Yeqiu Chen

Ziyan Liu, Zhezheng Hao, Yeqiu Chen et al.

Memory-augmented LLM agents tackle complex long-horizon tasks by recursively summarizing interaction trajectories into compact memory. However, existing approaches typically train these memory policies using outcome-based reinforcement learning, failing to localize where intermediate memory quality degrades. As interactions unfold, ambiguous recursive summaries progressively discard task-relevant information and introduce semantic noise. This exacerbates belief deviation, obscuring the agent's estimate of the latent task state and ultimately derailing long-horizon reasoning. We therefore argue that memory optimization should focus not merely on trajectory-level success, but on the clarity of the belief induced by intermediate summaries. To this end, we introduce Belief Entropy, a self-supervised proxy that probes how uncertain the model remains about the latent task state given its current memory. Based on this proxy, we propose Metacognitive Memory Policy Optimization (MMPO). Instead of relying only on sparse outcome-based signals, MMPO provides fine-grained, memory-specific supervision via explicitly penalizing summaries that induce high epistemic uncertainty. Experiments show that MMPO consistently outperforms existing methods on diverse long-horizon tasks, maintaining 97.1% performance even when scaled to 1.75M-token contexts.

Yeqiu Chen, Ziyan Liu, Zhenxin Huang et al.

Recent progress in LLM reasoning has increasingly shifted from single-pass generation to explicit search over intermediate reasoning states. Tree-of-Thoughts (ToT) organizes inference to tree-structured search with branching and backtracking, but it substantially amplifies the Key--Value (KV) cache: retaining KV states for a frontier of partial trajectories quickly becomes a memory bottleneck that limits throughput and constrains search depth and width under fixed hardware budgets. We address this challenge by observing that KV reuse in ToT-style inference is governed by search dynamics: near-term decoding depends primarily on the active branch and its ancestors, whereas inactive subtrees have low short-term reuse probability yet must remain recoverable for backtracking. Motivated by this, we propose ArborKV, a structure-aware eviction framework that couples a lightweight value estimator with a tree-aware allocation policy, and performs purely token-extractive eviction with lazy rehydration to support revisits. Experiments on ToT-style reasoning benchmarks show that ArborKV achieves up to ~4x peak KV-memory reduction while preserving near-full-retention accuracy, enabling larger search configurations under fixed device budgets that would otherwise run out of memory.

Yeqiu Chen, Ziyan Liu, Hong Wang

Solving large-scale Generalized Eigenvalue Problems (GEPs) is a fundamental yet computationally prohibitive task in science and engineering. As a promising direction, contour integral (CI) methods, such as the CIRR algorithm, offer an efficient and parallelizable framework. However, their performance is critically dependent on the selection of integration contours -- improper selection without reliable prior knowledge of eigenvalue distribution can incur significant computational overhead and compromise numerical accuracy. To address this challenge, we propose DeepContour, a novel hybrid framework that integrates a deep learning-based spectral predictor with Kernel Density Estimation for principled contour design. Specifically, DeepContour first employs a Fourier Neural Operator (FNO) to rapidly predict the spectral distribution of a given GEP. Subsequently, Kernel Density Estimation (KDE) is applied to the predicted spectrum to automatically and systematically determine proper integration contours. Finally, these optimized contours guide the CI solver to efficiently find the desired eigenvalues. We demonstrate the effectiveness of our method on diverse challenging scientific problems. In our main experiments, DeepContour accelerates GEP solving across multiple datasets, achieving up to a 5.63$\times$ speedup. By combining the predictive power of deep learning with the numerical rigor of classical solvers, this work pioneers an efficient and robust paradigm for tackling difficult generalized eigenvalue involving matrices of high dimension.

Ziyan Liu, Yeqiu Chen, Hongyi Cai et al.

Vision-Language-Action (VLA) models have shown great promise for embodied AI, yet the heavy computational cost of processing continuous visual streams severely limits their real-time deployment. Token pruning (keeping salient visual tokens and dropping redundant ones) has emerged as an effective approach for accelerating Vision-Language Models (VLMs), offering a solution for efficient VLA. However, these VLM-specific token pruning methods select tokens based solely on semantic salience metrics (e.g., prefill attention), while overlooking the VLA's intrinsic dual-system nature of high-level semantic understanding and low-level action execution. Consequently, these methods bias token retention toward semantic cues, discard critical information for action generation, and significantly degrade VLA performance. To bridge this gap, we propose VLA-Pruner, a versatile plug-and-play VLA-specific token prune method that aligns with the dual-system nature of VLA models and exploits the temporal continuity in robot manipulation. Specifically, VLA-Pruner adopts a dual-level importance criterion for visual token retention: vision-language prefill attention for semantic-level relevance and action decode attention, estimated via temporal smoothing, for action-level importance. Based on this criterion, VLA-Pruner proposes a novel dual-level token selection strategy that adaptively preserves a compact, informative set of visual tokens for both semantic understanding and action execution under given compute budget. Experiments show that VLA-Pruner achieves state-of-the-art performance across multiple VLA architectures and diverse robotic tasks.

Hong Wang, Jie Wang, Jian Luo et al.

Eigenvalue problems are among the most important topics in many scientific disciplines. With the recent surge and development of machine learning, neural eigenvalue methods have attracted significant attention as a forward pass of inference requires only a tiny fraction of the computation time compared to traditional solvers. However, a key limitation is the requirement for large amounts of labeled data in training, including operators and their eigenvalues. To tackle this limitation, we propose a novel method, named Sorting Chebyshev Subspace Filter (SCSF), which significantly accelerates eigenvalue data generation by leveraging similarities between operators -- a factor overlooked by existing methods. Specifically, SCSF employs truncated fast Fourier transform sorting to group operators with similar eigenvalue distributions and constructs a Chebyshev subspace filter that leverages eigenpairs from previously solved problems to assist in solving subsequent ones, reducing redundant computations. To the best of our knowledge, SCSF is the first method to accelerate eigenvalue data generation. Experimental results show that SCSF achieves up to a $3.5\times$ speedup compared to various numerical solvers.