论文检索

Reinforcement Learning · Online

Hanyang Wang, Lu Wang, Chaoyun Zhang, Tianjun Mao, Si Qin, Qingwei Lin, Saravan Rajmohan, Dongmei Zhang

Traditional RLHF optimizes language models with coarse, scalar rewards that mask the fine-grained reasons behind success or failure, leading to slow, opaque learning. Recent work augments RL with textual critiques through prompting or reflection, improving interpretability but leaving model parameters untouched. We introduce **Text2Grad**, a reinforcement-learning paradigm that *turns free-form textual feedback into span-level gradients*. Given human (or programmatic) critiques, Text2Grad aligns each feedback phrase with the relevant token spans, converts these alignments into differentiable reward signals, and performs gradient updates that directly refine the offending portions of the model's policy. This yields precise, feedback-conditioned adjustments instead of global nudges. Text2Grad is realized through three components: (1) a high-quality feedback–annotation pipeline that pairs critiques with token spans; (2) a fine-grained reward model that predicts span-level reward on answers while generating explanatory critiques; and (3) a span-level policy optimizer that back-propagates *natural-language gradients*. Across summarization, code generation, and question answering, Text2Grad consistently surpasses scalar-reward RL and prompt-only baselines, providing both higher task metrics and richer interpretability. Our results suggest that natural-language feedback can serve not only as explanations, but also as actionable training signals for fine-grained alignment. The code for our method is available at *[https://github.com/microsoft/Text2Grad](https://github.com/microsoft/Text2Grad)*.

Computer Vision · Image and Video Generation

Zhiwen Zheng, Yuheng Qiao, Xiaoshuai Zhang, Zhao Huang, Tao Zhang, Huiyu Zhou, Shaowei Jiang, Jin Liu, Wenwen Tang, Xingru Huang

Noncontact exfiltration of electronic screen content poses a security challenge, with side-channel incursions as the principal vector. We introduce an optical projection side-channel paradigm that confronts two core instabilities: (i) the near-singular Jacobian spectrum of projection mapping breaches Hadamard stability, rendering inversion hypersensitive to perturbations; (ii) irreversible compression in light transport obliterates global semantic cues, magnifying reconstruction ambiguity. Exploiting passive speckle patterns formed by diffuse reflection, our Irradiance Robust Radiometric Inversion Network (IR$^4$Net) fuses a Physically Regularized Irradiance Approximation (PRIrr‑Approximation), which embeds the radiative transfer equation in a learnable optimizer, with a contour-to-detail cross-scale reconstruction mechanism that arrests noise propagation. Moreover, an Irreversibility Constrained Semantic Reprojection (ICSR) module reinstates lost global structure through context-driven semantic mapping. Evaluated across four scene categories, IR$^4$Net achieves fidelity beyond competing neural approaches while retaining resilience to illumination perturbations.

Computer Vision · Vision Models & Multimodal

Enxin Song, Wenhao Chai, Shusheng Yang, Ethan Armand, Xiaojun Shan, Haiyang Xu, Jianwen Xie, Zhuowen Tu

Video understanding in multimodal language models remains limited by context length: models often miss key transition frames and struggle to maintain coherence across long time scales. To address this, we adapt Native Sparse Attention (NSA) to video-language models. **Our method, VideoNSA, adapts Qwen2.5-VL through end-to-end training on a 216K video instruction dataset. We employ a hardware-aware hybrid approach to attention, preserving dense attention for text, while employing NSA for video.** Compared to token-compression and training-free sparse baselines, VideoNSA achieves improved performance on long-video understanding, temporal reasoning, and spatial benchmarks. Further ablation analysis reveals four key findings: (1) reliable scaling to 128K tokens; (2) an optimal global–local attention allocation at a fixed budget; (3) task-dependent branch usage patterns; and (4) the learnable combined sparse attention help induce dynamic attention sinks.

Optimization · Learning for Optimization

Zhuoli Yin, Yi Ding, Reem Khir, Hua Cai

Solving the Traveling Salesman Problem (TSP) is NP-hard yet fundamental for a wide range of real-world applications. Classical exact methods face challenges in scaling, and heuristic methods often require domain-specific parameter calibration. While learning-based approaches have shown promise, they suffer from poor generalization and limited scalability due to fixed training data. This work proposes ViTSP, a novel framework that leverages pre-trained vision language models (VLMs) to visually guide the solution process for large-scale TSPs. The VLMs function to identify promising small-scale subproblems from a visualized TSP instance, which are then efficiently optimized using an off-the-shelf solver to improve the global solution. ViTSP bypasses the dedicated model training at the user end while maintaining effectiveness across diverse instances. Experiments on real-world TSP instances ranging from 1k to 88k nodes demonstrate that ViTSP consistently achieves solutions with average optimality gaps of 0.24\%, outperforming existing learning-based methods. Under the same runtime budget, it surpasses the best-performing heuristic solver, LKH-3, by reducing its gaps by 3.57\% to 100\%, particularly on very-large-scale instances with more than 10k nodes. Our framework offers a new perspective in hybridizing pre-trained generative models and operations research solvers in solving combinatorial optimization problems. The framework holds potential for integration into more complex real-world logistics systems. The code is available at https://github.itap.purdue.edu/uSMART/ViTSP_ICLR2026.

Deep Learning · Algorithms

Minghui Chen, Wenlong Deng, James Y Zou, Han Yu, Xiaoxiao Li

Large language models (LLMs) are increasingly deployed as part of compound AI systems which coordinate multiple modules (e.g., retrievers, tools, verifiers) over long-horizon workflows. Although recent frameworks that propagate textual feedback globally (e.g., TextGrad make it feasible to optimize such pipelines, we identify two depth-scaling failure modes in long-horizon agentic workflows: 1) exploding textual gradient, where textual feedback grows exponentially with depth, leading to prohibitively long message and amplifies evaluation biases; and 2) vanishing textual gradient, where limited long-context ability causes models overemphasize recent or early feedback, while compression of lengthy feedback causes downstream messages to lose specificity gradually as they propagate many hops upstream. To mitigate these issues, we introduce Textual Equilibrium Propagation (TEP), a local learning principle inspired by Equilibrium Propagation in energy-based models. TEP includes two phases: 1) a free phase where a local LLM critics iteratively refine prompts until reaching equilibrium (no further improvements are suggested); and 2) a nudged phase which applies proximal prompt edits with bounded modification intensity, using task-level objectives that propagate via forward signaling rather than backward feedback chains. This design supports local prompt optimization followed by controlled adaptation toward global goals without the computational burden and signal degradation of global textual backpropagation. Across long-horizon QA benchmarks and multi-agent tool-use dataset, TEP consistently improves accuracy and efficiency over global propagation methods such as TextGrad, with gains that increase at greater depths, while preserving the practicality of black-box LLM components in deep compound AI system.

Applications · Language, Speech and Dialog

Zican Hu, Shilin Zhang, Yafu Li, Jianhao (Elliott) Yan, Xuyang Hu, Leyang Cui, Xiaoye Qu, Chunlin Chen, Yu Cheng, Zhi Wang

Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a crucial paradigm for incentivizing reasoning capabilities in Large Language Models (LLMs). Due to vast state-action spaces and reward sparsity in reasoning tasks, existing methods often struggle with deficient exploration and poor sample efficiency. In the paper, we propose **DIVER** (**D**iversity-**I**ncentivized Exploration for **V**ersatil**E** **R**easoning), an innovative framework that highlights the pivotal role of global sequence-level diversity to incentivize deep exploration for versatile reasoning. We first conduct a primary empirical study to reveal a strong positive correlation between global diversity and reasoning capacity. Building on this insight, we introduce global diversity incentives as an intrinsic reward to promote deep exploration in a semantically structured space. Incorporating the intrinsic reward, we develop a potential-based reward shaping mechanism to preserve optimal policy invariance and design simple heuristics to mitigate possible reward hacking. Experimental results show that DIVER outperforms competitive RLVR baselines with various exploration strategies on both in-domain and out-of-domain tasks, excelling in both Pass@1 and Pass@k evaluations. Our code is available at https://github.com/NJU-RL/DIVER.

General Machine Learning · Transfer, Multitask and Meta-learning

Zhiliang Chen, Gregory Kang Ruey Lau, Chuan Sheng Foo, Bryan Kian Hsiang Low

The performance of an LLM depends heavily on how well the training data matches the downstream evaluation task. However, in many practical settings, we typically do not know the data in the evaluation task (e.g., conversations between a chatbot and users are end-to-end encrypted). We refer to such tasks as unseen evaluation tasks. We can only deploy the LLM on these unseen evaluation tasks to gather multiple rounds of feedback on how well the model performs (e.g., gathering user ratings from a chatbot). In addition, this feedback can be noisy. How can we exploit such noisy feedback efficiently to optimize the LLM training data-mixture? Our paper presents DUET, a novel global-to-local algorithm that optimizes training data mixtures by interleaving data selection with Bayesian optimization to exploit coarse and noisy feedback from a downstream evaluation task. DUET is flexible enough to incorporate different data selection methods, each with different performance-compute tradeoffs. By analyzing DUET's cumulative regret, we theoretically show that DUET converges to the optimal training data mixture even without any fine-grained data information from an unseen task. Finally, our experiments across a variety of language tasks demonstrate that DUET attains substantial performance improvements over existing data selection and mixing methods in the unseen-task setting. Our library, which is flexible enough to optimize different LLM training ingredients, can be found at https://github.com/chenzhiliang94/BO-for-LLM.

General Machine Learning · Causality

Xinshuai Dong, Ignavier Ng, Haoyue Dai, Jiaqi Sun, Xiangchen Song, Peter Spirtes, Kun Zhang

Identifying the structure of a partially observed causal system is essential to various scientific fields. Recent advances have focused on constraint-based causal discovery to solve this problem, and yet in practice these methods often face challenges related to multiple testing and error propagation. These issues could be mitigated by a score-based method and thus it has raised great attention whether there exists a score-based greedy search method that can handle the partially observed scenario. In this work, we propose the first score-based greedy search method for the identification of structure involving latent variables with identifiability guarantees. Specifically, we propose Generalized N Factor Model and establish the global consistency: the true structure including latent variables can be identified up to the Markov equivalence class by using score. We then design Latent variable Greedy Equivalence Search (LGES), a greedy search algorithm for this class of model with well-defined operators, which search very efficiently over the graph space to find the optimal structure. Our experiments on both synthetic and real-life data validate the effectiveness of our method.

General Machine Learning · Representation Learning

Hangke Sui, Yuqing Wang, Minh Do

Contrastive objectives power state-of-the-art multimodal models, but their training remains slow, relying on long stochastic optimization. We propose a Unified Framework for Efficient Contrastive Alignment via Kernels (UniCon), which spans linear and nonlinear encoders as well as one-to-one and many-to-many alignments. At its core, UniCon introduces the contrastive similarity weight matrix $S(\gamma)$, which enables closed-form global solutions that provably replace minibatch back-propagation with exact updates. Through the lens of reproducing kernel Hilbert spaces (RKHS), UniCon provides a kernelized perspective that unifies contrastive alignment and reveals its connection to spectral methods. To validate the theory, we conduct experiments on synthetic, unimodal, multimodal, and zero-shot tasks, demonstrating that UniCon achieves substantial efficiency gains while preserving generality and strong empirical performance.

Deep Learning · Algorithms

Wenbing Li, Zikai Song, Hang Zhou, Junqing Yu, Yunyao Zhang, Wei Yang

Recent attempts to combine low-rank adaptation (LoRA) with mixture-of-experts (MoE) for multi-task adaptation of Large Language Models (LLMs) often replace whole attention/FFN layers with switch experts or append parallel expert branches, undermining parameter efficiency and limiting task specialization. We introduce **LoRA-Mixer**, a modular MoE framework that routes task-specific LoRA experts into the core projection matrices of the attention module (input/output linear layers), rather than primarily targeting FFN blocks. The design delivers fine-grained token-level specialization by fully exploiting the attention mechanism, while remaining drop-in compatible with Transformers and state-space models (SSMs) as the linear projection layers are ubiquitous. To train robust routers from limited data while promoting stable, selective decisions and high expert reuse, **LoRA-Mixer** employs an adaptive **Routing Specialization Loss (RSL)** that jointly enforces global load balance and input-aware specialization via an entropy-shaping objective. The framework supports two regimes: (i) joint optimization of adapters and router with a differentiable hard–soft top-k routing scheme, and (ii) plug-and-play routing over frozen, pre-trained LoRA modules sourced from public repositories. Across 15 benchmarks—including MedQA, GSM8K, HumanEval, and GLUE—RSL-optimized LoRA-Mixer outperforms state-of-the-art routing and LoRA-MoE baselines while using 48% of their trainable parameters, with gains of +3.79%, +2.90%, and +3.95% on GSM8K, CoLA, and ARC-C, respectively. Cross-model transfer and adapter reuse experiments further demonstrate the approach’s versatility and data efficiency.

Optimization · Global Optimization

Youjun Wang, Boqi Li, Xin Zou, Weiwei Liu

Neural collapse (NC) plays a key role in understanding deep neural networks. However, existing empirical and theoretical studies of NC focus on one single task. This paper studies neural collapse in multi-task learning. We consider two standard feature-based multi-task learning scenarios: Single-Source Multi-Task Classification (SSMTC) and Multi-Source Multi-Task Classification (MSMTC). Interestingly, we find that the task-specific linear classifier and features converge to the Simplex Equiangular Tight Frame (ETF) in the setting of MSMTC. In the setting of SSMTC, task-specific linear classifier converges to the task-specific ETF and these task-specific ETFs are mutually orthogonal. Moreover, the shared features across tasks converge to the scaled sum of the weight vectors associated with the task-specific labels in each task's classifier. We also provide the theoretical guarantee for our empirical findings. Through detailed analysis, we uncover the mechanism of MTL where each task learns task-specific latent features that together form the shared features. Moreover, we reveal an inductive bias in MTL that task correlation reconfigures the geometry of task-specific classifiers and promotes alignment among the features learned by each task.

Applications · Robotics

Haisheng Su, Wei Wu, Feixiang Song, Junjie Zhang, Zhenjie Yang, Junchi Yan

Recent advances towards End-to-End Autonomous Driving (E2E-AD) focus on integrating modular designs into a unified framework for joint optimization. Most of these advances follow a sequential paradigm (i.e., perception-prediction-planning) based on separable Transformer decoders and rely on dense BEV features to encode scene representations. However, such manual ordering design can inevitably cause information loss and cumulative errors, lacking flexible and diverse relation modeling among different modules and sensors. Meanwhile, insufficient training of image backbone and quadratic-complexity of attention mechanism also hinder the scalability and efficiency of E2E-AD system to handle spatiotemporal input. To this end, we propose DriveMamba, a Task-Centric Scalable paradigm for efficient E2E-AD, which integrates dynamic task relation modeling, implicit view correspondence learning and long-term temporal fusion into a single-stage Unified Mamba decoder. Specifically, both extracted image features and expected task outputs are converted into token-level sparse representations in advance, which are then sorted by their instantiated positions in 3D space. The linear-complexity operator enables efficient long-context sequential token modeling to capture task-related inter-dependencies simultaneously. Additionally, a bidirectional trajectory-guided "local-to-global" scan method is designed to preserve spatial locality from ego-perspective, thus facilitating the ego-planning. Extensive experiments conducted on nuScenes and Bench2Drive datasets demonstrate the superiority, generalizability and great efficiency of DriveMamba.

Jiangyong Yu, Xiaomeng Han, Xing Hu, XUCHEN, Zhe jiang, Dawei Yang

Large Language Models (LLMs) have demonstrated remarkable performance across a wide range of tasks, but their deployment is often constrained by substantial memory footprints and computational costs. While prior work has achieved significant progress in compressing and accelerating linear layers, nonlinear layers—such as SiLU, RMSNorm, and Softmax—still heavily depend on high-precision floating-point operations. In this paper, we propose a calibration-free, dynamic-programming-optimal, and hardware-friendly framework called \underline{N}on-uniform \underline{L}inear \underline{I}nterpolation (NLI). NLI is capable of efficiently approximating a variety of nonlinear functions, enabling seamless integration into LLMs and other deep neural networks with almost no loss in accuracy. NLI ingeniously recasts cutpoint selection as a dynamic-programming problem, achieving the \emph{globally} minimal interpolation error in $\mathcal{O}(M \times N^2)$ time via Bellman’s optimality principle. Based on the NLI algorithm, we also design and implement a plug-and-play universal nonlinear computation unit. Hardware experiments demonstrate that the NLI Engine achieves more than 4× improvement in computational efficiency compared to the state-of-the-art designs.

Shaobo Wang, Yantai Yang, Guo Chen, Peiru Li, Kaixin Li, Yufa Zhou, Zhaorun Chen, Linfeng Zhang

Dataset Distillation (DD) seeks to create a compact dataset from a large, real-world dataset. While recent methods often rely on heuristic approaches to balance efficiency and quality, the fundamental relationship between original and synthetic data remains underexplored. This paper revisits knowledge distillation-based dataset distillation within a solid theoretical framework. We introduce the concepts of Informativeness and Utility, capturing crucial information within a sample and essential samples in the training set, respectively. Building on these principles, we define \textit{optimal dataset distillation} mathematically. We then present InfoUtil, a framework that balances informativeness and utility in synthesizing the distilled dataset. InfoUtil incorporates two key components: (1) game-theoretic informativeness maximization using Shapley Value attribution to extract key information from samples, and (2) principled utility maximization by selecting globally influential samples based on Gradient Norm. These components ensure that the distilled dataset is both informative and utility-optimized. Experiments demonstrate that our method achieves a 6.1\% performance improvement over the previous state-of-the-art approach on ImageNet-1K dataset using ResNet-18.

Tianrui Zhu, Houyuan Chen, Ruihao Gong, Michele Magno, Haotong Qin, Kai Zhang

Video matting is crucial for applications such as film production and virtual reality, yet deploying its computationally intensive models on resource-constrained devices presents challenges. Quantization is a key technique for model compression and acceleration. As an efficient approach, Post-Training Quantization (PTQ) is still in its nascent stages for video matting, facing significant hurdles in maintaining accuracy and temporal coherence. To address these challenges, this paper proposes a novel and general PTQ framework specifically designed for video matting models, marking, to the best of our knowledge, the first systematic attempt in this domain. Our contributions include: (1) A two-stage PTQ strategy that combines block reconstruction-based optimization for fast, stable initial quantization and local dependency capture, followed by a global calibration of quantization parameters to minimize accuracy loss. (2) A Statistically-Driven Global Affine Calibration (GAC) method that enables the network to compensate for cumulative statistical distortions arising from factors such as neglected BN layer effects, even reducing the error of existing PTQ methods on video matting tasks up to 20%. (3) An Optical Flow Assistance (OFA) component that leverages temporal and semantic priors from frames to guide the PTQ process, enhancing the model’s ability to distinguish moving foregrounds in complex scenes and ultimately achieving near full-precision performance even under ultra-low-bit quantization. Comprehensive quantitative and visual results show that our PTQ4VM achieves the state-of-the-art accuracy performance across different bit-widths compared to the existing quantization methods. We highlight that the 4-bit PTQ4VM even achieves performance close to the full-precision counterpart while enjoying 8× FLOP savings.

Deep Learning · Graph Neural Networks

Weining Shi, Zhisen Wen, Qinggang Zhang, Chentao Zhang, Zhihong Zhang

Temporal graphs are commonly employed as conceptual models for capturing time-evolving interactions in real-world systems. Representation learning on such non-Euclidean data typically depends on aggregating information from neighbors, and the presence of temporal dynamics further complicates this process. However, neighbors often contain noisy information in practice, making the unreliable propagation of knowledge and may even lead to the model failure. Although existing methods employ adaptive spatiotemporal neighbor sampling strategies or temporal dependency modeling frameworks to enhance model robustness, their constrained sampling scope limits handling of severe noise and long-term dependencies. This limitation can be attributed to a fundamental cause: neglecting global evolution inherently overlooks the temporal regularities encoded in continuous dynamics. To address this, we propose the **T**emporal **G**raph **T**humbnail (**TGT**), encapsulating a temporal graph’s global evolutionary skeleton as a thumbnail to characterize temporal regularities and enhance model robustness. Specifically, we model the thumbnail by leveraging von Neumann graph entropy and node mutual information to extract essential evolutionary skeleton from the raw temporal graph, and subsequently use it to guide optimization for model learning. In addition to rigorous theoretical derivation, extensive experiments demonstrate that TGT achieves superior capability and robustness compared to baselines, particularly in rapidly evolving and noisy environments. The code is available at https://anonymous.4open.science/r/TGT-BDF2.

Yifeng Wu, Huimin Huang, Shangjie Zhou, Yawen Huang, Hao Zheng, Yun Chen, Xian Wu, Ruize Han, Guanhua Chen

High-resolution visual representation learning remains challenging due to the quadratic complexity of Vision Transformers and the limitations of existing efficient approaches, where fixed scanning patterns in recent Mamba-based models hinder content-adaptive perception. To address these limitations, a novel Information-aware Scanning mechanism (InfoScan) tailored for state-space visual backbones is proposed, which dynamically allocates computational resources to the most salient regions of an image. Specifically, InfoScan rigorously assesses the informativeness of image patches by integrating entropy with local structural analyses, formulates a joint optimization objective balancing fine-grained detail preservation and broader contextual coherence, and learns an adaptive scanning policy via reinforcement learning. Built upon the innovative Visual Information State Space (VISS) block, InfoScan establishes a new family of models that achieve superior efficiency-accuracy trade-offs across diverse tasks. Extensive empirical evaluation in different downstream vision tasks demonstrates that our information-driven dynamic scanning paradigm offers a robust and principled alternative to fixed or global-first traversal methods. Collectively, our work positions adaptive, content-aware processing as a promising and effective new paradigm for efficient high-resolution visual representation.

Reinforcement Learning · Deep RL

Zhongwen Xu, Zihan Ding

We revisit policy-gradient optimization for Large Language Models (LLMs) from a single-stream perspective. Prevailing group-based methods like GRPO reduce variance with on-the-fly baselines but suffer from critical flaws: frequent degenerate groups erase learning signals, and synchronization barriers hinder scalability. We introduce Single-stream Policy Optimization (SPO), which eliminates these issues by design. SPO replaces per-group baselines with a persistent, KL-adaptive value tracker and normalizes advantages globally across the batch, providing a stable, low-variance learning signal for every sample. Being group-free, SPO enables higher throughput and scales effectively in long-horizon or tool-integrated settings where generation times vary. Furthermore, the persistent value tracker naturally enables an adaptive curriculum via prioritized sampling. Experiments using Qwen3-8B show that SPO converges more smoothly and attains higher accuracy than GRPO, while eliminating computation wasted on degenerate groups. Ablation studies confirm that SPO's gains stem from its principled approach to baseline estimation and advantage normalization, offering a more robust and efficient path for LLM reasoning. Across five hard math benchmarks with Qwen3-8B, SPO improves the average maj@32 by $+3.4\ \text{percentage points} (\mathrm{pp})$ over GRPO, driven by substantial absolute point gains on challenging datasets, including $+7.3\ \mathrm{pp}$ on BRUMO 25, $+4.4\ \mathrm{pp}$ on AIME 25, $+3.3\ \mathrm{pp}$ on HMMT 25, and achieves consistent relative gain in pass@$k$ across the evaluated $k$ values. SPO's success challenges the prevailing trend of adding incidental complexity to RL algorithms, highlighting a path where fundamental principles, not architectural workarounds, drive the next wave of progress in LLM reasoning.

Seyed Mohammad Eshtehardian, Mohammad Hossein Yassaee, Babak HosseinKhalaj

Kolmogorov-Arnold Networks (KANs) have emerged as a promising alternative to traditional neural networks, offering enhanced interpretability based on the Kolmogorov-Arnold representation theorem. While their empirical success is growing, a theoretical understanding of their training dynamics remains nascent. This paper investigates the optimization of a two-layer KAN in the overparameterized regime, focusing on a simplified yet insightful setting where only the first-layer coefficients are trained via gradient descent. Our main result establishes that, provided the network is sufficiently wide, this training method is guaranteed to converge to a global minimum and achieve zero training error. Furthermore, we derive a novel, fine-grained convergence rate that explicitly connects the optimization speed to the structure of the data labels through the eigenspectrum of the KAN Tangent Kernel (KAN-TK). Our analysis reveals a key advantage of this architecture: guaranteed convergence is achieved with a hidden layer width of $m=\mathcal{O}(n^2)$, a significant polynomial improvement over the $m=\mathcal{O}(n^6)$ requirement for classic two-layer neural networks using ReLU activation functions and analyzed within the same Tangent Kernel framework. We validate our theoretical findings with numerical experiments that corroborate our predictions on convergence speed and the impact of label structure.

Applications · Robotics

Kangye Ji, Yuan Meng, Hanyun Cui, Ye Li, Jianbo Zhou, Shengjia Hua, Lei Chen, Zhi Wang

Diffusion Policy has demonstrated strong visuomotor modeling capabilities, but its high computational cost renders it impractical for real-time robotic control. Despite huge redundancy across repetitive denoising steps, existing diffusion acceleration techniques fail to generalize to Diffusion Policy due to fundamental architectural and data divergences. In this paper, we propose **B**lock-wise **A**daptive **C**aching (**BAC**), a method to accelerate Diffusion Policy by caching intermediate action features. BAC achieves lossless action generation acceleration by adaptively updating and reusing cached features at the block level, based on a key observation that feature similarities exhibit non-uniform temporal dynamics and distinct block-specific patterns. To operationalize this insight, we first design an Adaptive Caching Scheduler to identify optimal update timesteps by maximizing the global feature similarities between cached and skipped features. However, applying this scheduler for each block leads to significant error surges due to the inter-block propagation of caching errors, particularly within Feed-Forward Network (FFN) blocks. To mitigate this issue, we develop the Bubbling Union Algorithm, which truncates these errors by updating the upstream blocks with significant caching errors before downstream FFNs. As a training-free plugin, BAC is readily integrable with existing transformer-based Diffusion Policy and vision-language-action models. Extensive experiments on multiple robotic benchmarks demonstrate that BAC achieves up to 3x inference speedup for free. Project page: https://block-wise-adaptive-caching.github.io.