Dayu Wang

Dayu Wang, Jiaye Yang, Weikang Li et al.

Reinforcement Learning with Verifiable Rewards (RLVR) has demonstrated remarkable success in enhancing the reasoning capabilities of Large Language Models (LLMs). However, recent studies question whether RL genuinely expands reasoning capacity or merely aligns existing latent capabilities, arguing that exploration remains confined within the pre-trained model's low-rank bias manifold. In this work, we challenge this accessibility boundary hypothesis by demonstrating that the latent reasoning space can be fundamentally expanded through targeted geometric interventions. We propose Manifold-Reshaping Policy Optimization (MRPO), a geometric framework designed to fundamentally restructure the inference space of LLMs. MRPO operates in two stages: first, we employ Spectral Orthogonal Exploration (SOE) to eject the policy initialization into the null space of the bias manifold; second, we integrate an Effective Rank regularization term into the policy optimization objective. This approach incentivizes the discovery and maintenance of high-dimensional reasoning trajectories against the entropy-reducing tendency of standard RL. Empirically, our 4B-parameter method achieves state-of-the-art performance on mathematical tasks, significantly outperforming larger models (e.g., Qwen3-32B) and expanding the capability boundary beyond standard GRPO. Our code is available at https://anonymous.4open.science/r/MRPO-D57B/

Dayu Wang, Jiaye Yang, Weikang Li et al.

Large Language Models (LLMs) often suffer from ''Reasoning Collapse'' on challenging mathematical reasoning tasks, where stochastic sampling produces lexical variations of the same erroneous logic rather than genuine semantic exploration. We observe that failed reasoning traces are often associated with a low-rank bias manifold in the model's hidden-state geometry, which reduces exploration toward corrective solution directions. To address this, we propose Spectral Orthogonal Exploration (SOE), a geometric inference framework under a ''Student Guides Teacher'' paradigm. Instead of using a weak auxiliary agent for imitation, SOE uses it as an orthogonal probe to introduce semantically heterogeneous reasoning signals into the teacher's orthogonal complement of its dominant subspace. This intervention steers the teacher toward more diverse reasoning trajectories and improves exploration beyond standard sampling. Experiments on mathematical benchmarks show that SOE improves average accuracy by 62.4\% and average sampling efficiency by 113.7\% over baseline methods, suggesting that geometric interventions can be effective for mitigating reasoning collapse in mathematical reasoning. We further provide preliminary evidence that SOE is also effective on logic and code generation benchmarks.

Guoqi Yu, Yaoming Li, Xiaoyu Guo et al.

Forecasting models are pivotal in a data-driven world with vast volumes of time series data that appear as a compound of vast Linear and Nonlinear patterns. Recent deep time series forecasting models struggle to utilize seasonal and trend decomposition to separate the entangled components. Such a strategy only explicitly extracts simple linear patterns like trends, leaving the other linear modes and vast unexplored nonlinear patterns to the residual. Their flawed linear and nonlinear feature extraction models and shallow-level decomposition limit their adaptation to the diverse patterns present in real-world scenarios. Given this, we innovate Recursive Residual Decomposition by introducing explicit extraction of both linear and nonlinear patterns. This deeper-level decomposition framework, which is named LiNo, captures linear patterns using a Li block which can be a moving average kernel, and models nonlinear patterns using a No block which can be a Transformer encoder. The extraction of these two patterns is performed alternatively and recursively. To achieve the full potential of LiNo, we develop the current simple linear pattern extractor to a general learnable autoregressive model, and design a novel No block that can handle all essential nonlinear patterns. Remarkably, the proposed LiNo achieves state-of-the-art on thirteen real-world benchmarks under univariate and multivariate forecasting scenarios. Experiments show that current forecasting models can deliver more robust and precise results through this advanced Recursive Residual Decomposition. We hope this work could offer insight into designing more effective forecasting models. Code is available at this Repository: https://github.com/Levi-Ackman/LiNo.

Dayu Wang, Jiaye Yang, Weikang Li et al.

Flow Matching (FM) has recently emerged as a leading approach for high-fidelity visual generation, offering a robust continuous-time alternative to ordinary differential equation (ODE) based models. However, despite their success, FM models are highly sensitive to dataset biases, which cause severe semantic degradation when generating out-of-distribution or minority-class samples. In this paper, we provide a rigorous mathematical formalization of the ``Bias Manifold'' within the FM framework. We identify that this performance drop is driven by conditional expectation smoothing, a mechanism that inevitably leads to trajectory lock-in during inference. To resolve this, we introduce InjectFlow, a novel, training-free method by injecting orthogonal semantics during the initial velocity field computation, without requiring any changes to the random seeds. This design effectively prevents the latent drift toward majority modes while maintaining high generative quality. Extensive experiments demonstrate the effectiveness of our approach. Notably, on the GenEval dataset, InjectFlow successfully fixes 75% of the prompts that standard flow matching models fail to generate correctly. Ultimately, our theoretical analysis and algorithm provide a ready-to-use solution for building more fair and robust visual foundation models.

Yingjie Qi, Jianlei Yang, Yiou Wang et al.

Digital Compute-in-Memory (CIM) architectures have shown great promise in Deep Neural Network (DNN) acceleration by effectively addressing the "memory wall" bottleneck. However, the development and optimization of digital CIM accelerators are hindered by the lack of comprehensive tools that encompass both software and hardware design spaces. Moreover, existing design and evaluation frameworks often lack support for the capacity constraints inherent in digital CIM architectures. In this paper, we present CIMFlow, an integrated framework that provides an out-of-the-box workflow for implementing and evaluating DNN workloads on digital CIM architectures. CIMFlow bridges the compilation and simulation infrastructures with a flexible instruction set architecture (ISA) design, and addresses the constraints of digital CIM through advanced partitioning and parallelism strategies in the compilation flow. Our evaluation demonstrates that CIMFlow enables systematic prototyping and optimization of digital CIM architectures across diverse configurations, providing researchers and designers with an accessible platform for extensive design space exploration.

Dayu Wang, Jiaye Yang, Weikang Li et al.

Recent advances in multi-agent systems highlight the potential of specialized small agents that collaborate via division of labor. Existing tool-integrated reasoning systems, however, often follow a single-agent paradigm in which one large model interleaves long-horizon reasoning with precise tool operations, leading to cognitive-load interference and unstable coordination. We present MSARL, a Multi-Small-Agent Reinforcement Learning framework that explicitly decouples reasoning from tool use. In MSARL, a Reasoning Agent decomposes problems and plans tool invocations, while multiple Tool Agents specialize in specific external tools, each trained via a combination of imitation learning and reinforcement learning with role-specific rewards. On mathematical problem solving with code execution, MSARL significantly improves reasoning stability and final-answer accuracy over single-agent baselines. Moreover, the architecture generalizes to diverse tool-use tasks, demonstrating that cognitive-role decoupling with small agents is a scalable blueprint for multi-agent AI design.