arXiv subjects

Feifan Liu

Publications and source records attributed to Feifan Liu.

12 recordsLinked to original sources

arXiv2026

Safe Task Planning with Long-Term Graph Memory for Embodied Agents

Large language models (LLMs) and vision-language models (VLMs) have significantly advanced zero-shot task planning for embodied agents. However, most LLM- and VLM-driven methods struggle to generate safe high-level actions due to a lack of physical risk awareness, particularly under partial observability, where hazards lie outside the immediate field of view. To address this challenge, we propose a novel safe task-planning framework, SafeMem, which constructs and maintains a long-term semantic graph memory of the open and dynamic environment. Based on egocentric observations, the proposed framework incrementally accumulates knowledge about surrounding objects and their relationships with a graph. Then, an LLM-based risk predictor evaluates candidate actions using the graph memory, triggering a conservatism-modulated replanning loop with explanations for detected hazards. Extensive experiments on the IS-Bench benchmark and a real-world robot platform demonstrate that the SafeMem framework substantially improves safe success rates compared to state-of-the-art VLM-driven task planners. Video results are available on our webpage: https://sites.google.com/view/safemem.

cs.RO↗

arXiv2026

ConceptTree: Bringing Semantic Transparency to Black-Box Decision Making for Robotic Manipulation

Establishing interpretable decision-making processes in long-horizon robotic manipulation is critical for enabling reliable human oversight and intervention. However, existing approaches to robotic manipulation largely treat skill selection as opaque mappings from observations to actions, offering limited transparency into how decisions are formed. In this work, we propose ConceptTree, a framework that reframes high-level manipulation skill selection as reasoning over human-interpretable concepts, representing high-level policies as a sequence of concept-level predicates over visual observations. Rather than relying on implicit latent representations, our method learns a normalized concept space grounded in visual inputs, over which a decision tree is trained to predict high-level skills. This formulation yields a transparent decision process that is both traceable and intervenable, enabling direct inspection and modification of policy behavior. We evaluate our approach on a set of real-world robotic manipulation tasks with increasing complexity. Experimental results show that ConceptTree consistently outperforms existing concept-based baselines, particularly in complex, long-horizon scenarios. Furthermore, we provide qualitative case studies showing that our model supports fine-grained intervention by modifying individual concepts, enabling targeted correction of decision errors without retraining.

cs.RO↗

arXiv2026

Exploring Accurate and Transparent Domain Adaptation in Predictive Healthcare via Concept-Grounded Orthogonal Inference

Deep learning models for clinical event prediction on electronic health records (EHR) often suffer performance degradation when deployed under different data distributions. While domain adaptation (DA) methods can mitigate such shifts, their "black-box" nature prevents widespread adoption in clinical practice where transparency is essential for trust and safety. We propose ExtraCare to decompose patient representations into invariant and covariant components. By supervising these two components and enforcing their orthogonality during training, our model preserves label information while exposing domain-specific variation at the same time for more accurate predictions than most feature alignment models. More importantly, it offers human-understandable explanations by mapping sparse latent dimensions to medical concepts and quantifying their contributions via targeted ablations. ExtraCare is evaluated on two real-world EHR datasets across multiple domain partition settings, demonstrating superior performance along with enhanced transparency, as evidenced by its accurate predictions and explanations from extensive case studies.

cs.LG↗

arXiv2025

Efficient Context Scaling with LongCat ZigZag Attention

We introduce LongCat ZigZag Attention (LoZA), which is a sparse attention scheme designed to transform any existing full-attention models into sparse versions with rather limited compute budget. In long-context scenarios, LoZA can achieve significant speed-ups both for prefill-intensive (e.g., retrieval-augmented generation) and decode-intensive (e.g., tool-integrated reasoning) cases. Specifically, by applying LoZA to LongCat-Flash during mid-training, we serve LongCat-Flash-Exp as a long-context foundation model that can swiftly process up to 1 million tokens, enabling efficient long-term reasoning and long-horizon agentic capabilities.

cs.CL↗

arXiv2025

Optimizing Frequent Checkpointing via Low-Cost Differential for Distributed Training Systems

Distributed training of large deep-learning models often leads to failures, so checkpointing is commonly employed for recovery. State-of-the-art studies focus on frequent checkpointing for fast recovery from failures. However, frequent checkpointing generates numerous checkpoints, incurring substantial costs and thus degrading training performance. Recently, differential checkpointing has been proposed to reduce costs, but it is limited to recommendation systems, so its application to general distributed training systems remains unexplored. In this paper, we find that gradients generated during distributed training can be reused to construct differential checkpoints, while the former's size is smaller than the latter's, motivating us to reuse gradients for low-cost differential checkpointing. Based on this main idea, we propose \sysname, a frequent checkpointing framework for compression-enabled training systems that reuses compressed gradients as differential checkpoints, eliminating redundant differential computation and reducing checkpoint transmission cost. Furthermore, we extend gradient reuse to scenarios without gradient compression and propose \sysnameplus, which employs layer-wise-reuse snapshotting and incremental-merging persistence to overlap checkpointing with training execution. Experiments on diverse workloads, including billion-parameter-scale models, demonstrate that \sysname and \sysnameplus significantly reduce checkpointing overhead and enable checkpointing at frequencies as high as once per iteration, reducing training time by up to 89.2\% and 81.2\%, respectively.

cs.DC↗

arXiv2025

DHEvo: Data-Algorithm Based Heuristic Evolution for Generalizable MILP Solving

Primal heuristics play a critical role in improving the efficiency of mixed integer programming (MILP) solvers. As large language models (LLMs) have demonstrated superior code generation abilities, recent MILP works are devoted to leveraging the evolutionary computation approaches with LLMs to generate effective primal heuristics. Although the generated heuristics have achieved better solving performance than the hand-crafted ones with little adaptability, the advantage of current LLM-based methods is limited to few MILP instances in one problem class, as they fail to capture the instance characteristics in the problem class (the MILP instances generated from the same mathematical model are defined as a problem class). Since MILP instances often differ significantly in structure and feature distribution, the neglect of their characteristics in the evolution process results in poor generalization within the same problem class. To overcome this challenge, we propose a data-algorithm co-evolution framework (DHEvo) that iteratively selects representative instances and evolves corresponding heuristics. With the initial instance distribution, we develop an LLM-based multi-agent system to generate data-code pairs simultaneously. These data-code pairs are iteratively refined based on their fitness scores, leading to the identification of the most effective heuristic over the entire problem class. Extensive experiments across diverse MILP benchmarks demonstrate that our approach significantly outperforms both human-designed heuristics and existing LLM-based methods.

cs.NE↗

arXiv2025

RACE-Align: Retrieval-Augmented and Chain-of-Thought Enhanced Preference Alignment for Large Language Models

Large Language Models (LLMs) struggle with accuracy, domain-specific reasoning, and interpretability in vertical domains. Traditional preference alignment methods like Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO) often overlook the underlying knowledge sources and reasoning logic. This paper introduces RACE-Align (Retrieval-Augmented and Chain-of-Thought Enhanced Alignment), a novel framework designed to address these limitations. RACE-Align systematically constructs a binary preference dataset incorporating external knowledge support and explicit Chain-of-Thought (CoT) reasoning, then aligns LLMs using the DPO algorithm. The core innovation lies in its preference data construction strategy: it integrates AI-driven retrieval for factual grounding, enhancing knowledgeability and accuracy, and emphasizes the optimization of domain-specific CoT, treating the reasoning process itself as a key preference dimension. A multi-stage, AI-driven refinement pipeline cost-effectively generates these preference pairs. Experimental validation in Traditional Chinese Medicine (TCM) using Qwen3-1.7B as the base model demonstrates that RACE-Align significantly outperforms the original base model and a model fine-tuned only with Supervised Fine-Tuning (SFT). Improvements were observed across multiple dimensions, including answer accuracy, information richness, application of TCM thinking patterns, logicality and depth of reasoning, and interpretability. These findings suggest RACE-Align offers an effective pathway to enhance LLMs' knowledge application, reasoning reliability, and process transparency in complex vertical domains.

cs.CL↗

arXiv2024

Safe Planner: Empowering Safety Awareness in Large Pre-Trained Models for Robot Task Planning

Robot task planning is an important problem for autonomous robots in long-horizon challenging tasks. As large pre-trained models have demonstrated superior planning ability, recent research investigates utilizing large models to achieve autonomous planning for robots in diverse tasks. However, since the large models are pre-trained with Internet data and lack the knowledge of real task scenes, large models as planners may make unsafe decisions that hurt the robots and the surrounding environments. To solve this challenge, we propose a novel Safe Planner framework, which empowers safety awareness in large pre-trained models to accomplish safe and executable planning. In this framework, we develop a safety prediction module to guide the high-level large model planner, and this safety module trained in a simulator can be effectively transferred to real-world tasks. The proposed Safe Planner framework is evaluated on both simulated environments and real robots. The experiment results demonstrate that Safe Planner not only achieves state-of-the-art task success rates, but also substantially improves safety during task execution. The experiment videos are shown in https://sites.google.com/view/safeplanner .

cs.RO↗

arXiv2024

Initial Burst of Disruptive Efforts Ensuring Scientific Career Viability

Despite persistent efforts to understand the dynamics of creativity of scientists over careers in terms of productivity, impact, and prize, little is known about the dynamics of scientists' disruptive efforts that affect individual academic careers and drive scientific advance. Drawing on millions of data over six decades and across nineteen disciplines, associating the publication records of individual scientists with the disruption index, we systematically quantify the temporal pattern of disruptive ideas over individual scientific careers, providing a detailed understanding of the macro phenomenon of scientific stagnation from the individual perspective. We start by checking the relationship between disruption-based and citation-based publication profiles. Next, we observe the finite inequality in the disruptive productivity of scientists, diminishing gradually as the level of disruption increases. We then identify the initial burst phenomenon in disruption dynamics. It is further revealed that while early engagement in high disruption frictions away initial productivity, compared to initial advantage in productivity or impact, initial high disruption ensures more subsequent academic viability evidenced by a longer career span and relatively final higher productivity, but does not necessarily guarantee academic success throughout careers. Further analysis shows that increasing disruptive work is uncorrelated to overall productivity but negatively correlated with the overall impact. However, increasing disruptive work in the early career is associated with higher overall productivity, yet lower overall productivity in the later career. Our research underscores the urgent need for a policy shift that encourages a balance between the pursuit of disruptive efforts and the achievement of impactful outcomes.

physics.soc-ph↗

arXiv2023

Science as Exploration in a Knowledge Landscape: Tracing Hotspots or Seeking Opportunity?

The selection of research topics by scientists can be viewed as an exploration process conducted by individuals with cognitive limitations traversing a complex cognitive landscape influenced by both individual and social factors. While existing theoretical investigations have provided valuable insights, the intricate and multifaceted nature of modern science hinders the implementation of empirical experiments. This study leverages advancements in deep learning techniques to investigate the patterns and dynamic mechanisms of topic-transition among scientists. By constructing the knowledge space across 6 large-scale disciplines, we depict the trajectories of scientists' topic transitions within this space, measuring the flow and distance of research regions across different sub-spaces. Our findings reveal a predominantly conservative pattern of topic transition at the individual level, with scientists primarily exploring local knowledge spaces. Furthermore, simulation modeling analysis identifies research intensity, driven by the concentration of scientists within a specific region, as the key facilitator of topic transition. Conversely, the knowledge distance between fields serves as a significant barrier to exploration. Notably, despite potential opportunities for breakthrough discoveries at the intersection of subfields, empirical evidence suggests that these opportunities do not exert a strong pull on scientists, leading them to favor familiar research areas. Our study provides valuable insights into the exploration dynamics of scientific knowledge production, highlighting the influence of individual cognition, social factors, and the intrinsic structure of the knowledge landscape itself. These findings offer a framework for understanding and potentially shaping the course of scientific progress.

physics.soc-ph↗

arXiv2023

Scientists' bounded mobility on the epistemic landscape

Despite persistent efforts in revealing the temporal patterns in scientific careers, little attention has been paid to the spatial patterns of scientific activities in the knowledge space. Here, drawing on millions of papers in six disciplines, we consider scientists' publication sequence as "walks" on the quantifiable epistemic landscape constructed from large-scale bibliometric corpora by combining embedding and manifold learning algorithms, aiming to reveal the individual research topic dynamics and association between research radius with academic performance, along their careers. Intuitively, the visualization shows the localized and bounded nature of mobile trajectories. We further find that the distributions of scientists' transition radius and transition pace are both left-skewed compared with the results of controlled experiments. Then, we observe the mixed exploration and exploitation pattern and the corresponding strategic trade-off in the research transition, where scientists both deepen their previous research with frequency bias and explore new research with knowledge proximity bias. We further develop a bounded exploration-exploitation (BEE) model to reproduce the observed patterns. Moreover, the association between scientists' research radius and academic performance shows that extensive exploration will not lead to a sustained increase in academic output but a decrease in impact. In addition, we also note that disruptive findings are more derived from an extensive transition, whereas there is a saturation in this association. Our study contributes to the comprehension of the mobility patterns of scientists in the knowledge space, thereby providing significant implications for the development of scientific policy-making.

physics.soc-ph↗

arXiv2016

Learning for Biomedical Information Extraction: Methodological Review of Recent Advances

Biomedical information extraction (BioIE) is important to many applications, including clinical decision support, integrative biology, and pharmacovigilance, and therefore it has been an active research. Unlike existing reviews covering a holistic view on BioIE, this review focuses on mainly recent advances in learning based approaches, by systematically summarizing them into different aspects of methodological development. In addition, we dive into open information extraction and deep learning, two emerging and influential techniques and envision next generation of BioIE.

cs.CL↗