arXiv subjects

Jinpeng Li

Publications and source records attributed to Jinpeng Li.

At least 37 records · Page 2Linked to original sources

arXiv2026

Med-Evo: Test-time Self-evolution for Medical Multimodal Large Language Models

Medical Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities across diverse healthcare tasks. However, current post-training strategies, such as supervised fine-tuning and reinforcement learning, heavily depend on substantial annotated data while overlooking the potential of unlabeled test data for model enhancement. This limitation becomes particularly pronounced in medical domains, where acquiring extensive labeled medical data is difficult due to the strict data sensitivity and annotation complexity. Moreover, leveraging test data poses challenges in generating reliable supervision signals from unlabeled samples and maintaining stable self-evolution. To address these limitations, we propose Med-Evo, the first self-evolution framework for medical MLLMs that utilizes label-free reinforcement learning to promote model performance without requiring additional labeled data. Our framework introduces two key innovations: $1)$ Feature-driven Pseudo Labeling (FPL) that identifies semantic centroids from all heterogeneous candidate responses to select pseudo labels in each rollout, and $2)$ Hard-Soft Reward (HSR) that combines exact match with token-level assessment and semantic similarity to provide hierarchical reward. Experiments on three medical VQA benchmarks and two base MLLMs show clear advantages of our approach over SOTA methods, with significant improvements of 10.43\% accuracy and 4.68\% recall on the SLAKE dataset using Qwen2.5-VL, showing the effectiveness of our method.

cs.CV↗

arXiv2026

DualWeaver: Synergistic Feature Weaving Surrogates for Multivariate Forecasting with Univariate Time Series Foundation Models

Time-series foundation models (TSFMs) have achieved strong univariate forecasting through large-scale pre-training, yet effectively extending this success to multivariate forecasting remains challenging. To address this, we propose DualWeaver, a novel framework that adapts univariate TSFMs (Uni-TSFMs) for multivariate forecasting by using a pair of learnable, structurally symmetric surrogate series. Generated by a shared auxiliary feature-fusion module that captures cross-variable dependencies, these surrogates are mapped to TSFM-compatible series via the forecasting objective. The symmetric structure enables parameter-free reconstruction of final predictions directly from the surrogates, without additional parametric decoding. A theoretically grounded regularization term is further introduced to enhance robustness against adaptation collapse. Extensive experiments on diverse real-world datasets show that DualWeaver outperforms state-of-the-art multivariate forecasters in both accuracy and stability. We release the code at https://github.com/li-jinpeng/DualWeaver.

cs.LG↗

arXiv2026

Teaching Large Reasoning Models Effective Reflection

Large Reasoning Models (LRMs) have recently shown impressive performance on complex reasoning tasks, often by engaging in self-reflective behaviors such as self-critique and backtracking. However, not all reflections are beneficial-many are superficial, offering little to no improvement over the original answer and incurring computation overhead. In this paper, we identify and address the problem of superficial reflection in LRMs. We first propose Self-Critique Fine-Tuning (SCFT), a training framework that enhances the model's reflective reasoning ability using only self-generated critiques. SCFT prompts models to critique their own outputs, filters high-quality critiques through rejection sampling, and fine-tunes the model using a critique-based objective. Building on this strong foundation, we further introduce Reinforcement Learning with Effective Reflection Rewards (RLERR). RLERR leverages the high-quality reflections initialized by SCFT to construct reward signals, guiding the model to internalize the self-correction process via reinforcement learning. Experiments on two challenging benchmarks, AIME2024 and AIME2025, show that SCFT and RLERR significantly improve both reasoning accuracy and reflection quality, outperforming state-of-the-art baselines. All data and codes are available at https://github.com/wanghanbinpanda/SCFT.

cs.AI↗

arXiv2026

Group Pattern Selection Optimization: Let LRMs Pick the Right Pattern for Reasoning

Large reasoning models (LRMs) exhibit diverse high-level reasoning patterns (e.g., direct solution, reflection-and-verification, and exploring multiple solutions), yet prevailing training recipes implicitly bias models toward a limited set of dominant patterns. Through a systematic analysis, we identify substantial accuracy variance across these patterns on mathematics and science benchmarks, revealing that a model's default reasoning pattern is often sub-optimal for a given problem. To address this, we introduce Group Pattern Selection Optimization (GPSO), a reinforcement learning framework that extends GRPO by incorporating multi-pattern rollouts, verifier-guided optimal pattern selection per problem, and attention masking during optimization to prevent the leakage of explicit pattern suffixes into the learned policy. By exploring a portfolio of diverse reasoning strategies and optimizing the policy on the most effective ones, GPSO enables the model to internalize the mapping from problem characteristics to optimal reasoning patterns. Extensive experiments demonstrate that GPSO delivers consistent and substantial performance gains across various model backbones and benchmarks, effectively mitigating pattern sub-optimality and fostering more robust, adaptable reasoning. All data and codes are available at https://github.com/wanghanbinpanda/GPSO.

cs.AI↗

arXiv2026

First Submillimeter Lights from Dome A: Tracing the Carbon Cycle in the Feedback of Massive Stars

The cycling of carbon between its ionized, atomic, and molecular phases shapes the chemical compositions and physical conditions of the interstellar medium (ISM). However, ground-based studies of the full carbon cycle have been limited by atmospheric absorption. Dome~A, the most promising site for submillimeter astronomy, has long resisted successful submillimeter astronomical observations. Using the 60~cm Antarctic Terahertz Explorer, we present the first successful CO ($4-3$) and [CI] ($^3P_1 - ^3P_0$) mapping observations of two archetypal triggered massive star-formation regions at Dome~A. These data, together with archival [CII], provide the first complete characterization of all three carbon phases in these environments. We find elevated C$^{0}$/CO abundance ratios in high-extinction regions, plausibly driven by deep penetration of intense radiation fields from massive stars into a clumpy ISM. These findings mark a major milestone for submillimeter astronomy at Dome~A and offer valuable insights into the impact of massive star feedback on the surrounding ISM.

astro-ph.GA↗

arXiv2026

Boosting In-Silicon Directed Evolution with Fine-Tuned Protein Language Model and Tree Search

Protein evolution through amino acid mutations is a cornerstone of life sciences. Recent advances in protein language models have shown rich evolutionary patterns, offering unprecedented potential for in-silicon directed evolution. However, existing directed evolution methods largely rely on heuristic evolution strategies and have yet to efficiently integrate the transformative protein language models with advanced optimization techniques, such as reinforcement learning, to adaptively learn superior evolution policies. To bridge this gap, we propose AlphaDE, a novel framework that evolves protein sequences by harnessing the innovative paradigms of large language models, such as fine-tuning and test-time inference. First, AlphaDE fine-tunes pretrained protein language models using masked language modeling on homologous protein sequences to activate the evolutionary plausibility of the interested protein family. Second, AlphaDE introduces test-time inference based on Monte Carlo tree search, which effectively evolves proteins with evolutionary guidance from the fine-tuned protein language model. Extensive benchmark experiments show that AlphaDE remarkably outperforms previous state-of-the-art methods even with few-shot fine-tuning. A case study further demonstrates that AlphaDE supports condensing the protein sequence space of avGFP through computational evolution.

cs.AI↗

arXiv2025

Perceive and Calibrate: Analyzing and Enhancing Robustness of Medical Multi-Modal Large Language Models

Medical Multi-modal Large Language Models (MLLMs) have shown promising clinical performance. However, their sensitivity to real-world input perturbations, such as imaging artifacts and textual errors, critically undermines their clinical applicability. Systematic analysis of such noise impact on medical MLLMs remains largely unexplored. Furthermore, while several works have investigated the MLLMs' robustness in general domains, they primarily focus on text modality and rely on costly fine-tuning. They are inadequate to address the complex noise patterns and fulfill the strict safety standards in medicine. To bridge this gap, this work systematically analyzes the impact of various perturbations on medical MLLMs across both visual and textual modalities. Building on our findings, we introduce a training-free Inherent-enhanced Multi-modal Calibration (IMC) framework that leverages MLLMs' inherent denoising capabilities following the perceive-and-calibrate principle for cross-modal robustness enhancement. For the visual modality, we propose a Perturbation-aware Denoising Calibration (PDC) which leverages MLLMs' own vision encoder to identify noise patterns and perform prototype-guided feature calibration. For text denoising, we design a Self-instantiated Multi-agent System (SMS) that exploits the MLLMs' self-assessment capabilities to refine noisy text through a cooperative hierarchy of agents. We construct a benchmark containing 11 types of noise across both image and text modalities on 2 datasets. Experimental results demonstrate our method achieves the state-of-the-art performance across multiple modalities, showing potential to enhance MLLMs' robustness in real clinical scenarios.

cs.CV↗

arXiv2025

Short-wave infrared broadband up-conversion imaging by using a noncritical phase matched bulk KTiOPO$_4$ crystal

Compared to cryogenically cooled conventional detectors, up-conversion detection enables efficient room-temperature short-wave infrared (SWIR) imaging. Although quasi-phase-matching (QPM) in periodically poled crystals offers advantages, the small crystal aperture (typically 1 mm$\times$3 mm) limits resolution. Non-poled crystals enable larger apertures but suffer walk-off aberrations. This work overcomes these limitations by using a noncritical phase matched (NCPM) KTiOPO$_4$ crystal (6 mm$\times$7 mm aperture, 0.5 mm length). Results show resolutions 6$\times$ and 2$\times$ higher than periodically poled crystals in orthogonal directions, with broad conversion band (1.3-2.2 $μ$m) covering biological and atmospheric windows. The absence of walk-off ensures better image fidelity in up-conversion process. This study presents the first comprehensive characterization of NCPM-based broadband up-conversion imaging, demonstrating performance at the theoretical resolution limit while circumventing drawbacks inherent in alternative up-conversion schemes and conventional detectors.

physics.optics↗

arXiv2025

MEJO: MLLM-Engaged Surgical Triplet Recognition via Inter- and Intra-Task Joint Optimization

Surgical triplet recognition, which involves identifying instrument, verb, target, and their combinations, is a complex surgical scene understanding challenge plagued by long-tailed data distribution. The mainstream multi-task learning paradigm benefiting from cross-task collaborative promotion has shown promising performance in identifying triples, but two key challenges remain: 1) inter-task optimization conflicts caused by entangling task-generic and task-specific representations; 2) intra-task optimization conflicts due to class-imbalanced training data. To overcome these difficulties, we propose the MLLM-Engaged Joint Optimization (MEJO) framework that empowers both inter- and intra-task optimization for surgical triplet recognition. For inter-task optimization, we introduce the Shared-Specific-Disentangled (S$^2$D) learning scheme that decomposes representations into task-shared and task-specific components. To enhance task-shared representations, we construct a Multimodal Large Language Model (MLLM) powered probabilistic prompt pool to dynamically augment visual features with expert-level semantic cues. Additionally, comprehensive task-specific cues are modeled via distinct task prompts covering the temporal-spatial dimensions, effectively mitigating inter-task ambiguities. To tackle intra-task optimization conflicts, we develop a Coordinated Gradient Learning (CGL) strategy, which dissects and rebalances the positive-negative gradients originating from head and tail classes for more coordinated learning behaviors. Extensive experiments on the CholecT45 and CholecT50 datasets demonstrate the superiority of our proposed framework, validating its effectiveness in handling optimization conflicts.

cs.CV↗

arXiv2025

SurgPub-Video: A Comprehensive Surgical Video Dataset for Enhanced Surgical Intelligence in Vision-Language Model

Vision-Language Models (VLMs) have shown significant potential in surgical scene analysis, yet existing models are limited by frame-level datasets and lack high-quality video data with procedural surgical knowledge. To address these challenges, we make the following contributions: (i) SurgPub-Video, a comprehensive dataset of over 3,000 surgical videos and 25 million annotated frames across 11 specialties, sourced from peer-reviewed clinical journals, (ii) SurgLLaVA-Video, a specialized VLM for surgical video understanding, built upon the TinyLLaVA-Video architecture that supports both video-level and frame-level inputs, and (iii) a video-level surgical Visual Question Answering (VQA) benchmark, covering diverse 11 surgical specialities, such as vascular, cardiology, and thoracic. Extensive experiments, conducted on the proposed benchmark and three additional surgical downstream tasks (action recognition, skill assessment, and triplet recognition), show that SurgLLaVA-Video significantly outperforms both general-purpose and surgical-specific VLMs with only three billion parameters. The dataset, model, and benchmark will be released to enable further advancements in surgical video understanding.

q-bio.OT↗

arXiv2025

From Learning to Unlearning: Biomedical Security Protection in Multimodal Large Language Models

The security of biomedical Multimodal Large Language Models (MLLMs) has attracted increasing attention. However, training samples easily contain private information and incorrect knowledge that are difficult to detect, potentially leading to privacy leakage or erroneous outputs after deployment. An intuitive idea is to reprocess the training set to remove unwanted content and retrain the model from scratch. Yet, this is impractical due to significant computational costs, especially for large language models. Machine unlearning has emerged as a solution to this problem, which avoids complete retraining by selectively removing undesired knowledge derived from harmful samples while preserving required capabilities on normal cases. However, there exist no available datasets to evaluate the unlearning quality for security protection in biomedical MLLMs. To bridge this gap, we propose the first benchmark Multimodal Large Language Model Unlearning for BioMedicine (MLLMU-Med) built upon our novel data generation pipeline that effectively integrates synthetic private data and factual errors into the training set. Our benchmark targets two key scenarios: 1) Privacy protection, where patient private information is mistakenly included in the training set, causing models to unintentionally respond with private data during inference; and 2) Incorrectness removal, where wrong knowledge derived from unreliable sources is embedded into the dataset, leading to unsafe model responses. Moreover, we propose a novel Unlearning Efficiency Score that directly reflects the overall unlearning performance across different subsets. We evaluate five unlearning approaches on MLLMU-Med and find that these methods show limited effectiveness in removing harmful knowledge from biomedical MLLMs, indicating significant room for improvement. This work establishes a new pathway for further research in this promising field.

cs.CV↗

arXiv2025

StepHint: Multi-level Stepwise Hints Enhance Reinforcement Learning to Reason

Reinforcement learning with verifiable rewards (RLVR) is a promising approach for improving the complex reasoning abilities of large language models (LLMs). However, current RLVR methods face two significant challenges: the near-miss reward problem, where a small mistake can invalidate an otherwise correct reasoning process, greatly hindering training efficiency; and exploration stagnation, where models tend to focus on solutions within their ``comfort zone,'' lacking the motivation to explore potentially more effective alternatives. To address these challenges, we propose StepHint, a novel RLVR algorithm that utilizes multi-level stepwise hints to help models explore the solution space more effectively. StepHint generates valid reasoning chains from stronger models and partitions these chains into reasoning steps using our proposed adaptive partitioning method. The initial few steps are used as hints, and simultaneously, multiple-level hints (each comprising a different number of steps) are provided to the model. This approach directs the model's exploration toward a promising solution subspace while preserving its flexibility for independent exploration. By providing hints, StepHint mitigates the near-miss reward problem, thereby improving training efficiency. Additionally, the external reasoning pathways help the model develop better reasoning abilities, enabling it to move beyond its ``comfort zone'' and mitigate exploration stagnation. StepHint outperforms competitive RLVR enhancement methods across six mathematical benchmarks, while also demonstrating superior generalization and excelling over baselines on out-of-domain benchmarks.

cs.AI↗

arXiv2025

xbench: Tracking Agents Productivity Scaling with Profession-Aligned Real-World Evaluations

We introduce xbench, a dynamic, profession-aligned evaluation suite designed to bridge the gap between AI agent capabilities and real-world productivity. While existing benchmarks often focus on isolated technical skills, they may not accurately reflect the economic value agents deliver in professional settings. To address this, xbench targets commercially significant domains with evaluation tasks defined by industry professionals. Our framework creates metrics that strongly correlate with productivity value, enables prediction of Technology-Market Fit (TMF), and facilitates tracking of product capabilities over time. As our initial implementations, we present two benchmarks: Recruitment and Marketing. For Recruitment, we collect 50 tasks from real-world headhunting business scenarios to evaluate agents' abilities in company mapping, information retrieval, and talent sourcing. For Marketing, we assess agents' ability to match influencers with advertiser needs, evaluating their performance across 50 advertiser requirements using a curated pool of 836 candidate influencers. We present initial evaluation results for leading contemporary agents, establishing a baseline for these professional domains. Our continuously updated evalsets and evaluations are available at https://xbench.org.

cs.LG↗

arXiv2025

Human-like object concept representations emerge naturally in multimodal large language models

Understanding how humans conceptualize and categorize natural objects offers critical insights into perception and cognition. With the advent of Large Language Models (LLMs), a key question arises: can these models develop human-like object representations from linguistic and multimodal data? In this study, we combined behavioral and neuroimaging analyses to explore the relationship between object concept representations in LLMs and human cognition. We collected 4.7 million triplet judgments from LLMs and Multimodal LLMs (MLLMs) to derive low-dimensional embeddings that capture the similarity structure of 1,854 natural objects. The resulting 66-dimensional embeddings were stable, predictive, and exhibited semantic clustering similar to human mental representations. Remarkably, the dimensions underlying these embeddings were interpretable, suggesting that LLMs and MLLMs develop human-like conceptual representations of objects. Further analysis showed strong alignment between model embeddings and neural activity patterns in brain regions such as EBA, PPA, RSC, and FFA. This provides compelling evidence that the object representations in LLMs, while not identical to human ones, share fundamental similarities that reflect key aspects of human conceptual knowledge. Our findings advance the understanding of machine intelligence and inform the development of more human-like artificial cognitive systems.

cs.AI↗

arXiv2025

KDRL: Post-Training Reasoning LLMs via Unified Knowledge Distillation and Reinforcement Learning

Recent advances in large language model (LLM) post-training have leveraged two distinct paradigms to enhance reasoning capabilities: reinforcement learning (RL) and knowledge distillation (KD). While RL enables the emergence of complex reasoning behaviors, it often suffers from low sample efficiency when the initial policy struggles to explore high-reward trajectories. Conversely, KD improves learning efficiency via mimicking the teacher model but tends to generalize poorly to out-of-domain scenarios. In this work, we present \textbf{KDRL}, a \textit{unified post-training framework} that jointly optimizes a reasoning model through teacher supervision (KD) and self-exploration (RL). Specifically, KDRL leverages policy gradient optimization to simultaneously minimize the reverse Kullback-Leibler divergence (RKL) between the student and teacher distributions while maximizing the expected rule-based rewards. We first formulate a unified objective that integrates GRPO and KD, and systematically explore how different KL approximations, KL coefficients, and reward-guided KD strategies affect the overall post-training dynamics and performance. Empirical results on multiple reasoning benchmarks demonstrate that KDRL outperforms GRPO and various KD baselines while achieving a favorable balance between performance and reasoning token efficiency. These findings indicate that integrating KD and RL serves as an effective and efficient strategy to train reasoning LLMs.

cs.LG↗

arXiv2025

Pangu Embedded: An Efficient Dual-system LLM Reasoner with Metacognition

This work presents Pangu Embedded, an efficient Large Language Model (LLM) reasoner developed on Ascend Neural Processing Units (NPUs), featuring flexible fast and slow thinking capabilities. Pangu Embedded addresses the significant computational costs and inference latency challenges prevalent in existing reasoning-optimized LLMs. We propose a two-stage training framework for its construction. In Stage 1, the model is finetuned via an iterative distillation process, incorporating inter-iteration model merging to effectively aggregate complementary knowledge. This is followed by reinforcement learning on Ascend clusters, optimized by a latency-tolerant scheduler that combines stale synchronous parallelism with prioritized data queues. The RL process is guided by a Multi-source Adaptive Reward System (MARS), which generates dynamic, task-specific reward signals using deterministic metrics and lightweight LLM evaluators for mathematics, coding, and general problem-solving tasks. Stage 2 introduces a dual-system framework, endowing Pangu Embedded with a "fast" mode for routine queries and a deeper "slow" mode for complex inference. This framework offers both manual mode switching for user control and an automatic, complexity-aware mode selection mechanism that dynamically allocates computational resources to balance latency and reasoning depth. Experimental results on benchmarks including AIME 2024, GPQA, and LiveCodeBench demonstrate that Pangu Embedded with 7B parameters, outperforms similar-size models like Qwen3-8B and GLM4-9B. It delivers rapid responses and state-of-the-art reasoning quality within a single, unified model architecture, highlighting a promising direction for developing powerful yet practically deployable LLM reasoners.

cs.CL↗

arXiv2025

Pangu Pro MoE: Mixture of Grouped Experts for Efficient Sparsity

The surgence of Mixture of Experts (MoE) in Large Language Models promises a small price of execution cost for a much larger model parameter count and learning capacity, because only a small fraction of parameters are activated for each input token. However, it is commonly observed that some experts are activated far more often than others, leading to system inefficiency when running the experts on different devices in parallel. Therefore, we introduce Mixture of Grouped Experts (MoGE), which groups the experts during selection and balances the expert workload better than MoE in nature. It constrains tokens to activate an equal number of experts within each predefined expert group. When a model execution is distributed on multiple devices, this architectural design ensures a balanced computational load across devices, significantly enhancing throughput, particularly for the inference phase. Further, we build Pangu Pro MoE on Ascend NPUs, a sparse model based on MoGE with 72 billion total parameters, 16 billion of which are activated for each token. The configuration of Pangu Pro MoE is optimized for Ascend 300I Duo and 800I A2 through extensive system simulation studies. Our experiments indicate that MoGE indeed leads to better expert load balancing and more efficient execution for both model training and inference on Ascend NPUs. The inference performance of Pangu Pro MoE achieves 1148 tokens/s per card and can be further improved to 1528 tokens/s per card by speculative acceleration, outperforming comparable 32B and 72B Dense models. Furthermore, we achieve an excellent cost-to-performance ratio for model inference on Ascend 300I Duo. Our studies show that Ascend NPUs are capable of training Pangu Pro MoE with massive parallelization to make it a leading model within the sub-100B total parameter class, outperforming prominent open-source models like GLM-Z1-32B and Qwen3-32B.

cs.CL↗

arXiv2025

GigaSpeech 2: An Evolving, Large-Scale and Multi-domain ASR Corpus for Low-Resource Languages with Automated Crawling, Transcription and Refinement

The evolution of speech technology has been spurred by the rapid increase in dataset sizes. Traditional speech models generally depend on a large amount of labeled training data, which is scarce for low-resource languages. This paper presents GigaSpeech 2, a large-scale, multi-domain, multilingual speech recognition corpus. It is designed for low-resource languages and does not rely on paired speech and text data. GigaSpeech 2 comprises about 30,000 hours of automatically transcribed speech, including Thai, Indonesian, and Vietnamese, gathered from unlabeled YouTube videos. We also introduce an automated pipeline for data crawling, transcription, and label refinement. Specifically, this pipeline involves Whisper for initial transcription, MMS for forced alignment, and multi-dimensional filtering for data quality assurance. A modified Noisy Student Training is developed to further refine flawed pseudo labels iteratively, thereby enhancing model performance. Experimental results on our manually transcribed evaluation set and two public test sets from Common Voice and FLEURS confirm our corpus's high quality and broad applicability. Notably, ASR models trained on GigaSpeech 2 can reduce the word error rate for Thai, Indonesian, and Vietnamese on our challenging and realistic YouTube test set by 25% to 40% compared to Whisper large-v3, with merely 10% model parameters. Furthermore, our ASR models trained on GigaSpeech 2 yield superior performance compared to commercial services. We hope that our newly introduced corpus and pipeline will open a new avenue for low-resource speech recognition and significantly facilitate research in this area.

eess.AS↗