arXiv subjects

Jinyi Liu

Publications and source records attributed to Jinyi Liu.

At least 19 recordsLinked to original sources

arXiv2026

WarpSAC: Towards the Pinnacle of Scalable Off-policy RL by Rethinking Exploration and Exploitation

Massively parallel simulation changes the data regime in which off-policy reinforcement learning (RL) is trained, challenging stabilizers designed for data-limited replay. Through controlled experiments across eight benchmark families, we show that these stabilizers are data-regime-dependent: parameter normalization helps with narrow replay coverage but restricts value fitting when data are abundant, while clipped double-Q can be relaxed in high-throughput manipulation. Age-biased replay weighting improves learning efficiency across regimes, especially with limited network capacity. Based on these findings, we propose WarpSAC, a regime-aware family of off-policy RL algorithms. WarpSAC uses Sample Weight Decay for efficient exploitation and provides two variants: WarpSAC-L (Norm ON, clipped double-Q) for data-limited CPU-scale training, and WarpSAC-A (Norm OFF, single-Q) for data-abundant GPU-parallel training. WarpSAC improves normalized score--step AUC over FlashSAC by 4.5% across nine CPU-scale environments and 23.1% across fourteen GPU-parallel environments. It increases UnitreeG1TransportBox-v1 success rate from 19.8% to 96.4%, improves mean normalized wall-time AUC on MuJoCo Playground by 19.1%, and achieves 36.4% faster sim-to-real deployment on Unitree G1 than FlashSAC. These results show that scalable off-policy RL should adapt its stabilizers to the available data regime.

cs.LG↗

arXiv2026

Prox: Training-Free FFN Activation Sparsity via Approximate Intermediate-Channel Salience in LLMs

Feed-forward networks (FFNs) dominate memory traffic and computation in large language model (LLM) inference, making them a primary target for activation sparsification. However, existing training-free methods suffer substantial model-quality degradation at high sparsity due to limitations in their channel-selection strategies. We observe that the SwiGLU intermediate state provides a highly effective channel-selection signal, but obtaining it requires costly dense computation. To address this, we present \emph{Prox}, a two-stage training-free framework for sparse SwiGLU FFNs. Prox hinges on the key insight: sparse execution requires only the channel mask induced by the intermediate state, which can be constructed from the magnitude ranking of its entries rather than their exact values. Specifically, Stage 1 uses input sparsity and quantized proxy weights to construct a shared mask; Stage 2 computes the selected channels exactly, enabling sparse execution of all three projections. Across ten LLMs from six model families, Prox outperforms training-free baselines at all sparsity levels, achieves up to a $1.99\times$ end-to-end decoding speedup at 70\% FFN sparsity, and is compatible with quantization and sparse attention.

cs.LG↗

arXiv2026

Demystifying Deep Learning Compiler Frontend Bugs: An LLM-Aided Empirical Study

Deep learning compilers (DLCs) are designed to translate deep learning programs into optimized, hardware-specific code. Typically, DLC frontends translate programs into graph-based intermediate representations (IRs) to enable optimizations. Defects introduced during this stage (termed \emph{fBug}s) are severe yet understudied, as prior work predominantly focuses on low-level APIs and operators or treats DLCs as monolithic entities. To bridge this gap, we conduct the first systematic empirical study of \emph{fBug}s in TorchDynamo, the default DLC frontend for PyTorch 2, the most popular DL framework. Leveraging a domain-knowledge-enhanced LLM-aided methodology, we analyze 123 \emph{fBug}s and construct a taxonomy comprising 7 root cause categories and 15 subcategories. Our findings provide actionable insights for DLC development and testing. Furthermore, we leverage the LLM to generate targeted, root cause-aware test cases to detect new bugs. We uncovered 23 previously unknown \emph{fBug}s in recent releases (15 confirmed) across eight (sub)categories, demonstrating the efficacy of our methodology in testing and hardening DLC frontends.

cs.PL↗

arXiv2026

The Mirage of Optimizing Training Policies: Monotonic Inference Policies as the Real Objective for LLM Reinforcement Learning

Reinforcement learning (RL) has gained growing attention in large language model (LLM) post-training, yet RL training remains fragile and can suffer from instability or collapse. One vital cause is training-inference mismatch: LLM adopts separate inference and training engines for generation efficiency and training precision, which in practice exhibits inconsistent probabilities for the same trajectories on training and inference sides, even with synchronized model parameters. This naturally induces a special type of off-policyness ever existing and poisoning the training. Prior works have made various efforts in addressing the off-policyness to stabilize the training policies under the mismatch. In this paper, we point out the objective misalignment neglected by existing works that an effective update to the policy in the training engine not necessarily ensures the improvement of the inference policy, i.e., the one used in deployment. To this end, we propose a new policy optimization objective for LLM RL, named Monotonic Inference Policy Improvement (MIPI). Following this principle, we introduce Monotonic Inference Policy Update (MIPU), a two-step LLM RL framework that constructs sampler-referenced candidate updates and selectively accepts synchronized candidates using an inference-side gap proxy. Experiments conducted on two model scales under high mismatch show that MIPU improves average reasoning performance and training stability.

cs.LG↗

arXiv2026

Intensity-guided pose-free multiview fusion for single photon sensing

Single-photon light detection and ranging (LiDAR) extends active three-dimensional sensing at the fundamental level and has found applications in extreme environments involving long-range operation, low-reflectance targets, and adverse visibility. However, the acquired measurements often give rise to single-photon point clouds that are sparse, spatially non-uniform, and corrupted by outliers and depth distortions, making multi-view registration challenging especially when sensor poses are not accurately known. In this work, we present a geometry-intensity coupled registration framework (GIC-Reg) of pose-free multi-view fusion for single-photon sensing. It is established by combining physical-aware preprocessing, joint geometry-intensity grid feature aggregation, global matching, and local ambiguity disambiguation to estimate inter-view rigid transformations and hence to construct a globally consistent reconstruction. On the synthetic benchmark, it admits the lowest relative rotation error (RRE), relative translation error, and root mean square error across all background-noise and dropout rates, in comparison to baselines. Notably, under the most degraded dropout, it reduces the RRE from $13.167^\circ$ to $8.459^\circ$ compared with the learning-based baseline. Furthermore, experimental results on real multi-view data acquired at about 80~m show that it achieves more reliable global orientation and local alignment. Our results show that photon intensity provides an effective physical cue for stabilizing multiview registration in single-photon point cloud, and thus our work aids significant progress in exploring practical utility of single-photon sensing.

physics.optics↗

arXiv2025

Key Decision-Makers in Multi-Agent Debates: Who Holds the Power?

Recent studies on LLM agent scaling have highlighted the potential of Multi-Agent Debate (MAD) to enhance reasoning abilities. However, the critical aspect of role allocation strategies remains underexplored. In this study, we demonstrate that allocating roles with differing viewpoints to specific positions significantly impacts MAD's performance in reasoning tasks. Specifically, we find a novel role allocation strategy, "Truth Last", which can improve MAD performance by up to 22% in reasoning tasks. To address the issue of unknown truth in practical applications, we propose the Multi-Agent Debate Consistency (MADC) strategy, which systematically simulates and optimizes its core mechanisms. MADC incorporates path consistency to assess agreement among independent roles, simulating the role with the highest consistency score as the truth. We validated MADC across a range of LLMs (9 models), including the DeepSeek-R1 Distilled Models, on challenging reasoning tasks. MADC consistently demonstrated advanced performance, effectively overcoming MAD's performance bottlenecks and providing a crucial pathway for further improvements in LLM agent scaling.

cs.AI↗

arXiv2025

Point upsampling networks for single-photon sensing

Single-photon sensing has generated great interest as a prominent technique of long-distance and ultra-sensitive imaging, however, it tends to yield sparse and spatially biased point clouds, thus limiting its practical utility. In this work, we propose using point upsampling networks to increase point density and reduce spatial distortion in single-photon point cloud. Particularly, our network is built on the state space model which integrates a multi-path scanning mechanism to enrich spatial context, a bidirectional Mamba backbone to capture global geometry and local details, and an adaptive upsample shift module to correct offset-induced distortions. Extensive experiments are implemented on commonly-used datasets to confirm its high reconstruction accuracy and strong robustness to the distortion noise, and also on real-world data to demonstrate that our model is able to generate visually consistent, detail-preserving, and noise suppressed point clouds. Our work is the first to establish the upsampling framework for single-photon sensing, and hence opens a new avenue for single-photon sensing and its practical applications in the downstreaming tasks.

physics.optics↗

arXiv2025

Squeeze the Soaked Sponge: Efficient Off-policy Reinforcement Finetuning for Large Language Model

Reinforcement Learning (RL) has demonstrated its potential to improve the reasoning ability of Large Language Models (LLMs). One major limitation of most existing Reinforcement Finetuning (RFT) methods is that they are on-policy RL in nature, i.e., data generated during the past learning process is not fully utilized. This inevitably comes at a significant cost of compute and time, posing a stringent bottleneck on continuing economic and efficient scaling. To this end, we launch the renaissance of off-policy RL and propose Reincarnating Mix-policy Proximal Policy Gradient (ReMix), a general approach to enable on-policy RFT methods like PPO and GRPO to leverage off-policy data. ReMix consists of three major components: (1) Mix-policy proximal policy gradient with an increased Update-To-Data (UTD) ratio for efficient training; (2) KL-Convex policy constraint to balance the trade-off between stability and flexibility; (3) Policy reincarnation to achieve a seamless transition from efficient early-stage learning to steady asymptotic improvement. In our experiments, we train a series of ReMix models upon PPO, GRPO and 1.5B, 7B base models. ReMix shows an average Pass@1 accuracy of 52.10% (for 1.5B model) with 0.079M response rollouts, 350 training steps and achieves 63.27%/64.39% (for 7B model) with 0.007M/0.011M response rollouts, 50/75 training steps, on five math reasoning benchmarks (i.e., AIME'24, AMC'23, Minerva, OlympiadBench, and MATH500). Compared with 15 recent advanced models, ReMix shows SOTA-level performance with an over 30x to 450x reduction in training cost in terms of rollout data volume. In addition, we reveal insightful findings via multifaceted analysis, including the implicit preference for shorter responses due to the Whipping Effect of off-policy discrepancy, the collapse mode of self-reflection behavior under the presence of severe off-policyness, etc.

cs.LG↗

arXiv2025

Prospects for radio weak lensing: studies using LOFAR observations in the ELAIS-N1 field

We carry out a shape and weak lensing analysis of Low Frequency Array (LOFAR) radio sources and Hyper Suprime-Cam (HSC) optical sources within the European Large Area Infrared Space Observatory Survey-North 1 (ELAIS-N1) field. Using HSC data alone, we detect a cosmic shear correlation signal at a significance of $\sim$$9\sigma$ over a $\sim$$6.4$ deg$^2$ region. For the radio dataset, we analyse observations from both the LOFAR Two Metre Sky Survey (LoTSS) and the International LOFAR Telescope (ILT). While LoTSS provides the deepest radio imaging of ELAIS-N1 with a central source density of $\sim$2.7 arcmin$^{-2}$, its $6^{\prime\prime}$ resolution limits the accuracy of shape measurements. But, using LoTSS-matched HSC sources, we show that accurate radio shape measurements would enable us to measure the amplitude of the shear correlation function at least at $\sim$2$\sigma$ significance. In contrast, ILT observation of the field offers a superior $0.3^{\prime\prime}$ resolution. By cross-matching HSC and ILT samples, we measure a position angle correlation of $R_{\cos(2\alpha)} = 0.15 \pm 0.02$. This result highlights ILT's ability to resolve extended and diffuse emission. The current ILT observations lack the required depth for robust weak lensing measurements. To assess the potential of ILT, we use simulated data with increased observation hours. Our analysis indicates that with 3200 hours of ILT observations or deeper data, and assuming that statistical errors dominate over systematics, a shear correlation could be detected with moderate significance. To achieve this will require precise radio shear measurements and effective mitigation of point spread function (PSF) systematics.

astro-ph.CO↗

arXiv2025

From Seeing to Doing: Bridging Reasoning and Decision for Robotic Manipulation

Achieving generalization in robotic manipulation remains a critical challenge, particularly for unseen scenarios and novel tasks. Current Vision-Language-Action (VLA) models, while building on top of general Vision-Language Models (VLMs), still fall short of achieving robust zero-shot performance due to the scarcity and heterogeneity prevalent in embodied datasets. To address these limitations, we propose FSD (From Seeing to Doing), a novel vision-language model that generates intermediate representations through spatial relationship reasoning, providing fine-grained guidance for robotic manipulation. Our approach combines a hierarchical data pipeline for training with a self-consistency mechanism that aligns spatial coordinates with visual signals. Through extensive experiments, we comprehensively validated FSD's capabilities in both "seeing" and "doing," achieving outstanding performance across 8 benchmarks for general spatial reasoning and embodied reference abilities, as well as on our proposed more challenging benchmark VABench. We also verified zero-shot capabilities in robot manipulation, demonstrating significant performance improvements over baseline methods in both SimplerEnv and real robot settings. Experimental results show that FSD achieves 40.6% success rate in SimplerEnv and 72% success rate across 8 real-world tasks, outperforming the strongest baseline by 30%.

cs.RO↗

arXiv2025

DualRAG: A Dual-Process Approach to Integrate Reasoning and Retrieval for Multi-Hop Question Answering

Multi-Hop Question Answering (MHQA) tasks permeate real-world applications, posing challenges in orchestrating multi-step reasoning across diverse knowledge domains. While existing approaches have been improved with iterative retrieval, they still struggle to identify and organize dynamic knowledge. To address this, we propose DualRAG, a synergistic dual-process framework that seamlessly integrates reasoning and retrieval. DualRAG operates through two tightly coupled processes: Reasoning-augmented Querying (RaQ) and progressive Knowledge Aggregation (pKA). They work in concert: as RaQ navigates the reasoning path and generates targeted queries, pKA ensures that newly acquired knowledge is systematically integrated to support coherent reasoning. This creates a virtuous cycle of knowledge enrichment and reasoning refinement. Through targeted fine-tuning, DualRAG preserves its sophisticated reasoning and retrieval capabilities even in smaller-scale models, demonstrating its versatility and core advantages across different scales. Extensive experiments demonstrate that this dual-process approach substantially improves answer accuracy and coherence, approaching, and in some cases surpassing, the performance achieved with oracle knowledge access. These results establish DualRAG as a robust and efficient solution for complex multi-hop reasoning tasks.

cs.LG↗

arXiv2025

From Chaos to Order: The Atomic Reasoner Framework for Fine-grained Reasoning in Large Language Models

Recent advances in large language models (LLMs) have shown remarkable progress, yet their capacity for logical ``slow-thinking'' reasoning persists as a critical research frontier. Current inference scaling paradigms suffer from two fundamental constraints: fragmented thought flows compromising logical coherence, and intensively computational complexity that escalates with search space dimensions. To overcome these limitations, we present \textbf{Atomic Reasoner} (\textbf{AR}), a cognitive inference strategy that enables fine-grained reasoning through systematic atomic-level operations. AR decomposes the reasoning process into atomic cognitive units, employing a cognitive routing mechanism to dynamically construct reasoning representations and orchestrate inference pathways. This systematic methodology implements stepwise, structured cognition, which ensures logical coherence while significantly reducing cognitive load, effectively simulating the cognitive patterns observed in human deep thinking processes. Extensive experimental results demonstrate AR's superior reasoning capabilities without the computational burden of exhaustive solution searches, particularly excelling in linguistic logic puzzles. These findings substantiate AR's effectiveness in enhancing LLMs' capacity for robust, long-sequence logical reasoning and deliberation.

cs.CL↗

arXiv2024

ENOTO: Improving Offline-to-Online Reinforcement Learning with Q-Ensembles

Offline reinforcement learning (RL) is a learning paradigm where an agent learns from a fixed dataset of experience. However, learning solely from a static dataset can limit the performance due to the lack of exploration. To overcome it, offline-to-online RL combines offline pre-training with online fine-tuning, which enables the agent to further refine its policy by interacting with the environment in real-time. Despite its benefits, existing offline-to-online RL methods suffer from performance degradation and slow improvement during the online phase. To tackle these challenges, we propose a novel framework called ENsemble-based Offline-To-Online (ENOTO) RL. By increasing the number of Q-networks, we seamlessly bridge offline pre-training and online fine-tuning without degrading performance. Moreover, to expedite online performance enhancement, we appropriately loosen the pessimism of Q-value estimation and incorporate ensemble-based exploration mechanisms into our framework. Experimental results demonstrate that ENOTO can substantially improve the training stability, learning efficiency, and final performance of existing offline RL methods during online fine-tuning on a range of locomotion and navigation tasks, significantly outperforming existing offline-to-online RL methods.

cs.LG↗

arXiv2024

CellAgent: An LLM-driven Multi-Agent Framework for Automated Single-cell Data Analysis

Single-cell RNA sequencing (scRNA-seq) data analysis is crucial for biological research, as it enables the precise characterization of cellular heterogeneity. However, manual manipulation of various tools to achieve desired outcomes can be labor-intensive for researchers. To address this, we introduce CellAgent (http://cell.agent4science.cn/), an LLM-driven multi-agent framework, specifically designed for the automatic processing and execution of scRNA-seq data analysis tasks, providing high-quality results with no human intervention. Firstly, to adapt general LLMs to the biological field, CellAgent constructs LLM-driven biological expert roles - planner, executor, and evaluator - each with specific responsibilities. Then, CellAgent introduces a hierarchical decision-making mechanism to coordinate these biological experts, effectively driving the planning and step-by-step execution of complex data analysis tasks. Furthermore, we propose a self-iterative optimization mechanism, enabling CellAgent to autonomously evaluate and optimize solutions, thereby guaranteeing output quality. We evaluate CellAgent on a comprehensive benchmark dataset encompassing dozens of tissues and hundreds of distinct cell types. Evaluation results consistently show that CellAgent effectively identifies the most suitable tools and hyperparameters for single-cell analysis tasks, achieving optimal performance. This automated framework dramatically reduces the workload for science data analyses, bringing us into the "Agent for Science" era.

cs.AI↗

arXiv2024

A Novel Paradigm for Neural Computation: X-Net with Learnable Neurons and Adaptable Structure

Multilayer perception (MLP) has permeated various disciplinary domains, ranging from bioinformatics to financial analytics, where their application has become an indispensable facet of contemporary scientific research endeavors. However, MLP has obvious drawbacks. 1), The type of activation function is single and relatively fixed, which leads to poor `representation ability' of the network, and it is often to solve simple problems with complex networks; 2), the network structure is not adaptive, it is easy to cause network structure redundant or insufficient. In this work, we propose a novel neural network paradigm X-Net promising to replace MLPs. X-Net can dynamically learn activation functions individually based on derivative information during training to improve the network's representational ability for specific tasks. At the same time, X-Net can precisely adjust the network structure at the neuron level to accommodate tasks of varying complexity and reduce computational costs. We show that X-Net outperforms MLPs in terms of representational capability. X-Net can achieve comparable or even better performance than MLP with much smaller parameters on regression and classification tasks. Specifically, in terms of the number of parameters, X-Net is only 3% of MLP on average and only 1.1% under some tasks. We also demonstrate X-Net's ability to perform scientific discovery on data from various disciplines such as energy, environment, and aerospace, where X-Net is shown to help scientists discover new laws of mathematics or physics.

cs.AI↗

arXiv2024

vMFER: Von Mises-Fisher Experience Resampling Based on Uncertainty of Gradient Directions for Policy Improvement

Reinforcement Learning (RL) is a widely employed technique in decision-making problems, encompassing two fundamental operations -- policy evaluation and policy improvement. Enhancing learning efficiency remains a key challenge in RL, with many efforts focused on using ensemble critics to boost policy evaluation efficiency. However, when using multiple critics, the actor in the policy improvement process can obtain different gradients. Previous studies have combined these gradients without considering their disagreements. Therefore, optimizing the policy improvement process is crucial to enhance learning efficiency. This study focuses on investigating the impact of gradient disagreements caused by ensemble critics on policy improvement. We introduce the concept of uncertainty of gradient directions as a means to measure the disagreement among gradients utilized in the policy improvement process. Through measuring the disagreement among gradients, we find that transitions with lower uncertainty of gradient directions are more reliable in the policy improvement process. Building on this analysis, we propose a method called von Mises-Fisher Experience Resampling (vMFER), which optimizes the policy improvement process by resampling transitions and assigning higher confidence to transitions with lower uncertainty of gradient directions. Our experiments demonstrate that vMFER significantly outperforms the benchmark and is particularly well-suited for ensemble structures in RL.

cs.LG↗

arXiv2024

Uni-RLHF: Universal Platform and Benchmark Suite for Reinforcement Learning with Diverse Human Feedback

Reinforcement Learning with Human Feedback (RLHF) has received significant attention for performing tasks without the need for costly manual reward design by aligning human preferences. It is crucial to consider diverse human feedback types and various learning methods in different environments. However, quantifying progress in RLHF with diverse feedback is challenging due to the lack of standardized annotation platforms and widely used unified benchmarks. To bridge this gap, we introduce Uni-RLHF, a comprehensive system implementation tailored for RLHF. It aims to provide a complete workflow from real human feedback, fostering progress in the development of practical problems. Uni-RLHF contains three packages: 1) a universal multi-feedback annotation platform, 2) large-scale crowdsourced feedback datasets, and 3) modular offline RLHF baseline implementations. Uni-RLHF develops a user-friendly annotation interface tailored to various feedback types, compatible with a wide range of mainstream RL environments. We then establish a systematic pipeline of crowdsourced annotations, resulting in large-scale annotated datasets comprising more than 15 million steps across 30+ popular tasks. Through extensive experiments, the results in the collected datasets demonstrate competitive performance compared to those from well-designed manual rewards. We evaluate various design choices and offer insights into their strengths and potential areas of improvement. We wish to build valuable open-source platforms, datasets, and baselines to facilitate the development of more robust and reliable RLHF solutions based on realistic human feedback. The website is available at https://uni-rlhf.github.io/.

cs.LG↗

arXiv2024

SheetAgent: Towards A Generalist Agent for Spreadsheet Reasoning and Manipulation via Large Language Models

Spreadsheets are ubiquitous across the World Wide Web, playing a critical role in enhancing work efficiency across various domains. Large language model (LLM) has been recently attempted for automatic spreadsheet manipulation but has not yet been investigated in complicated and realistic tasks where reasoning challenges exist (e.g., long horizon manipulation with multi-step reasoning and ambiguous requirements). To bridge the gap with the real-world requirements, we introduce SheetRM, a benchmark featuring long-horizon and multi-category tasks with reasoning-dependent manipulation caused by real-life challenges. To mitigate the above challenges, we further propose SheetAgent, a novel autonomous agent that utilizes the power of LLMs. SheetAgent consists of three collaborative modules: Planner, Informer, and Retriever, achieving both advanced reasoning and accurate manipulation over spreadsheets without human interaction through iterative task reasoning and reflection. Extensive experiments demonstrate that SheetAgent delivers 20--40\% pass rate improvements on multiple benchmarks over baselines, achieving enhanced precision in spreadsheet manipulation and demonstrating superior table reasoning abilities. More details and visualizations are available at the project website: https://sheetagent.github.io/. The datasets and source code are available at https://anonymous.4open.science/r/SheetAgent.

cs.AI↗