ICML 2025 Papers

Predictive tasks on relational databases are critical in real-world applications spanning e-commerce, healthcare, and social media. To address these tasks effectively, Relational Deep Learning (RDL) encodes relational data as graphs, enabling Graph Neural Networks (GNNs) to exploit relational structures for improved predictions. However, existing RDL methods often overlook the intrinsic structural properties of the graphs built from relational databases, leading to modeling inefficiencies, particularly in handling many-to-many relationships. Here we introduce RelGNN, a novel GNN framework specifically designed to leverage the unique structural characteristics of the graphs built from relational databases. At the core of our approach is the introduction of atomic routes, which are simple paths that enable direct single-hop interactions between the source and destination nodes. Building upon these atomic routes, RelGNN designs new composite message passing and graph attention mechanisms that reduce redundancy, highlight key signals, and enhance predictive accuracy. RelGNN is evaluated on 30 diverse real-world tasks from Relbench (Fey et al., 2024), and achieves state-of-the-art performance on the vast majority of tasks, with improvements of up to 25%.

We introduce $\textbf{STRING}$: Separable Translationally Invariant Position Encodings. STRING extends Rotary Position Encodings, a recently proposed and widely used algorithm in large language models, via a unifying theoretical framework. Importantly, STRING still provides $\textbf{exact}$ translation invariance, including token coordinates of arbitrary dimensionality, whilst maintaining a low computational footprint. These properties are especially important in robotics, where efficient 3D token representation is key. We integrate STRING into Vision Transformers with RGB(-D) inputs (color plus optional depth), showing substantial gains, e.g. in open-vocabulary object detection and for robotics controllers. We complement our experiments with a rigorous mathematical analysis, proving the universality of our methods. Videos of STRING-based robotics controllers can be found here: https://sites.google.com/view/string-robotics.

Spatial Transcriptomics (ST) is a method that captures gene expression profiles aligned with spatial coordinates. The discrete spatial distribution and the super-high dimensional sequencing results make ST data challenging to be modeled effectively. In this paper, we manage to model ST in a continuous and compact manner by the proposed tool, SUICA, empowered by the great approximation capability of Implicit Neural Representations (INRs) that can enhance both the spatial density and the gene expression. Concretely within the proposed SUICA, we incorporate a graph-augmented Autoencoder to effectively model the context information of the unstructured spots and provide informative embeddings that are structure-aware for spatial mapping. We also tackle the extremely skewed distribution in a regression-by-classification fashion and enforce classification-based loss functions for the optimization of SUICA. By extensive experiments of a wide range of common ST platforms under varying degradations, SUICA outperforms both conventional INR variants and SOTA methods regarding numerical fidelity, statistical correlation, and bio-conservation. The prediction by SUICA also showcases amplified gene signatures that enriches the bio-conservation of the raw data and benefits subsequent analysis.

To avoid myopic behavior, multi-step lookahead Bayesian optimization (BO) algorithms consider the sequential nature of BO and have demonstrated promising results in recent years. However, owing to the curse of dimensionality, most of these methods make significant approximations or suffer scalability issues. This paper presents a novel reinforcement learning (RL)-based framework for multi-step lookahead BO in high-dimensional black-box optimization problems. The proposed method enhances the scalability and decision-making quality of multi-step lookahead BO by efficiently solving the sequential dynamic program of the BO process in a near-optimal manner using RL. We first introduce an Attention-DeepSets encoder to represent the state of knowledge to the RL agent and subsequently propose a multi-task, fine-tuning procedure based on end-to-end (encoder-RL) on-policy learning. We evaluate the proposed method, EARL-BO (Encoder Augmented RL for BO), on synthetic benchmark functions and hyperparameter tuning problems, finding significantly improved performance compared to existing multi-step lookahead and high-dimensional BO methods.

Dynamic data selection aims to accelerate training with lossless performances. However, reducing training data inherently limits data diversity, potentially hindering generalization. While data augmentation is widely used to enhance diversity, it is typically not optimized in conjunction with selection. As a result, directly combining these techniques fails to fully exploit their synergies. To tackle the challenge, we propose a novel online data training framework that, for the first time, unifies dynamic data selection and augmentation, achieving both training efficiency and enhanced performance. Our method estimates each sample's joint distribution of local density and multimodal semantic consistency, allowing for the targeted selection of augmentation-suitable samples while suppressing the inclusion of noisy or ambiguous data. This enables a more significant reduction in dataset size without sacrificing model generalization. Experimental results demonstrate that our method outperforms existing state-of-the-art approaches on various benchmark datasets and architectures, e.g., reducing 50% training costs on ImageNet-1k with lossless performance. Furthermore, our approach enhances noise resistance and improves model robustness, reinforcing its practical utility in real-world scenarios.

Continual Instruction Tuning (CIT) is adopted to continually instruct Large Models to follow human intent data by data. It is observed that existing gradient update would heavily destroy the performance on previous datasets during CIT process. Instead, Exponential Moving Average (EMA), owns the ability to trace previous parameters, which can aid in decreasing forgetting. Nonetheless, its stable balance weight fails to deal with the ever-changing datasets, leading to the out-of-balance between plasticity and stability. In this paper, we propose a general continual instruction tuning framework to address the challenge. Starting from the trade-off prerequisite and EMA update, we propose the plasticity and stability ideal condition. Based on Taylor expansion in the loss function, we find the optimal balance weight can be automatically determined by the gradients and learned parameters. Therefore, we propose a stable-plasticity balanced coefficient to avoid knowledge interference. Based on the semantic similarity of the instructions, we can determine whether to retrain or expand the training parameters and allocate the most suitable parameters for the testing instances. Extensive experiments across multiple continual instruction tuning benchmarks demonstrate that our approach not only enhances anti-forgetting capabilities but also significantly improves overall continual tuning performance. Our code is available at https://github.com/JingyangQiao/CoIN.

A key component of transformers is the attention mechanism orchestrating how each token influences the propagation of every other token along the layers of a transformer. In this paper we provide a rigorous, mathematical analysis of the asymptotic properties of attention in transformers. Although we present several results based on different assumptions, all of them point to the same conclusion, all tokens asymptotically converge to each other, a phenomenon that has been empirically reported in the literature. Our findings are carefully compared with existing theoretical results and illustrated by simulations and experimental studies using the GPT-2 model.

Learning a latent dynamics model provides a task-agnostic representation of an agent's understanding of its environment. Leveraging this knowledge for model-based reinforcement learning (RL) holds the potential to improve sample efficiency over model-free methods by learning from imagined rollouts. Furthermore, because the latent space serves as input to behavior models, the informative representations learned by the world model facilitate efficient learning of desired skills. Most existing methods rely on holistic representations of the environment’s state. In contrast, humans reason about objects and their interactions, predicting how actions will affect specific parts of their surroundings. Inspired by this, we propose *Slot-Attention for Object-centric Latent Dynamics (SOLD)*, a novel model-based RL algorithm that learns object-centric dynamics models in an unsupervised manner from pixel inputs. We demonstrate that the structured latent space not only improves model interpretability but also provides a valuable input space for behavior models to reason over. Our results show that SOLD outperforms DreamerV3 and TD-MPC2 - state-of-the-art model-based RL algorithms - across a range of multi-object manipulation environments that require both relational reasoning and dexterous control. Videos and code are available at https:// slot-latent-dynamics.github.io.

Kernel method-based intensity estimators, formulated within reproducing kernel Hilbert spaces (RKHSs), and classical kernel intensity estimators (KIEs) have been among the most easy-to-implement and feasible methods for estimating the intensity functions of inhomogeneous Poisson processes. While both approaches share the term "kernel", they are founded on distinct theoretical principles, each with its own strengths and limitations. In this paper, we propose a novel regularized kernel method for Poisson processes based on the least squares loss and show that the resulting intensity estimator involves a specialized variant of the representer theorem: it has the dual coefficient of unity and coincides with classical KIEs. This result provides new theoretical insights into the connection between classical KIEs and kernel method-based intensity estimators, while enabling us to develop an efficient KIE by leveraging advanced techniques from RKHS theory. We refer to the proposed model as the *kernel method-based kernel intensity estimator* (K$^2$IE). Through experiments on synthetic datasets, we show that K$^2$IE achieves comparable predictive performance while significantly surpassing the state-of-the-art kernel method-based estimator in computational efficiency.

We study tensor completion (TC) through the lens of low-rank tensor decomposition (TD). Many TD algorithms use fast alternating minimization methods to solve _highly structured_ linear regression problems at each step (e.g., for CP, Tucker, and tensor-train decompositions). However, such algebraic structure is often lost in TC regression problems, making direct extensions unclear. This work proposes a novel _lifting_ method for approximately solving TC regression problems using structured TD regression algorithms as blackbox subroutines, enabling sublinear-time methods. We analyze the convergence rate of our approximate Richardson iteration-based algorithm, and our empirical study shows that it can be 100x faster than direct methods for CP completion on real-world tensors.

Soft Actor-Critic (SAC) has achieved notable success in continuous control tasks but struggles in sparse reward settings, where infrequent rewards make efficient exploration challenging. While novelty-based exploration methods address this issue by encouraging the agent to explore novel states, they are not trivial to apply to SAC. In particular, managing the interaction between novelty-based exploration and SAC’s stochastic policy can lead to inefficient exploration and redundant sample collection. In this paper, we propose KEA (Keeping Exploration Alive) which tackles the inefficiencies in balancing exploration strategies when combining SAC with novelty-based exploration. KEA integrates a novelty-augmented SAC with a standard SAC agent, proactively coordinated via a switching mechanism. This coordination allows the agent to maintain stochasticity in high-novelty regions, enhancing exploration efficiency and reducing repeated sample collection. We first analyze this potential issue in a 2D navigation task, and then evaluate KEA on the DeepSea hard-exploration benchmark as well as sparse reward control tasks from the DeepMind Control Suite. Compared to state-of-the-art novelty-based exploration baselines, our experiments show that KEA significantly improves learning efficiency and robustness in sparse reward setups.

Diffusion models have shown great promise in decision-making, also known as diffusion planning. However, the slow inference speeds limit their potential for broader real-world applications. Here, we introduce **Habi**, a general framework that transforms powerful but slow diffusion planning models into fast decision-making models, which mimics the cognitive process in the brain that costly goal-directed behavior gradually transitions to efficient habitual behavior with repetitive practice. Even using a laptop CPU, the habitized model can achieve an average **800+ Hz** decision-making frequency (faster than previous diffusion planners by orders of magnitude) on standard offline reinforcement learning benchmarks D4RL, while maintaining comparable or even higher performance compared to its corresponding diffusion planner. Our work proposes a fresh perspective of leveraging powerful diffusion models for real-world decision-making tasks. We also provide robust evaluations and analysis, offering insights from both biological and engineering perspectives for efficient and effective decision-making.

Recent studies have shown that supervised fine-tuning of LLMs on a small number of high-quality datasets can yield strong reasoning capabilities. However, full fine-tuning (Full FT), while powerful, is computationally expensive and susceptible to overfitting and catastrophic forgetting, particularly when data is limited. Sparse fine-tuning, which previously achieved notable success by updating only a small subset of model parameters, offers a promising trade-off between efficiency and effectiveness. Yet, it has lagged behind in the LLM era due to the difficulty of identifying parameters truly critical for reasoning. In this work, we state that weights with the largest magnitude after low-rank approximation are critical weights for fine-tuning, which we call *Principal Weights*. Surprisingly, while magnitude-based sparse fine-tuning performs poorly as a baseline on LLM fine-tuning, it becomes highly effective after rank reduction. These insights motivate our method: **L**ow-rank **I**nformed Sparse **F**ine-**T**uning ($\texttt{LIFT}$). $\texttt{LIFT}$ only updates the top 5% *Principal Weights* throughout training and consistently achieves better performance on reasoning tasks than Full FT, while maintaining memory efficiency on par with popular parameter-efficient fine-tuning methods. In addition to strong performance on target domains such as arithmetic reasoning, $\texttt{LIFT}$ also retains up to 20% more source-domain knowledge, compared to Full FT and LoRA. Our code is available at: https://github.com/zihanghliu/LIFT.

This paper investigates the convergence behavior of score-based graph generative models (SGGMs). Unlike common score-based generative models (SGMs) that are governed by a single stochastic differential equation (SDE), SGGMs utilize a system of dependent SDEs, where the graph structure and node features are modeled separately, while accounting for their inherent dependencies. This distinction makes existing convergence analyses from SGMs inapplicable for SGGMs. In this work, we present the first convergence analysis for SGGMs, focusing on the convergence bound (the risk of generative error) across three key graph generation paradigms: (1) feature generation with a fixed graph structure, (2) graph structure generation with fixed node features, and (3) joint generation of both graph structure and node features. Our analysis reveals several unique factors specific to SGGMs (e.g., the topological properties of the graph structure) which significantly affect the convergence bound. Additionally, we offer theoretical insights into the selection of hyperparameters (e.g., sampling steps and diffusion length) and advocate for techniques like normalization to improve convergence. To validate our theoretical findings, we conduct a controlled empirical study using a synthetic graph model. The results in this paper contribute to a deeper theoretical understanding of SGGMs and offer practical guidance for designing more efficient and effective SGGMs.

The application of deep learning for partial differential equation (PDE)-constrained control is gaining increasing attention. However, existing methods rarely consider safety requirements crucial in real-world applications. To address this limitation, we propose Safe Diffusion Models for PDE Control (SafeDiffCon), which introduce the uncertainty quantile as model uncertainty quantification to achieve optimal control under safety constraints through both post-training and inference phases. Firstly, our approach post-trains a pre-trained diffusion model to generate control sequences that better satisfy safety constraints while achieving improved control objectives via a reweighted diffusion loss, which incorporates the uncertainty quantile estimated using conformal prediction. Secondly, during inference, the diffusion model dynamically adjusts both its generation process and parameters through iterative guidance and fine-tuning, conditioned on control targets while simultaneously integrating the estimated uncertainty quantile. We evaluate SafeDiffCon on three control tasks: 1D Burgers' equation, 2D incompressible fluid, and controlled nuclear fusion problem. Results demonstrate that SafeDiffCon is the only method that satisfies all safety constraints, whereas other classical and deep learning baselines fail. Furthermore, while adhering to safety constraints, SafeDiffCon achieves the best control performance. The code can be found at https://github.com/AI4Science-WestlakeU/safediffcon.

Offline preference alignment for language models such as Direct Preference Optimization (DPO) is favored for its effectiveness and simplicity, eliminating the need for costly reinforcement learning. Various offline algorithms have been developed for different data settings, yet they lack a unified understanding. In this study, we introduce Pior-Informed Preference Alignment (PIPA), a unified, RL-free probabilistic framework that formulates language model preference alignment as a Maximum Likelihood Estimation (MLE) problem with prior constraints. This method effectively accommodates both paired and unpaired data, as well as answer and step-level annotations. We illustrate that DPO and KTO are special cases with different prior constraints within our framework. By integrating different types of prior information, we developed two variations of PIPA: PIPA-M and PIPA-N. Both algorithms demonstrate a $3\sim10\%$ performance enhancement on the GSM8K and MATH benchmarks across all configurations, achieving these gains without additional training or computational costs compared to existing algorithms.

In this paper, we theoretically investigate the effects of noisy labels in offline alignment, with a focus on the interplay between privacy and robustness against adversarial corruption. Specifically, under linear modeling assumptions, we present a unified analysis covering both reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO) under different privacy-corruption scenarios, such as Local differential privacy-then-Corruption (LTC), where human preference labels are privatized before being corrupted by an adversary, and Corruption-then-Local differential privacy (CTL), where labels are corrupted before privacy protection. Our analysis leverages a reduction framework that reduces the offline alignment problem under linear modeling assumptions to parameter estimation in logistic regression. This framework allows us to establish an interesting separation result between LTC and CTL, demonstrating that LTC presents a greater challenge than CTL in offline alignment, even under linear models. As important by-products, our findings also advance the state-of-the-art theoretical results in offline alignment under privacy-only or corruption-only scenarios.

*Policy gradient* (PG) methods are effective *reinforcement learning* (RL) approaches, particularly for continuous problems. While they optimize stochastic (hyper)policies via action- or parameter-space exploration, real-world applications often require deterministic policies. Existing PG convergence guarantees to deterministic policies assume a fixed stochasticity in the (hyper)policy, tuned according to the desired final suboptimality, whereas practitioners commonly use a dynamic stochasticity level. This work provides the theoretical foundations for this practice. We introduce PES, a phase-based method that reduces stochasticity via a deterministic schedule while running PG subroutines with fixed stochasticity in each phase. Under gradient domination assumptions, PES achieves last-iterate convergence to the optimal deterministic policy with a sample complexity of order $\widetilde{\mathcal{O}}(\epsilon^{-5})$. Additionally, we analyze the common practice, termed SL-PG, of jointly learning stochasticity (via an appropriate parameterization) and (hyper)policy parameters. We show that SL-PG also ensures last-iterate convergence with a rate $\widetilde{\mathcal{O}}(\epsilon^{-3})$, but to the optimal stochastic (hyper)policy only, requiring stronger assumptions compared to PES.

Federated prompt learning (FPL) for vision-language models is a powerful approach to collaboratively adapt models across distributed clients while preserving data privacy. However, existing FPL approaches suffer from a trade-off between performance and robustness, particularly in out-of-distribution (OOD) shifts, limiting their reliability in real-world scenarios. The inherent in-distribution (ID) data heterogeneity among different clients makes it more challenging to maintain this trade-off. To fill this gap, we introduce a Federated OOD-aware Context Optimization (FOCoOp) framework, which captures diverse distributions among clients using ID global prompts, local prompts, and OOD prompts. Specifically, FOCoOp leverages three sets of prompts to create both class-level and distribution-level separations, which adapt to OOD shifts through bi-level distributionally robust optimization. Additionally, FOCoOp improves the discrimination consistency among clients, i.e., calibrating global prompts, seemly OOD prompts, and OOD prompts by Semi-unbalanced optimal transport. The extensive experiments on real-world datasets demonstrate that FOCoOp effectively captures decentralized heterogeneous distributions and enhances robustness of different OOD shifts. The project is available at GitHub.