Siwei Chen

Yujie Wang, Siwei Chen, Longzan Luo et al.

Reinforcement Learning (RL) has become pivotal for improving model capabilities yet suffers from rollout efficiency bottlenecks due to the long-tail response length distribution. While existing works mitigate the impact of long tails via prompt-level tail scheduling, we focus on the root source of inefficiency: the distribution itself. Specifically, we characterize the long-tail distribution at a finer granularity, identifying intra-prompt long tails, and revealing that they frequently consist of ineffective verbosity. To address this, we propose a novel paradigm of active distribution shaping to shape the rollout distribution towards conciseness and certainty, thereby fundamentally resolving tail-induced overheads. We achieve this through a distribution-aware trajectory sampling mechanism, which selects trajectories from a redundant exploration space for each prompt, and an adaptive redundancy allocation scheme to maximize both shaping effectiveness and system efficiency. Experiments demonstrate significant acceleration over state-of-the-art systems by up to 1.77x without compromising model performance.

Jiawei Ren, Cunjun Yu, Siwei Chen et al.

Motion mimicking is a foundational task in physics-based character animation. However, most existing motion mimicking methods are built upon reinforcement learning (RL) and suffer from heavy reward engineering, high variance, and slow convergence with hard explorations. Specifically, they usually take tens of hours or even days of training to mimic a simple motion sequence, resulting in poor scalability. In this work, we leverage differentiable physics simulators (DPS) and propose an efficient motion mimicking method dubbed DiffMimic. Our key insight is that DPS casts a complex policy learning task to a much simpler state matching problem. In particular, DPS learns a stable policy by analytical gradients with ground-truth physical priors hence leading to significantly faster and stabler convergence than RL-based methods. Moreover, to escape from local optima, we utilize a Demonstration Replay mechanism to enable stable gradient backpropagation in a long horizon. Extensive experiments on standard benchmarks show that DiffMimic has a better sample efficiency and time efficiency than existing methods (e.g., DeepMimic). Notably, DiffMimic allows a physically simulated character to learn Backflip after 10 minutes of training and be able to cycle it after 3 hours of training, while the existing approach may require about a day of training to cycle Backflip. More importantly, we hope DiffMimic can benefit more differentiable animation systems with techniques like differentiable clothes simulation in future research.

Siwei Chen, Xiao Ma, Zhongwen Xu

Existing imitation learning (IL) methods such as inverse reinforcement learning (IRL) usually have a double-loop training process, alternating between learning a reward function and a policy and tend to suffer long training time and high variance. In this work, we identify the benefits of differentiable physics simulators and propose a new IL method, i.e., Imitation Learning via Differentiable Physics (ILD), which gets rid of the double-loop design and achieves significant improvements in final performance, convergence speed, and stability. The proposed ILD incorporates the differentiable physics simulator as a physics prior into its computational graph for policy learning. It unrolls the dynamics by sampling actions from a parameterized policy, simply minimizing the distance between the expert trajectory and the agent trajectory, and back-propagating the gradient into the policy via temporal physics operators. With the physics prior, ILD policies can not only be transferable to unseen environment specifications but also yield higher final performance on a variety of tasks. In addition, ILD naturally forms a single-loop structure, which significantly improves the stability and training speed. To simplify the complex optimization landscape induced by temporal physics operations, ILD dynamically selects the learning objectives for each state during optimization. In our experiments, we show that ILD outperforms state-of-the-art methods in a variety of continuous control tasks with Brax, requiring only one expert demonstration. In addition, ILD can be applied to challenging deformable object manipulation tasks and can be generalized to unseen configurations.

Zhepeng Wang, Feng Chen, Kanokphan Lertniphonphan et al.

This report presents our Le3DE2E solution for unified sensor-based detection, tracking, and forecasting in Argoverse Challenges at CVPR 2023 Workshop on Autonomous Driving (WAD). We propose a unified network that incorporates three tasks, including detection, tracking, and forecasting. This solution adopts a strong Bird's Eye View (BEV) encoder with spatial and temporal fusion and generates unified representations for multi-tasks. The solution was tested in the Argoverse 2 sensor dataset to evaluate the detection, tracking, and forecasting of 26 object categories. We achieved 1st place in Detection, Tracking, and Forecasting on the E2E Forecasting track in Argoverse Challenges at CVPR 2023 WAD.

Siwei Chen, Anxing Xiao, David Hsu

This work addresses the problem of long-horizon task planning with the Large Language Model (LLM) in an open-world household environment. Existing works fail to explicitly track key objects and attributes, leading to erroneous decisions in long-horizon tasks, or rely on highly engineered state features and feedback, which is not generalizable. We propose an open state representation that provides continuous expansion and updating of object attributes from the LLM's inherent capabilities for context understanding and historical action reasoning. Our proposed representation maintains a comprehensive record of an object's attributes and changes, enabling robust retrospective summary of the sequence of actions leading to the current state. This allows continuously updating world model to enhance context understanding for decision-making in task planning. We validate our model through experiments across simulated and real-world task planning scenarios, demonstrating significant improvements over baseline methods in a variety of tasks requiring long-horizon state tracking and reasoning. (Video\footnote{Video demonstration: \url{https://youtu.be/QkN-8pxV3Mo}.})

Peipei Xie, Siwei Chen, Zejun Xiang et al.

At SAC 2013, Berger et al. first proposed the Extended Generalized Feistel Networks (EGFN) structure for the design of block ciphers with efficient diffusion. Later, based on the Type-2 EGFN, they instantiated a new lightweight block cipher named Lilliput (published in IEEE Transactions on Computers, Vol. 65, Issue 7, 2016). According to published cryptanalysis results, Lilliput is sufficiently secure against theoretical attacks such as differential, linear, boomerang, and integral attacks, which rely on the statistical properties of plaintext and ciphertext. However, there is a lack of analysis regarding its resistance to physical attacks in real-world scenarios, such as fault attacks. In this paper, we present the first systematic differential fault analysis (DFA) of Lilliput under three nibble-oriented fault models with progressively relaxed adversarial assumptions to comprehensively assess its fault resilience. In Model I (multi-round fixed-location), precise fault injections at specific rounds recover the master key with a 98% success rate using only 8 faults. Model II (single-round fixed-location) relaxes the multi-round requirement, demonstrating that 8 faults confined to a single round are still sufficient to achieve a 99% success rate by exploiting Lilliput's diffusion properties and DDT-based constraints. Model III (single-round random-location) further weakens the assumption by allowing faults to occur randomly among the eight rightmost branches of round 27. By uniquely identifying the fault location from ciphertext differences with high probability, the attack remains highly feasible, achieving over 99% success with 33 faults and exceeding 99.5% with 36 faults. Our findings reveal a significant vulnerability of Lilliput to practical fault attacks across different adversary capabilities in real-world scenarios, providing crucial insights for its secure implementation.

Yinsong Wang, Siwei Chen, Ziyi Song et al.

Cooperative perception research is hindered by the limited availability of datasets that capture the complexity of real-world Vehicle-to-Everything (V2X) interactions, particularly under dynamic communication constraints. To address this gap, we introduce WHALES (Wireless enhanced Autonomous vehicles with Large number of Engaged agents), the first large-scale V2X dataset explicitly designed to benchmark communication-aware agent scheduling and scalable cooperative perception. WHALES introduces a new benchmark that enables state-of-the-art (SOTA) research in communication-aware cooperative perception, featuring an average of 8.4 cooperative agents per scene and 2.01 million annotated 3D objects across diverse traffic scenarios. It incorporates detailed communication metadata to emulate real-world communication bottlenecks, enabling rigorous evaluation of scheduling strategies. To further advance the field, we propose the Coverage-Aware Historical Scheduler (CAHS), a novel scheduling baseline that selects agents based on historical viewpoint coverage, improving perception performance over existing SOTA methods. WHALES bridges the gap between simulated and real-world V2X challenges, providing a robust framework for exploring perception-scheduling co-design, cross-data generalization, and scalability limits. The WHALES dataset and code are available at https://github.com/chensiweiTHU/WHALES.

Pengfei Zheng, Kanokphan Lertniphonphan, Feng Chen et al.

This report presents our Le3DE2E_Occ solution for 4D Occupancy Forecasting in Argoverse Challenges at CVPR 2023 Workshop on Autonomous Driving (WAD). Our solution consists of a strong LiDAR-based Bird's Eye View (BEV) encoder with temporal fusion and a two-stage decoder, which combines a DETR head and a UNet decoder. The solution was tested on the Argoverse 2 sensor dataset to evaluate the occupancy state 3 seconds in the future. Our solution achieved 18% lower L1 Error (3.57) than the baseline and got the 1 place on the 4D Occupancy Forecasting task in Argoverse Challenges at CVPR 2023.

Xiao Ma, Siwei Chen, David Hsu et al.

Deep reinforcement learning (DRL) has achieved significant success in various robot tasks: manipulation, navigation, etc. However, complex visual observations in natural environments remains a major challenge. This paper presents Contrastive Variational Reinforcement Learning (CVRL), a model-based method that tackles complex visual observations in DRL. CVRL learns a contrastive variational model by maximizing the mutual information between latent states and observations discriminatively, through contrastive learning. It avoids modeling the complex observation space unnecessarily, as the commonly used generative observation model often does, and is significantly more robust. CVRL achieves comparable performance with state-of-the-art model-based DRL methods on standard Mujoco tasks. It significantly outperforms them on Natural Mujoco tasks and a robot box-pushing task with complex observations, e.g., dynamic shadows. The CVRL code is available publicly at https://github.com/Yusufma03/CVRL.

Yaqi Zhao, Yuanyang Yin, Lin Li et al.

Does seeing always mean knowing? Large Vision-Language Models (LVLMs) integrate separately pre-trained vision and language components, often using CLIP-ViT as vision backbone. However, these models frequently encounter a core issue of "cognitive misalignment" between the vision encoder (VE) and the large language model (LLM). Specifically, the VE's representation of visual information may not fully align with LLM's cognitive framework, leading to a mismatch where visual features exceed the language model's interpretive range. To address this, we investigate how variations in VE representations influence LVLM comprehension, especially when the LLM faces VE-Unknown data-images whose ambiguous visual representations challenge the VE's interpretive precision. Accordingly, we construct a multi-granularity landmark dataset and systematically examine the impact of VE-Known and VE-Unknown data on interpretive abilities. Our results show that VE-Unknown data limits LVLM's capacity for accurate understanding, while VE-Known data, rich in distinctive features, helps reduce cognitive misalignment. Building on these insights, we propose Entity-Enhanced Cognitive Alignment (EECA), a method that employs multi-granularity supervision to generate visually enriched, well-aligned tokens that not only integrate within the LLM's embedding space but also align with the LLM's cognitive framework. This alignment markedly enhances LVLM performance in landmark recognition. Our findings underscore the challenges posed by VE-Unknown data and highlight the essential role of cognitive alignment in advancing multimodal systems.

Xinyue Zeng, Tuo Wang, Adithya Kulkarni et al.

Recent advances in protein structure prediction have achieved near-atomic accuracy for well-folded proteins. However, current benchmarks inadequately assess model performance in biologically challenging contexts, especially those involving intrinsically disordered regions (IDRs), limiting their utility in applications such as drug discovery, disease variant interpretation, and protein interface design. We introduce DisProtBench, a comprehensive benchmark for evaluating protein structure prediction models (PSPMs) under structural disorder and complex biological conditions. DisProtBench spans three key axes: (1) Data complexity, covering disordered regions, G protein-coupled receptor (GPCR) ligand pairs, and multimeric complexes; (2) Task diversity, benchmarking twelve leading PSPMs across structure-based tasks with unified classification, regression, and interface metrics; and (3) Interpretability, via the DisProtBench Portal, which provides precomputed 3D structures and visual error analyses. Our results reveal significant variability in model robustness under disorder, with low-confidence regions linked to functional prediction failures. Notably, global accuracy metrics often fail to predict task performance in disordered settings, emphasizing the need for function-aware evaluation. DisProtBench establishes a reproducible, extensible, and biologically grounded framework for assessing next-generation PSPMs in realistic biomedical scenarios.

Xiyuan Wang, Yewei Liu, Lexi Pang et al.

Diffusion models have gained popularity in graph generation tasks; however, the extent of their expressivity concerning the graph distributions they can learn is not fully understood. Unlike models in other domains, popular backbones for graph diffusion models, such as Graph Transformers, do not possess universal expressivity to accurately model the distribution scores of complex graph data. Our work addresses this limitation by focusing on the frequency of specific substructures as a key characteristic of target graph distributions. When evaluating existing models using this metric, we find that they fail to maintain the distribution of substructure counts observed in the training set when generating new graphs. To address this issue, we establish a theoretical connection between the expressivity of Graph Neural Networks (GNNs) and the overall performance of graph diffusion models, demonstrating that more expressive GNN backbones can better capture complex distribution patterns. By integrating advanced GNNs into the backbone architecture, we achieve significant improvements in substructure generation.

Yanfeng Yang, Siwei Chen, Pingping Hu et al.

Probabilistic forecasting of multivariate time series is challenging due to non-stationarity, inter-variable dependencies, and distribution shifts. While recent diffusion and flow matching models have shown promise, they often ignore informative priors such as conditional means and covariances. In this work, we propose Conditionally Whitened Generative Models (CW-Gen), a framework that incorporates prior information through conditional whitening. Theoretically, we establish sufficient conditions under which replacing the traditional terminal distribution of diffusion models, namely the standard multivariate normal, with a multivariate normal distribution parameterized by estimators of the conditional mean and covariance improves sample quality. Guided by this analysis, we design a novel Joint Mean-Covariance Estimator (JMCE) that simultaneously learns the conditional mean and sliding-window covariance. Building on JMCE, we introduce Conditionally Whitened Diffusion Models (CW-Diff) and extend them to Conditionally Whitened Flow Matching (CW-Flow). Experiments on five real-world datasets with six state-of-the-art generative models demonstrate that CW-Gen consistently enhances predictive performance, capturing non-stationary dynamics and inter-variable correlations more effectively than prior-free approaches. Empirical results further demonstrate that CW-Gen can effectively mitigate the effects of distribution shift.

Siwei Chen, Xiao Ma, Yunfan Lu et al.

This paper presents Particle-based Object Manipulation (Prompt), a new approach to robot manipulation of novel objects ab initio, without prior object models or pre-training on a large object data set. The key element of Prompt is a particle-based object representation, in which each particle represents a point in the object, the local geometric, physical, and other features of the point, and also its relation with other particles. Like the model-based analytic approaches to manipulation, the particle representation enables the robot to reason about the object's geometry and dynamics in order to choose suitable manipulation actions. Like the data-driven approaches, the particle representation is learned online in real-time from visual sensor input, specifically, multi-view RGB images. The particle representation thus connects visual perception with robot control. Prompt combines the benefits of both model-based reasoning and data-driven learning. We show empirically that Prompt successfully handles a variety of everyday objects, some of which are transparent. It handles various manipulation tasks, including grasping, pushing, etc,. Our experiments also show that Prompt outperforms a state-of-the-art data-driven grasping method on the daily objects, even though it does not use any offline training data.

Jeffrey Fong, Siwei Chen, Kaiqi Chen

Training neural networks with large batch is of fundamental significance to deep learning. Large batch training remarkably reduces the amount of training time but has difficulties in maintaining accuracy. Recent works have put forward optimization methods such as LARS and LAMB to tackle this issue through adaptive layer-wise optimization using trust ratios. Though prevailing, such methods are observed to still suffer from unstable and extreme trust ratios which degrades performance. In this paper, we propose a new variant of LAMB, called LAMBC, which employs trust ratio clipping to stabilize its magnitude and prevent extreme values. We conducted experiments on image classification tasks such as ImageNet and CIFAR-10 and our empirical results demonstrate promising improvements across different batch sizes.

Siwei Chen, Xiao Ma, David Hsu

It has been arduous to assess the progress of a policy learning algorithm in the domain of hierarchical task with high dimensional action space due to the lack of a commonly accepted benchmark. In this work, we propose a new light-weight benchmark task called Diner Dash for evaluating the performance in a complicated task with high dimensional action space. In contrast to the traditional Atari games that only have a flat structure of goals and very few actions, the proposed benchmark task has a hierarchical task structure and size of 57 for the action space and hence can facilitate the development of policy learning in complicated tasks. On top of that, we introduce Decomposed Policy Graph Modelling (DPGM), an algorithm that combines both graph modelling and deep learning to allow explicit domain knowledge embedding and achieves significant improvement comparing to the baseline. In the experiments, we have shown the effectiveness of the domain knowledge injection via a specially designed imitation algorithm as well as results of other popular algorithms.