NeurIPS 2025 Papers

Event-based cameras offer unique advantages such as high temporal resolution, high dynamic range, and low power consumption. However, the massive storage requirements and I/O burdens of existing synthetic data generation pipelines and the scarcity of real data prevent event-based training datasets from scaling up, limiting the development and generalization capabilities of event vision models. To address this challenge, we introduce Video-to-Voxel (V2V), an approach that directly converts conventional video frames into event-based voxel grid representations, bypassing the storage-intensive event stream generation entirely. V2V enables a 150× reduction in storage requirements while supporting on-the-fly parameter randomization for enhanced model robustness. Leveraging this efficiency, we train several video reconstruction and optical flow estimation model architectures on 10,000 diverse videos totaling 52 hours—an order of magnitude larger than existing event datasets, yielding substantial improvements.

Contrastive Language-Image Pretraining (CLIP) has demonstrated strong zero-shot performance across diverse downstream text-image tasks. Existing CLIP methods typically optimize a contrastive objective using negative samples drawn from each minibatch. To achieve robust representation learning, these methods require extremely large batch sizes and escalate computational demands to hundreds or even thousands of GPUs. Prior approaches to mitigate this issue often compromise downstream performance, prolong training duration, or face scalability challenges with very large datasets. To overcome these limitations, we propose AmorLIP, an efficient CLIP pretraining framework that amortizes expensive computations involved in contrastive learning through lightweight neural networks, which substantially improves training efficiency and performance. Leveraging insights from a spectral factorization of energy-based models, we introduce novel amortization objectives along with practical techniques to improve training stability. Extensive experiments across 38 downstream tasks demonstrate the superior zero-shot classification and retrieval capabilities of AmorLIP, consistently outperforming standard CLIP baselines with substantial relative improvements of up to 12.24%.

Machine learning models now influence decisions that directly affect people’s lives, making it important to understand not only their predictions, but also how individuals could act to obtain better results. Algorithmic recourse provides actionable input modifications to achieve more favorable outcomes, typically relying on counterfactual explanations to suggest such changes. However, when the Rashomon set - the set of near-optimal models - is large, standard counterfactual explanations can become unreliable, as a recourse action valid for one model may fail under another. We introduce ElliCE, a novel framework for robust algorithmic recourse that optimizes counterfactuals over an ellipsoidal approximation of the Rashomon set. The resulting explanations are provably valid over this ellipsoid, with theoretical guarantees on uniqueness, stability, and alignment with key feature directions. Empirically, ElliCE generates counterfactuals that are not only more robust but also more flexible, adapting to user-specified features constraints while being substantially faster than existing baselines. This provides a principled and practical solution for reliable recourse under model uncertainty, ensuring stable recommendations for users even as models evolve.

We examine the problem of efficiently learning coarse correlated equilibria (CCE) in polyhedral games, that is, normal-form games with an exponentially large number of actions per player and an underlying combinatorial structure—such as the classic Colonel Blotto game or congestion games. Achieving computational efficiency in this setting requires learning algorithms whose regret and per-iteration complexity scale at most polylogarithmically with the size of the players’ action sets. This challenge has recently been addressed in the full-information setting, primarily through the use of kernelization; however, in the more realistic partial information setting, the situation is much more challenging, and existing approaches result in suboptimal and impractical runtime complexity to learn CCE. We address this gap via a novel kernelization-based framework for payoff-based learning in polyhedral games, which we then apply to certain key classes of polyhedral games—namely Colonel Blotto, graphic matroid and network congestion games. In so doing, we obtain a range of computationally efficient payoff-based learning algorithms which significantly improve upon prior work in terms of the runtime for learning CCE.

Heterogeneous temporal graphs (HTGs) are ubiquitous data structures in the real world. Recently, to enhance representation learning on HTGs, numerous attention-based neural networks have been proposed. Despite these successes, existing methods rely on a decoupled temporal and spatial learning paradigm, which weakens interactions of spatio-temporal information and leads to a high model complexity. To bridge this gap, we propose a novel learning paradigm for HTGs called Simple and Efficient Heterogeneous Temporal Graph Neural Network (SE-HTGNN). Specifically, we innovatively integrate temporal modeling into spatial learning via a novel dynamic attention mechanism, which substantially reduces model complexity while enhancing discriminative representation learning on HTGs. Additionally, to comprehensively and adaptively understand HTGs, we leverage large language models to prompt SE-HTGNN, enabling the model to capture the implicit properties of node types as prior knowledge. Extensive experiments demonstrate that SE-HTGNN achieves up to 10× speed-up over the state-of-the-art and latest baseline while maintaining the best forecasting accuracy.

Algorithms for learning decision trees often include heuristic local-search operations such as (1) adjusting the threshold of a cut or (2) also exchanging the feature of that cut. We study minimizing the number of classification errors by performing a fixed number of a single type of these operations. Although we discover that the corresponding problems are NP-complete in general, we provide a comprehensive parameterized-complexity analysis with the aim of determining those properties of the problems that explain the hardness and those that make the problems tractable. For instance, we show that the problems remain hard for a small number $d$ of features or small domain size $D$ but the combination of both yields fixed-parameter tractability. That is, the problems are solvable in $(D + 1)^{2d} \cdot |\mathcal{I}|^{O(1)}$ time, where $|\mathcal{I}|$ is the size of the input. We also provide a proof-of-concept implementation of this algorithm and report on empirical results.

Large Language Models (LLMs) have achieved remarkable success across many applications, with Mixture of Experts (MoE) models demonstrating great potential. Compared to traditional dense models, MoEs achieve better performance with less computation. Speculative decoding (SD) is a widely used technique to accelerate LLM inference without accuracy loss, but it has been considered efficient only for dense models. In this work, we first demonstrate that, under medium batch sizes, MoE surprisingly benefits more from SD than dense models. Furthermore, as MoE becomes sparser -- the prevailing trend in MoE designs -- the batch size range where SD acceleration is expected to be effective becomes broader. To quantitatively understand tradeoffs involved in SD, we develop a reliable modeling based on theoretical analyses. While current SD research primarily focuses on improving acceptance rates of algorithms, changes in workload and model architecture can still lead to degraded SD acceleration even with high acceptance rates. To address this limitation, we introduce a new metric 'target efficiency' that characterizes these effects, thus helping researchers identify system bottlenecks and understand SD acceleration more comprehensively. For scenarios like private serving, this work unveils a new perspective to speed up MoE inference, where existing solutions struggle. Experiments on different GPUs show up to 2.29x speedup for Qwen2-57B-A14B at medium batch sizes and validate our theoretical predictions.

Dataset distillation aims to synthesize a small dataset from a large dataset, enabling the model trained on it to perform well on the original dataset. With the blooming of large language models and multimodal large language models, the importance of multimodal datasets, particularly image-text datasets, has grown significantly. However, existing multimodal dataset distillation methods are constrained by the Matching Training Trajectories algorithm, which significantly increases the computing resource requirement, and takes days to process the distillation. In this work, we introduce EDGE, a generative distillation method for efficient multimodal dataset distillation. Specifically, we identify two key challenges of distilling multimodal datasets with generative models: 1) The lack of correlation between generated images and captions. 2) The lack of diversity among generated samples. To address the aforementioned issues, we propose a novel generative model training workflow with a bi-directional contrastive loss and a diversity loss. Furthermore, we propose a caption synthesis strategy to further improve text-to-image retrieval performance by introducing more text information. Our method is evaluated on Flickr30K, COCO, and CC3M datasets, demonstrating superior performance and efficiency compared to existing approaches. Notably, our method achieves results 18$\times$ faster than the state-of-the-art method. Our code will be made public at https://github.com/ichbill/EDGE.

Training language models on long sequence data is a demanding requirement for enhancing the model's capability on complex tasks, e.g., long-chain reasoning. However, as the sequence length scales up, the memory cost for storing activation values becomes huge during the Backpropagation (BP) process, even with the application of gradient checkpointing technique. To tackle this challenge, we propose a *memory-efficient* and *exact* BP method called **StreamBP**, which performs a linear decomposition of the chain rule along the sequence dimension in a layer-wise manner, significantly reducing the memory cost of activation values and logits. The proposed method is applicable to common objectives such as SFT, GRPO, and DPO. From an implementation perspective, StreamBP achieves less computational FLOPs and faster BP speed by leveraging the causal structure of the language model. Compared to gradient checkpointing, StreamBP scales up the maximum sequence length of BP by $2.8-5.5 \times$ larger, while using comparable or even less BP time. Note that StreamBP's sequence length scaling ability can be directly transferred to batch size scaling for accelerating training. We further develop a communication-efficient distributed StreamBP to effectively support multi-GPU training and broaden its applicability. Our code can be easily integrated into the training pipeline of any transformer models and is available at https://github.com/Ledzy/StreamBP.

We present LrcSSM, a $\textit{non-linear}$ recurrent model that processes long sequences as fast as today's linear state-space layers. By forcing its Jacobian matrix to be diagonal, the full sequence can be solved in parallel, giving $\mathcal{O}(TD)$ time and memory and only $\mathcal{O}(\log T)$ sequential depth, for input-sequence length $T$ and a state dimension $D$. Moreover, LrcSSM offers a formal gradient-stability guarantee that other input-varying systems such as Liquid-S4 and Mamba do not provide. Importantly, the diagonal Jacobian structure of our model results in no performance loss compared to the original model with dense Jacobian, and the approach can be generalized to other non-linear recurrent models, demonstrating broader applicability. On a suite of long-range forecasting tasks, we demonstrate that LrcSSM outperforms Transformers, LRU, S5, and Mamba.

Effective data selection is critical for efficient training of modern Large Language Models (LLMs). This paper introduces Influence Distillation, a novel, mathematically-justified framework for data selection that employs second-order information to optimally weight training samples. By distilling each sample's influence on a target distribution, our method assigns model-specific weights that are used to select training data for LLM fine-tuning, guiding it toward strong performance on the target domain. We derive these optimal weights for both Gradient Descent and Adam optimizers. To ensure scalability and reduce computational cost, we propose a $\textit{landmark-based approximation}$: influence is precisely computed for a small subset of "landmark" samples and then efficiently propagated to all other samples to determine their weights. We validate Influence Distillation by applying it to instruction tuning on the Tulu V2 dataset, targeting a range of tasks including GSM8k, SQuAD, and MMLU, across several models from the Llama and Qwen families. Experiments show that Influence Distillation matches or outperforms state-of-the-art performance while achieving up to $3.5\times$ faster selection.

Vision Transformers have demonstrated exceptional performance across various computer vision tasks, yet their quadratic computational complexity concerning token length remains a significant challenge. To address this, token reduction methods have been widely explored. However, existing approaches often overlook the frequency characteristics of self-attention, such as rank collapsing and over-smoothing phenomenon. In this paper, we propose a frequency-aware token reduction strategy that improves computational efficiency while preserving performance by mitigating rank collapsing. Our method partitions tokens into high-frequency tokens and low-frequency tokens. high-frequency tokens are selectively preserved, while low-frequency tokens are aggregated into a compact direct current token to retain essential low-frequency components. Through extensive experiments and analysis, we demonstrate that our approach significantly improves accuracy while reducing computational overhead and mitigating rank collapsing and over smoothing. Furthermore, we analyze the previous methods, shedding light on their implicit frequency characteristics and limitations. The code is available in https://github.com/jhtwosun/frequency-aware-token-pruning.

Fine-tuning large language models (LLMs) poses significant computational burdens, especially in federated learning (FL) settings. We introduce Layer-wise Efficient Federated Fine-tuning (LEFF), a novel method designed to enhance the efficiency of FL fine-tuning while preserving model performance and minimizing client-side computational overhead. LEFF strategically selects layers for fine-tuning based on client computational capacity, thereby mitigating the straggler effect prevalent in heterogeneous environments. Furthermore, LEFF incorporates an importance-driven layer sampling mechanism, prioritizing layers with greater influence on model performance. Theoretical analysis demonstrates that LEFF achieves a convergence rate of $\mathcal{O}(1/\sqrt{T})$. Extensive experiments on diverse datasets demonstrate that LEFF attains superior computational efficiency and model performance compared to existing federated fine-tuning methods, particularly under heterogeneous conditions.

Modern language models (LMs) exhibit strong deductive reasoning capabilities, yet standard evaluations emphasize correctness while overlooking a key aspect of human-like reasoning: *efficiency*. In real-world reasoning scenarios, much of the available information is irrelevant, and effective deductive inference requires identifying and ignoring such distractions. We propose a framework for assessing LM reasoning efficiency through the lens of logic programming, introducing a simple method to align proofs written in natural language---as generated by an LM---with shortest proofs found by executing the logic program. Efficiency is quantified by measuring how well a model avoids unnecessary inference. Empirically, we construct a dataset of math word problems injected with various number of irrelevant axioms that vary in semantic overlap with the goal theorem. We find that current LMs show marked accuracy declines under such conditions---even with minimal, domain-consistent distractions---and the proofs they generate frequently exhibit detours through irrelevant inferences.

Low-rank gradient-based optimization methods have significantly improved memory efficiency during the training of large language models (LLMs), enabling operations within constrained hardware without sacrificing performance. However, these methods primarily emphasize memory savings, often overlooking potential acceleration in convergence due to their reliance on standard isotropic steepest descent techniques, which can perform suboptimally in the highly anisotropic landscapes typical of deep networks, particularly LLMs. In this paper, we propose SUMO (Subspace-Aware Moment-Orthogonalization), an optimizer that employs exact singular value decomposition (SVD) for moment orthogonalization within a dynamically adapted low-dimensional subspace, enabling norm-inducing steepest descent optimization steps. By explicitly aligning optimization steps with the spectral characteristics of the loss landscape, SUMO effectively mitigates approximation errors associated with commonly used methods like Newton-Schulz orthogonalization approximation. We theoretically establish an upper bound on these approximation errors, proving their dependence on the condition numbers of moments, conditions we analytically demonstrate are encountered during LLM training. Furthermore, we both theoretically and empirically illustrate that exact orthogonalization via SVD substantially improves convergence rates while reducing overall complexity. Empirical evaluations confirm that SUMO accelerates convergence, enhances stability, improves performance, and reduces memory requirements by up to 20\% compared to state-of-the-art methods.

Continual learning seeks to enable a model to assimilate knowledge from non-stationary data streams without catastrophic forgetting. Recently, methods based on Parameter-Efficient Tuning (PET) have achieved superior performance without even storing any historical exemplars, which train much fewer specific parameters for each task upon a frozen pre-trained model, and tailored parameters are retrieved to guide predictions during inference. However, reliance solely on pre-trained features for parameter matching exacerbates the inconsistency between the training and inference phases, thereby constraining the overall performance. To address this issue, we propose HRM-PET, which makes full use of the richer downstream knowledge inherently contained in the trained parameters. Specifically, we introduce a hybrid re-matching mechanism, which benefits from the initial predicted distribution to facilitate the parameter selections. The direct re-matching addresses misclassified samples identified with correct task identity in prediction, despite incorrect initial matching. Moreover, the confidence-based re-matching is specifically designed to handle other more challenging mismatched samples that cannot be calibrated by the former. Besides, to acquire task-invariant knowledge for better matching, we integrate a cross-task instance relationship distillation module into the PET-based method. Extensive experiments conducted on four datasets under five pre-trained settings demonstrate that HRM-PET performs favorably against the state-of-the-art methods. The code is available in the https://github.com/wei-cheng777/HRM-PET.

Visual navigation is essential for robotics and embodied AI. However, existing foundation models, particularly those with transformer decoders, suffer from high computational overhead and lack interpretability, limiting their deployment on edge devices. To address this, we propose DynaNav, a Dynamic Visual Navigation framework that adapts feature and layer selection based on scene complexity. It employs a trainable hard feature selector for sparse operations, enhancing efficiency and interpretability. Additionally, we integrate feature selection into an early-exit mechanism, with Bayesian Optimization determining optimal exit thresholds to reduce computational cost. Extensive experiments in real-world-based datasets and simulated environments demonstrate the effectiveness of DynaNav. Compared to ViNT, DynaNav achieves a $2.6\times$ reduction in FLOPs, 42.3% lower inference time, and 32.8% lower memory usage while improving navigation performance across four public datasets.

Dynamic coreset selection is a promising approach for improving the training efficiency of deep neural networks by periodically selecting a small subset of the most representative or informative samples, thereby avoiding the need to train on the entire dataset. However, it remains inherently challenging due not only to the complex interdependencies among samples and the evolving nature of model training, but also to a critical *coreset representativeness degradation issue* identified and explored in-depth in this paper, that is, the representativeness or information content of the coreset degrades over time as training progresses. Therefore, we argue that, in addition to designing accurate selection rules, it is equally important to endow the algorithms with the ability to assess the quality of the current coreset. Such awareness enables timely re-selection, mitigating the risk of overfitting to stale subsets—a limitation often overlooked by existing methods. To this end, this paper proposes an **E**fficient **R**epresentativeness-**A**ware **C**oreset **S**election method for deep neural networks, a lightweight framework that enables dynamic tracking and maintenance of coreset quality during training. While the ideal criterion—gradient discrepancy between the coreset and the full dataset—is computationally prohibitive, we introduce a scalable surrogate based on the signal-to-noise ratio (SNR) of gradients within the coreset, which is the main technical contribution of this paper and is also supported by our theoretical analysis. Intuitively, a decline in SNR indicates overfitting to the subset and declining representativeness. Leveraging this observation, our method triggers coreset updates without requiring costly Hessian or full-batch gradient computations, maintaining minimal computational overhead. Experiments on multiple datasets confirm the effectiveness of our approach. Notably, compared with existing gradient-based dynamic coreset selection baselines, our method achieves up to a 5.4\% improvement in test accuracy across multiple datasets.

Brain-inspired spiking neural networks (SNNs) promise to be a low-power alternative to computationally intensive artificial neural networks (ANNs), although performance gaps persist. Recent studies have improved the performance of SNNs through knowledge distillation, but rely on large teacher models or introduce additional training overhead. In this paper, we show that SNNs can be naturally deconstructed into multiple submodels for efficient self-distillation. We treat each timestep instance of the SNN as a submodel and evaluate its output confidence, thus efficiently identifying the strong and the weak. Based on this strong and weak relationship, we propose two efficient self-distillation schemes: (1) Strong2Weak: During training, the stronger "teacher" guides the weaker "student", effectively improving overall performance. (2) Weak2Strong: The weak serve as the "teacher", distilling the strong in reverse with underlying dark knowledge, again yielding significant performance gains. For both distillation schemes, we offer flexible implementations such as ensemble, simultaneous, and cascade distillation. Experiments show that our method effectively improves the discriminability and overall performance of the SNN, while its adversarial robustness is also enhanced, benefiting from the stability brought by self-distillation. This ingeniously exploits the temporal properties of SNNs and provides insight into how to efficiently train high-performance SNNs.