success to foundation models' ability to perform in-context learning and identify context parroting as a simple mechanism used by these models to capture the long-term behavior of chaotic dynamical systems. Our results highlight the potential of foundation models as a tool for probing nonlinear and complex systems.

The video-conditioned policy takes prompt videos of the desired tasks as a condition and is regarded for its prospective generalizability. Despite its promise, training a video-conditioned policy is non-trivial due to the need for abundant demonstrations. In some tasks, the expert rollouts are merely available as videos, and costly and time-consuming efforts are required to annotate action labels. To address this, we explore training video-conditioned policy on a mixture of demonstrations and unlabeled expert videos to reduce reliance on extensive manual annotation. We introduce the Joint Embedding Predictive Transformer (JEPT) to learn a video-conditioned policy through sequence modeling. JEPT is designed to jointly learn visual transition prediction and inverse dynamics. The visual transition is captured from both demonstrations and expert videos, on the basis of which the inverse dynamics learned from demonstrations is generalizable to the tasks without action labels. Experiments on a series of simulated visual control tasks evaluate that JEPT can effectively leverage the mixture dataset to learn a generalizable policy. JEPT outperforms baselines in the tasks without action-labeled data and unseen tasks. We also experimentally reveal the potential of JEPT as a simple visual priors injection approach to enhance the video-conditioned policy.

Designing mechanically efficient geometry for architectural structures like shells, towers, and bridges, is an expensive iterative process. Existing techniques for solving such inverse problems rely on traditional optimization methods, which are slow and computationally expensive, limiting iteration speed and design exploration. Neural networks would seem to offer a solution via data-driven amortized optimization, but they often require extensive fine-tuning and cannot ensure that important design criteria, such as mechanical integrity, are met. In this work, we combine neural networks with a differentiable mechanics simulator to develop a model that accelerates the solution of shape approximation problems for architectural structures represented as bar systems. This model explicitly guarantees compliance with mechanical constraints while generating designs that closely match target geometries. We validate our approach in two tasks, the design of masonry shells and cable-net towers. Our model achieves better accuracy and generalization than fully neural alternatives, and comparable accuracy to direct optimization but in real time, enabling fast and reliable design exploration. We further demonstrate its advantages by integrating it into 3D modeling software and fabricating a physical prototype. Our work opens up new opportunities for accelerated mechanical design enhanced by neural networks for the built environment.

The observed similarities in the behavior of humans and Large Language Models (LLMs) have prompted researchers to consider the potential of using LLMs as models of human cognition. However, several significant challenges must be addressed before LLMs can be legitimately regarded as cognitive models. For instance, LLMs are trained on far more data than humans typically encounter, and may have been directly trained on human data in specific cognitive tasks or aligned with human preferences. Consequently, the origins of these behavioral similarities are not well understood. In this paper, we propose a novel way to enhance the utility of language models as cognitive models. This approach involves (i) leveraging computationally equivalent tasks that both a language model and a rational agent need to master for solving a cognitive problem and (ii) examining the specific task distributions required for a language model to exhibit human-like behaviors. We apply this approach to decision-making -- specifically risky and intertemporal choice -- where the key computationally equivalent task is the arithmetic of expected value calculations. We show that a small language model pretrained on an ecologically valid arithmetic dataset, which we call Arithmetic-GPT, predicts human behavior better than many traditional cognitive models. Pretraining language models on ecologically valid arithmetic datasets is sufficient to produce a strong correspondence between these models and human decision-making. Our results also suggest that language models used as cognitive models should be carefully investigated via ablation studies of the pretraining data.

As online video content rapidly grows, the task of text-video retrieval (TVR) becomes increasingly important. A key challenge in TVR is the information asymmetry between video and text: videos are inherently richer in information, while their textual descriptions often capture only fragments of this complexity. This paper introduces a novel, data-centric framework to bridge this gap by enriching textual representations to better match the richness of video content. During training, videos are segmented into event-level clips and captioned to ensure comprehensive coverage. During retrieval, a large language model (LLM) generates semantically diverse queries to capture a broader range of possible matches. To enhance retrieval efficiency, we propose a query selection mechanism that identifies the most relevant and diverse queries, reducing computational cost while improving accuracy. Our method achieves state-of-the-art results across multiple benchmarks, demonstrating the power of data-centric approaches in addressing information asymmetry in TVR. This work paves the way for new research focused on leveraging data to improve cross-modal retrieval.

Recent advances in language models have demonstrated their capability to solve mathematical reasoning problems, achieving near-perfect accuracy on grade-school level math benchmarks like GSM8K. In this paper, we formally study how language models solve these problems. We design a series of controlled experiments to address several fundamental questions: (1) Can language models truly develop reasoning skills, or do they simply memorize templates? (2) What is the model's hidden (mental) reasoning process? (3) Do models solve math questions using skills similar to or different from humans? (4) Do models trained on GSM8K-like datasets develop reasoning skills beyond those necessary for solving GSM8K problems? (5) What mental process causes models to make reasoning mistakes? (6) How large or deep must a model be to effectively solve GSM8K-level math questions? Our study uncovers many hidden mechanisms by which language models solve mathematical questions, providing insights that extend beyond current understandings of LLMs.

Recent advances in text-to-image generation have achieved impressive aesthetic quality, making these models usable for both personal and commercial purposes. However, in the fields of marketing and advertising, images are often created to be more engaging, as reflected in user behaviors such as increasing clicks, likes, and purchases, in addition to being aesthetically pleasing. To this end, we introduce the challenge of optimizing the image generation process for improved viewer engagement. In order to study image engagement and utility in real-world marketing scenarios, we collect *EngagingImageNet*, the first large-scale dataset of images, along with associated user engagement metrics. Further, we find that existing image evaluation metrics like aesthetics, CLIPScore, PickScore, ImageReward, *etc.* are unable to capture viewer engagement. To address the lack of reliable metrics for assessing image utility, we use the *EngagingImageNet* dataset to train *EngageNet*, an engagement-aware Vision Language Model (VLM) that predicts viewer engagement of images by leveraging contextual information about the tweet content, enterprise details, and posting time. We then explore methods to enhance the engagement of text-to-image models, making initial strides in this direction. These include conditioning image generation on improved prompts, supervised fine-tuning of stable diffusion on high-performing images, and reinforcement learning to align stable diffusion with *EngageNet*-based reward signals, all of which lead to the generation of images with higher viewer engagement. Finally, we propose the *Engagement Arena*, to benchmark text-to-image models based on their ability to generate engaging images, using *EngageNet* as the evaluator, thereby encouraging the research community to measure further advances in the engagement of text-to-image modeling. These contributions provide a new pathway for advancing utility-driven image generation, with significant implications for the commercial application of image generation. We have released our code and dataset on [behavior-in-the-wild.github.io/image-engagement](https://behavior-in-the-wild.github.io/image-engagement).

The development and evaluation of Large Language Models (LLMs) have largely focused on individual capabilities. However, this overlooks the intersection of multiple abilities across different types of expertise that are often required for real-world tasks, which we term **cross capabilities**. To systematically explore this concept, we first define seven core individual capabilities and then pair them to form seven common cross capabilities, each supported by a manually constructed taxonomy. Building on these definitions, we introduce *CrossEval*, a benchmark comprising 1,400 human-annotated prompts, with 100 prompts for each individual and cross capability. To ensure reliable evaluation, we involve expert annotators to assess 4,200 model responses, gathering 8,400 human ratings with detailed explanations to serve as reference examples. Our findings reveal that current LLMs consistently exhibit the ``Law of the Weakest Link,'' where cross-capability performance is significantly constrained by the weakest component. Across 58 cross-capability scores from 17 models, 38 scores are lower than all individual capabilities, while 20 fall between strong and weak, but closer to the weaker ability. These results highlight LLMs' underperformance in cross-capability tasks, emphasizing the need to identify and improve their weakest capabilities as a key research priority. The code, benchmarks, and evaluations are available on our [project website](https://www.llm-cross-capabilities.org).

Preconditioning is at the heart of iterative solutions of large, sparse linear systems of equations in scientific disciplines. Several algebraic approaches, which access no information beyond the matrix itself, are widely studied and used, but ill-conditioned matrices remain very challenging. We take a machine learning approach and propose using graph neural networks as a general-purpose preconditioner. They show attractive performance for many problems and can be used when the mainstream preconditioners perform poorly. Empirical evaluation on over 800 matrices suggests that the construction time of these graph neural preconditioners (GNPs) is more predictable and can be much shorter than that of other widely used ones, such as ILU and AMG, while the execution time is faster than using a Krylov method as the preconditioner, such as in inner-outer GMRES. GNPs have a strong potential for solving large-scale, challenging algebraic problems arising from not only partial differential equations, but also economics, statistics, graph, and optimization, to name a few.

Reinforcement learning (RL) faces two challenges: (1) The RL agent lacks explainability. (2) The trained RL agent is, in many cases, non-optimal and even far from optimal. To address the first challenge, explainable reinforcement learning (XRL) is proposed to explain the decision-making of the RL agent. In this paper, we demonstrate that XRL can also be used to address the second challenge, i.e., improve RL performance. Our method has two parts. The first part provides a two-level explanation for why the RL agent is not optimal by identifying the mistakes made by the RL agent. Since this explanation includes the mistakes of the RL agent, it has the potential to help correct the mistakes and thus improve RL performance. The second part formulates a constrained bi-level optimization problem to learn how to best utilize the two-level explanation to improve RL performance. In specific, the upper level learns how to use the high-level explanation to shape the reward so that the corresponding policy can maximize the cumulative ground truth reward, and the lower level learns the corresponding policy by solving a constrained RL problem formulated using the low-level explanation. We propose a novel algorithm to solve this constrained bi-level optimization problem, and theoretically guarantee that the algorithm attains global optimality. We use MuJoCo experiments to show that our method outperforms state-of-the-art baselines.

Large Language Models (LLMs) have showcased remarkable reasoning capabilities in various domains, yet face challenges in complex embodied tasks due to the need for a coherent long-term policy and context-sensitive environmental understanding. Previous work performed LLM refinement relying on outcome-supervised feedback, which can be costly and ineffective. In this work, we introduce a novel framework, Discriminator-Guided Action Optimization (DGAP), for facilitating the optimization of LLM action plans via step-wise signals. Specifically, we employ a limited set of demonstrations to enable the discriminator to learn a score function, which assesses the alignment between LLM-generated actions and the underlying optimal ones at every step. Based on the discriminator, LLMs are prompted to generate actions that maximize the score, utilizing historical action-score pair trajectories as guidance. Under mild conditions, DGAP resembles critic-regularized optimization and has been demonstrated to achieve a stronger policy than the LLM planner. In experiments across different LLMs (GPT-4, Llama3-70B) in ScienceWorld and VirtualHome, our method achieves superior performance and better efficiency than previous methods.

DETR-based methods, which use multi-layer transformer decoders to refine object queries iteratively, have shown promising performance in 3D indoor object detection. However, the scene point features in the transformer decoder remain fixed, leading to minimal contributions from later decoder layers, thereby limiting performance improvement. Recently, State Space Models (SSM) have shown efficient context modeling ability with linear complexity through iterative interactions between system states and inputs. Inspired by SSMs, we propose a new 3D object DEtection paradigm with an interactive STate space model (DEST). In the interactive SSM, we design a novel state-dependent SSM parameterization method that enables system states to effectively serve as queries in 3D indoor detection tasks. In addition, we introduce four key designs tailored to the characteristics of point cloud and SSM: The serialization and bidirectional scanning strategies enable bidirectional feature interaction among scene points within the SSM. The inter-state attention mechanism models the relationships between state points, while the gated feed-forward network enhances inter-channel correlations. To the best of our knowledge, this is the first method to model queries as system states and scene points as system inputs, which can simultaneously update scene point features and query features with linear complexity. Extensive experiments on two challenging datasets demonstrate the effectiveness of our DEST-based method. Our method improves the GroupFree baseline in terms of $\text{AP}_{50}$ on ScanNet V2 (+5.3) and SUN RGB-D (+3.2) datasets. Based on the VDETR baseline, Our method sets a new state-of-the-art on the ScanNetV2 and SUN RGB-D datasets.

Solving high wavenumber Helmholtz equations is notoriously challenging. Traditional solvers have yet to yield satisfactory results, and most neural network methods struggle to accurately solve cases with extremely high wavenumbers within heterogeneous media. This paper presents an advanced multigrid-hierarchical AI solver, tailored specifically for high wavenumber Helmholtz equations. We adapt the MGCNN architecture to align with the problem setting and incorporate a novel Fourier neural network (FNN) to match the characteristics of Helmholtz equations. FNN, mathematically akin to the convolutional neural network (CNN), enables faster propagation of source influence during the solve phase, making it particularly suitable for handling large size, high wavenumber problems. We conduct supervised learning tests against numerous neural operator learning methods to demonstrate the superior learning capabilities of our solvers. Additionally, we perform scalability tests using an unsupervised strategy to highlight our solvers' significant speedup over the most recent specialized AI solver and AI-enhanced traditional solver for high wavenumber Helmholtz equations. We also carry out an ablation study to underscore the effectiveness of the multigrid hierarchy and the benefits of introducing FNN. Notably, our solvers exhibit optimal convergence of $\mathcal{O}(k)$ up to $k \approx 2000$.

We introduce the "Belief State Transformer", a next-token predictor that takes both a prefix and suffix as inputs, with a novel objective of predicting both the next token for the prefix and the previous token for the suffix. The Belief State Transformer effectively learns to solve challenging problems that conventional forward-only transformers struggle with, in a domain-independent fashion. Key to this success is learning a compact belief state that captures all relevant information necessary for accurate predictions. Empirical ablations show that each component of the model is essential in difficult scenarios where standard Transformers fall short. For the task of story writing with known prefixes and suffixes, our approach outperforms the Fill-in-the-Middle method for reaching known goals and demonstrates improved performance even when the goals are unknown. Altogether, the Belief State Transformer enables more efficient goal-conditioned decoding, better test-time inference, and high-quality text representations on small scale problems. Website: https://edwhu.github.io/bst-website

Reinforcement learning with human feedback (RLHF) has become the dominant method to align large models to user preferences. Unlike fine-tuning, for which there are many studies regarding training data memorization, it is not clear how memorization is affected by or introduced in the RLHF alignment process. Understanding this relationship is important as real user data may be collected and used to align large models; if user data is memorized during RLHF and later regurgitated, this could raise privacy concerns. In addition to RLHF, other methods such as Direct Preference Optimization (DPO) and $\Psi$PO have gained popularity for learning directly from human preferences, removing the need for optimizing intermediary reward models with reinforcement learning. In this work, we analyze how training data memorization can surface and propagate through each phase of RLHF and direct preference learning. We focus our study on code completion models, as code completion is one of the most popular use cases for large language models. We find that RLHF significantly decreases the chance that data used for reward modeling and reinforcement learning is memorized in comparison to directly fine-tuning on this data, but that examples already memorized during the fine-tuning stage of RLHF, will, in the majority of cases, remain memorized after RLHF. In contrast, we find that aligning by learning directly from human preference data via a special case of $\Psi$PO, Identity Preference Optimization (IPO), increases the likelihood that training data is regurgitated compared to RLHF. Our work suggests that RLHF, as opposed to direct preference learning, is a safer way to mitigate the risk of regurgitating sensitive preference data when aligning large language models. We find our conclusions are robust across multiple code completion datasets, tasks, and model scales.

Quantum machine learning (QML) explores the potential advantages of quantum computers for machine learning tasks, with variational QML among the main current approaches. While quantum computers promise to solve problems that are classically intractable, it has been recently shown that a particular quantum algorithm which outperforms all pre-existing classical algorithms can be matched by a newly developed classical approach (often inspired by the quantum algorithm). We say such algorithms have been dequantized. For QML models to be effective, they must be trainable and non-dequantizable. The relationship between these properties is still not fully understood and recent works raised into question to what extent we could ever have QML models which are both trainable and non-dequantizable. This challenges the potential of QML altogether. In this work we answer open questions regarding when trainability and non-dequantization are compatible. We first formalize the key concepts and put them in the context of prior research. We introduce the role of "variationalness" of QML models using well-known quantum circuit architectures as leading examples. Our results provide recipes for variational QML models that are trainable and non-dequantizable. By ensuring that variational QML models are both trainable and non-dequantizable, we pave the way toward practical relevance.

Neural Combinatorial Optimization (NCO) methods have exhibited promising performance in solving Vehicle Routing Problems (VRPs). However, most NCO methods rely on the conventional self-attention mechanism that induces excessive computational complexity, thereby struggling to contend with large-scale VRPs and hindering their practical applicability. In this paper, we propose a lightweight cross-attention mechanism with linear complexity, by which a Transformer network is developed to learn efficient and favorable solutions for large-scale VRPs. We also propose a Self-Improved Training (SIT) algorithm that enables direct model training on large-scale VRP instances, bypassing extensive computational overhead for attaining labels. By iterating solution reconstruction, the Transformer network itself can generate improved partial solutions as pseudo-labels to guide the model training. Experimental results on the Travelling Salesman Problem (TSP) and the Capacitated Vehicle Routing Problem (CVRP) with up to 100K nodes indicate that our method consistently achieves superior performance for synthetic and real-world benchmarks, significantly boosting the scalability of NCO methods.