arXiv subjects

Feng Lin

Publications and source records attributed to Feng Lin.

At least 19 recordsLinked to original sources

arXiv2026

Tool and Trainer? Experimental Evidence on Human Learning from Predictive AI in Cancer Diagnosis

Artificial intelligence (AI) is increasingly deployed to aid professional judgment, yet its implications for the development of human expertise remain unclear. We examine how the availability of AI assistance during training and task execution shapes novices' individual and collective judgment in the context of cancer diagnosis. Across two field experiments at a large teaching hospital, 576 medical novices learned to diagnose lung cancer from CT scans. Study 1 used a 2 * 2 factorial design that randomized AI assistance during training, task execution, both, or neither. Novices who received AI assistance during training achieved higher diagnostic accuracy in both AI-assisted and unaided task execution, with a larger improvement in the former. However, training-phase assistance was associated with more common errors than task-execution-phase assistance, suggesting that while the trainer effect improves individual accuracy, it limits the benefits of aggregating judgments. Study 2 shows that AI assistance during earlier task execution improved subsequent unaided performance and reduced common errors in the unaided phase of task execution, thus revealing a surprisingly beneficial "tool-as-trainer" effect. These findings suggest that, contingent on the right deployment strategy, AI assistance can support individual expertise development and improve collective judgment by reducing common errors across individuals.

cs.HC↗

arXiv2026

LocalAgent: Collaborative Agentic Verification for Fine-Grained Instance-Level Consistency

Multimodal large language models (MLLMs) have made rapid progress in general visual understanding, yet they still struggle with fine-grained instance-level consistency verification, where the goal is to determine whether two images depict the exact same physical object under varying viewpoints, illumination, and backgrounds. This task requires both global semantic perception and precise examination of subtle local evidence, since semantically similar objects may differ only in small suspicious regions. To address this challenge, we propose LocalAgent, a collaborative agentic framework that integrates the complementary strengths of a generalist MLLM and lightweight specialist verifiers. The MLLM first reasons globally to identify suspicious local regions, which are then delegated to a Local Evidence Verifier (LEV) toolkit for fine-grained appearance and geometric matching. LEV comprises Local Appearance Consistency (LAC), which estimates appearance similarity under illumination and color changes, and Local Geometric Correspondence (LGC), which estimates spatial correspondence under viewpoint changes. LEV returns structured scalar evidence that the MLLM integrates for the final decision. We further introduce an auxiliary-capped reinforcement fine-tuning (RFT) strategy, in which answer correctness dominates auxiliary tool-use and format signals, enabling the policy to learn when verification is necessary. Experiments show that LocalAgent improves mean accuracy by 4.9% over the strongest CoT-RFT baseline while invoking a verifier on only 12.13% of samples and preserving the general multimodal capabilities of the underlying MLLM. The open-source repository is available at https://github.com/Firefly-zoo/LocalAgent-main.

cs.CV↗

arXiv2026

SPARE: Sparsity Allocation via Routing Entropy for Efficient MoEs

Mixture-of-Experts (MoE) architectures scale model capacity efficiently, yet their substantial memory footprint poses a major barrier to deployment. Expert pruning reduces this footprint, but most methods focus on which experts to remove and prune every layer equally. We propose SPARE (Sparsity Allocation via Routing Entropy), a training-free, selector-agnostic method that decides how many experts to remove from each layer using routing entropy, the entropy of tokens' normalized Top-K gate weights on calibration data. When a token's gate weight is spread evenly, removing one active expert discards little of it, so SPARE prunes layers with more diffuse routing more aggressively. A perturbation bound under random pruning formalizes this intuition, and controlled interventions with a fixed selector link concentrated routing to larger output changes. With the same selector and budget, SPARE improves average accuracy over uniform allocation on Mixtral, Qwen3, and DeepSeek models, especially at high pruning ratios. At 50% expert pruning, it raises the eight-task average of Qwen3-30B-A3B from 57.2% to 63.8%, and the math, code, and science average of DeepSeek-R1 from 80.1 to 82.8.

cs.LG↗

arXiv2026

From Scientific Crop Imagery to Agronomic Reasoning: A Multimodal LLM Benchmark for Plant Phenotyping

High-throughput, effective and comprehensive phenotyping is a critical prerequisite for crop improvement. While phenotypic analysis was traditionally performed manually, recent advances in multimodal foundation models, especially in vision-language models, have offered new opportunities for automated plant phenotyping. However, plant science remains particularly challenging for general-purpose VLMs because it requires domain-specific knowledge, fine-grained visual interpretation, and complex agronomic reasoning. We introduce PlantXpert, an evidence-grounded multimodal reasoning benchmark for plant phenotyping across four major crops: soybean, cotton, rice, and corn. Our benchmark provides a structured and reproducible framework for agronomic adaptation of VLMs, and enables controlled comparison between base models and their domain-adapted counterparts. We constructed a dataset comprising 545 digital images and 4,351 benchmark samples spanning key agronomic domains including disease, pest control, weed management and yield. The benchmark can assess diverse capabilities including visual expertise, quantitative reasoning, and multi-step agronomic reasoning. We trained and evaluated a total of 11 state-of-the-art VLMs. Fine-tuning on our dataset improves overall accuracy for all models, with many achieving around 70% accuracy, and improvements on 9 of 11 remaining statistically significant after Holm correction. We also identify diminishing returns from model scaling, uneven performance across crop subsets, and persistent weaknesses in quantitative reasoning and specialized image analysis. These limitations show that general multimodal competence alone is insufficient for reliable agronomic reasoning and demonstrate the value of domain-specific adaptation. We hope PlantXpert can serve as a foundation for advancing evidence-grounded multimodal reasoning in plant science.

cs.CV↗

arXiv2026

From Agents to Edges: Paired Optimization for Multi-Agent Collaboration

Strong individual agents do not guarantee effective collaboration. Communication must align what one agent conveys with how another interprets and uses it. This dependency makes the communication edge a natural unit of optimization. We introduce Paired Automated Communication Optimization (PACO), which jointly optimizes the instructions at both ends of each edge and their shared communication convention in systems built on black-box language models. PACO keeps model weights, agent roles, and workflow topology fixed. A state-aware discounted bandit selects edges for refinement, while execution feedback guides paired revisions evaluated through staged end-to-end execution. The optimized instructions integrate into existing agent prompts without additional model calls at deployment. On a four-agent workflow evaluated across diverse multi-agent benchmarks, PACO consistently outperforms leading prompt-optimization and communication baselines. Configurations learned on HotpotQA retain their advantage in cross-dataset transfer without target-specific optimization. These results demonstrate that optimizing how agents communicate can substantially improve collaboration within an otherwise fixed system.

cs.CL↗

arXiv2026

A Star's Death by a Thousand Cuts: The Runaway Periodic Eruptions of AT2023uqm

Stars on bound orbits around a supermassive black hole may undergo repeated partial tidal disruption events (rpTDEs), producing periodic flares. While several candidates have been suggested, definitive confirmation of these events remains elusive. We report the discovery of AT2023uqm, a nuclear transient that has exhibited at least five periodic optical flares, making it only the second confirmed case of periodicity after ASASSN-14ko. Uniquely, the flares from AT2023uqm show a nearly exponential increase in energy--a "runaway" phenomenon signaling the star's progressive destruction. This behavior is consistent with rpTDEs of low-mass, main-sequence stars or evolved giant stars. Multiwavelength observations and spectroscopic analysis of the two most recent flares reinforce its interpretation as an rpTDE. Intriguingly, each flare displays a similar double-peaked structure, potentially originating from a double-peaked mass fallback rate or two discrete collisions per orbit. The extreme ratio of peak separation to orbital period draws attention to the possibility of a giant star being disrupted, which could be distinguished from a low-mass main-sequence star by its future mass-loss evolution. Our analysis demonstrates the power of rpTDEs to probe the properties of disrupted stars and the physical processes of tidal disruption, though it is currently limited by our knowledge of these events. AT2023uqm emerges as the most compelling rpTDE thus far, serving as a crucial framework for modeling and understanding these phenomena.

astro-ph.HE↗

arXiv2026

CA-Observability of Discrete Event Systems under Cyber Attacks

In a previous paper, we investigate the control problem of discrete event systems under cyber attacks, where CA-controllability and CA-observability are proposed. We prove that a discrete event system can be controlled to generate a specification language K if and only if K is CA-controllable and CA-observable. While CA controllability can be "converted" to (conventional) controllability, CA-observability cannot be "converted" to (conventional) observability. In this paper, we further investigate CA-observability. We develop a method and an algorithm to check CA-observability. The method involves constructing an augmented automaton whose states are pairs of the current state and state estimate of the supervised system. We prove that K is CA-observable if and only if the language generated by the augmented automaton is equal to K. An algorithm is then developed to check CA-observability. We also investigate the properties of CA-observability. We show that CA-observability is not closed under (set) in tersection. Note that (conventional) observability is closed under intersection. We further show that CA-observability is not closed under union. This is similar to observability. If K is not CA-controllable and CA-observable, we pro pose a method to calculate a CA-controllable and CA observable sublanguage. We first calculate the supremal CA-controllable sublanguage for K which can be described by a sub-automaton of the automaton for K. We then propose a state-estimate-based supervisor based on the subautomaton, which ensures the supervised system stays within K.

eess.SY↗

arXiv2026

Privacy-Preserving Prompted Policy Search for Robotic Control

Large language models (LLMs) have recently demonstrated promising capabilities as in-context policy optimizers for Reinforcement Learning (RL), enabling policy search driven by both numerical reward signals and natural language reasoning. However, deploying such methods in practice requires transmitting raw policy parameters and rewards history to cloud-based LLM APIs, exposing proprietary control strategies to third-party service providers. To address this issue, this paper introduces Privacy-Preserving Prompted Policy Search (PP-ProPS), a framework that enables LLM-guided policy optimization while keeping policy and environmental parameters confidential. PP-ProPS encodes policy parameters and reward values using secret client-side transformations before they are included in each API request, ensuring that the LLM provider observes only encoded policy parameters and scaled reward information. Furthermore, unlike Vanilla ProPS, the proposed framework does not require the true optimal episodic return to be known or disclosed to the LLM. Beyond protecting the optimization data, PP-ProPS improves the search process in two ways. First, it provides the LLM with individual reward components instead of only a single total return, offering more informative feedback about each candidate policy. Second, it uses a bounded history that prevents the prompt from growing indefinitely, improving search with high-dimensional policies and supporting the use of open-weight LLMs. The proposed PP-ProPS is evaluated on both continuous and discrete control problems spanning Multi-Joint dynamics with Contact (MuJoCo) locomotion, classic control, highway driving, and robotic arm manipulation. Compared to Vanilla ProPS, the proposed PP-ProPS outperforms ProPS in seven of the ten evaluated tasks, and surpasses conventional RL methods including PPO, SAC, and TRPO, in five of the six tasks.

cs.RO↗

arXiv2026

Headroom-Aware Stochastic Adaptive Model Predictive Control for Load Frequency Control in Microgrids

As the penetration of inverter-based resources (IBRs) increases in microgrids, they are increasingly expected to play a greater role in load frequency control (LFC). Model predictive control (MPC) is attractive for LFC because it incorporates system dynamics and operational constraints. However, most MPC-based LFC formulations rely on fixed reserve headroom based on forecasted renewable availability or storage systems. Under short-term renewable intermittency, IBR headroom is stochastic and time-varying, causing optimal control commands to exceed the physically deliverable regulation capability and cause stochastic saturation. This control-actuator mismatch degrades LFC performance. Accordingly, this study develops two headroom-aware strategies for PV-dominated microgrids. First, stochastic headroom constrained MPC (SHCMPC) incorporates headroom predictions through time-varying input constraints to enforce control feasibility. Second, stochastic adaptive MPC (SAMPC) embeds headroom awareness into the MPC objective function by adaptively penalizing control actions based on predicted headroom, reducing reliance on units with limited headroom without hard time-varying constraints. Simulation results show that stochastic saturation degrades conventional MPC-based LFC, particularly under tight reserve margins. Both strategies improve regulation performance. SHCMPC eliminates saturation events, while SAMPC achieves substantial saturation mitigation with lower computational effort, offering a computationally efficient alternative for real-time LFC under stochastic renewable availability.

eess.SY↗

arXiv2026

Contingency Detection Integrated Model Predictive Control for Resilient Load Frequency Control

Contingencies can alter power-system dynamics and introduce prediction mismatch in model predictive control (MPC)-based load frequency control (LFC). Although such events may be detected or cleared by protection systems, the corresponding post-contingency dynamic model may not be available to the MPC controller on the LFC time scale. This paper proposes a contingency detection-integrated MPC (CDI-MPC) framework that combines disturbance-aware contingency detection with predictive frequency regulation. Contingencies are modeled as stochastic discrete events of a stochastic hybrid system (SHS), and a disturbance-aware residual formulation is developed to jointly identify the active mode and estimate unknown disturbances. The detected mode is then used to update the MPC prediction model, reducing contingency-induced prediction mismatch under changing operating conditions. Simulation results demonstrate accurate contingency detection and substantial improvements in closed-loop LFC performance under multiple contingency scenarios and unknown disturbances.

eess.SY↗

arXiv2026

MindForge: Teaching Small Language Models Whole-Life-Cycle Software Engineering via Source-Free Program Synthesis

Coding agents have made substantial progress on software engineering tasks that modify existing codebases, including bug fixing and feature implementation. However, constructing a complete program from scratch remains a major challenge: even the frontier models evaluated on ProgramBench fully resolve fewer than 1% of tasks. One obstacle is the lack of scalable training environments for this from-scratch setting, spanning the whole software engineering life cycle, as existing environment-construction frameworks focus only on a single phase in software development. To address this gap, we introduce MindForge, an automated pipeline that converts open-source command-line programs into source-free environments that expose only a compiled reference executable and its documentation. Using MindForge, we construct training environments from repositories disjoint from those in ProgramBench, and curate a high-quality data recipe consisting of program synthesis trajectories using GLM-5.2 as the teacher agent. Fine-tuning Qwen3.6-27B on these trajectories increases its ProgramBench average test pass rate from 37.98% to 49.51%, achieving performance comparable to substantially larger frontier models. Moreover, the fine-tuned model consistently improves over the base model across all seven unseen software engineering benchmarks, spanning long-horizon repository generation and translation, bug fixing, feature implementation, and cross-language issue resolution, with absolute gains of 31.00 points on RepoZero-C2Rust, 14.16 on DeepSWE, 10.70/4.56 on NL2Repo-Bench (with/without tests), 5.04 on SWE-bench Verified, 5.93 on SWE-bench Pro, 5.22 on SWE-bench Multilingual, and 4.94 on FeatBench.

cs.SE↗

arXiv2026

SpecFirst: Behavioral Specification Elicitation as a First-Class Step in Agent-Based Program Synthesis from Scratch

LLM-based agents excel at software engineering tasks where an existing codebase provides context, but constructing a program from scratch remains fundamentally harder. Recent benchmarks such as ProgramBench quantify this gap: given only natural-language documentation and an execute-only binary as a behavioral oracle, even frontier models solve fewer than 1% of instances. Existing frameworks conflate documentation reading, behavioral exploration, and code synthesis into a single pass, causing agents to probe insufficiently, lose behavioral intent as context drifts, and propagate early misinterpretations into the final implementation. Inspired by classical requirements engineering, we argue that behavioral specification elicitation should be a first-class phase that precedes implementation. We present SpecFirst, a two-stage framework that forces the specification elicitation before code synthesis. A dedicated spec agent first probes the binary and combines observations with documentation into a structured specification. Next, a code synthesis agent then uses this specification to drive implementation. This decomposition resolves documentation ambiguities before coding begins and provides a stable behavioral reference throughout synthesis. We evaluate SpecFirst on all 200 ProgramBench instances across four models spanning two families and an order of magnitude of capability. SpecFirst consistently outperforms the single-loop baseline, improving test pass rates by 6.9%-21.3% and binary exploration coverage by 9.4%-18.5%, all statistically significant. Behavioral analysis on code synthesis further shows that a prior specification enables earlier and more sustained code construction. Our results demonstrate that an explicit requirements-engineering phase is an effective paradigm for from-scratch program construction.

cs.SE↗

arXiv2026

Deterministic and Nonblocking Supervisory Control of Discrete Event Systems under Cyber Attacks

We investigate deterministic and nonblocking supervisory control of discrete event systems under cyber-attacks using the ALTER (Attack Language for Transition-basEd Replacement) model. While prior works consider supervisory control that achieves either the large (upper bound) language or small (lower bound) language separately, deterministic supervisory control achieves both large language and small language at the same time to ensure that the language generated by the supervised system is unique and deterministic. We introduce two new concepts of CA-D-controllability and CA-D-observability and prove that they are necessary and sufficient for the existence of a deterministic supervisor. For nonblocking supervisory control, the objective is to ensure that the supervised system can always reach marked states under any attack scenario. We prove that relative closure, CA-D-controllability, and CA-D-observability together are necessary and sufficient for the existence of a nonblocking supervisor. We further develop methods to verify CA-D-controllability and CA-D-observability. We also illustrate our results using a robotic system example.

eess.SY↗

arXiv2026

Agentic Kernel Optimization: Generating State-of-the-Art GPU Kernels Without Hand-Written CUDA

We study whether general-purpose code agents can produce state-of-the-art GPU kernels without any manually written CUDA code. We investigate this question using representative workloads from FlashInfer-Bench, focusing on the Fused MoE, DSA TopK Indexer, and DSA Sparse Attention, and evaluate all generated kernels under the correctness-gated FlashInfer-Bench protocol on NVIDIA B200 GPUs. Starting from the PyTorch implementations, workload definitions, benchmark commands, and a compact set of CUDA optimization skills, we build a kernel optimization workflow in Houmao, a multi-agent orchestration framework for heterogeneous coding agents, to generate, debug, profile, and optimize the kernels. Humans remain strictly in an orchestration role: defining the workflow, enforcing correctness and anti-hacking constraints, supplying key references, and redirecting the search when progress stalls, without reviewing or editing the kernel code itself. Across roughly 1.9 billion agent tokens, the resulting kernels achieve speedups of 92.68x on Fused MoE, 1101.02x on DSA TopK Indexer, and 181.35x on DSA Sparse Attention relative to the PyTorch reference implementations, while also significantly outperforming the corresponding FlashInfer baselines. In the official evaluation of the MLSys 2026 FlashInfer AI Kernel Generation Contest, our generated Fused MoE kernel achieves a 1.71x speedup over the FlashInfer baseline, exceeding the top result of the Fused MoE agent-assisted track, which reports a 1.68x speedup. These results suggest that, under a disciplined correctness-first workflow, code agents can serve as effective autonomous optimizers for modern GPU kernel development.

cs.DC↗

arXiv2026

Coordination Control of Discrete Event Systems under Cyber Attacks

In this paper, coordination control of discrete event systems under joint sensor and actuator attacks is investigated. Sensor attacks are described by a set of attack languages using a proposed ALTER model. Several local supervisors are used to control the system. The goal is to design local supervisors to ensure safety of the system even under cyber attacks (CA). The necessary and sufficient conditions for the existence of such supervisors are derived in terms of conditional decomposability, CA-controllability and CA-observability. A method is developed to calculate local state estimates under sensor attacks. Two methods are also developed to design local supervisors, one for discrete event systems satisfying conditional decomposability, CA-controllability and CA-observability, and one for discrete event systems satisfying conditional decomposability only. The approach works for both stealthy and non-stealthy attacks. A practical example is given to illustrate the results.

eess.SY↗

arXiv2026

Bridging Coarse and Fine Recognition: A Hybrid Approach for Open-Ended Multi-Granularity Object Recognition in Interactive Educational Games

Recent advances in Multimodal Large Language Models (MLLMs) have enabled open-ended object recognition, yet they struggle with fine-grained tasks. In contrast, CLIP-style models excel at fine-grained recognition but lack broad coverage of general object categories. To bridge this gap, we propose \textbf{HyMOR}, a \textbf{Hy}brid \textbf{M}ulti-granularity open-ended \textbf{O}bject \textbf{R}ecognition framework that integrates an MLLM with a CLIP model. In HyMOR, the MLLM performs open-ended and coarse-grained object recognition, while the CLIP model specializes in fine-grained identification of domain-specific objects such as animals and plants. This hybrid design enables accurate object understanding across multiple semantic granularities, serving as a robust perceptual foundation for downstream multi-modal content generation and interactive gameplay. To support evaluation in content-rich and educational scenarios, we introduce TBO (TextBook Objects), a dataset containing 20,942 images annotated with 8,816 object categories extracted from textbooks. Extensive experiments demonstrate that HyMOR narrows the fine-grained recognition gap with CLIP to 0.2\% while improving general object recognition by 2.5\% over a baseline MLLM, measured by average Sentence-BERT (SBert) similarity. Overall, HyMOR achieves a 23.2\% improvement in average SBert across all evaluated datasets, highlighting its effectiveness in enabling accurate perception for multi-modal game content generation and interactive learning applications.

cs.CV↗

arXiv2026

A Diamagnetic, Light-Driven Tesla Engine Based on a Mechanically Displaced, Magnetically Levitated Graphene Disk

Ferromagnetic materials are widely used in Tesla thermomagnetic engines, whereas diamagnetic counterparts have remained unexplored. Here, we demonstrate the first diamagnetic Tesla engine by exploiting the strong diamagnetism of graphene. A graphene disk, fabricated by stacking graphene sheets, serves as the engine wheel. We first show that the conventional Tesla engine design using a permanent magnet placed near the disk edge to create unbalanced thermomagnetic forces under asymmetric local heating fails to generate rotation. We achieve stable operation by laterally displacing the levitated disk from equilibrium, creating a strong restoring force that drives rotation under light excitation. Calculations and measurements establish the displacement-dependent force, with an optimal offset of 0.8 mm yielding speeds up to 2000 rpm under laser heating and 1000 rpm under direct sunlight. Adding vanes allows the disk to function as a gear, powering a graphene vehicle and transferring energy to another disk. This design utilizes the strong and anisotropic diamagnetism of graphene and paves the way for light-powered sensors, actuators, and micro-vehicles.

physics.app-ph↗

arXiv2026

Modular Control of Discrete Event System for Modeling and Mitigating Power System Cascading Failures

Cascading failures in power systems caused by sequential tripping of components are a serious concern as they can lead to complete or partial shutdowns, disrupting vital services and causing damage and inconvenience. In prior work, we developed a new approach for identifying and preventing cascading failures in power systems. The approach uses supervisory control technique of discrete event systems (DES) by incorporating both on-line lookahead control and forcible events. In this paper, we use modular supervisory control of DES to reduce computation complexity and increase the robustness and reliability of control. Modular supervisory control allows us to predict and mitigate cascading failures in power systems more effectively. We implemented the proposed control technique on a simulation platform developed in MATLAB and applied the proposed DES controller. The calculations of modular supervisory control of DES are performed using an external tool and imported into the MATLAB platform. We conduct simulation studies for the IEEE 30-bus, 118-bus and 300-bus systems, and the results demonstrate the effectiveness of our proposed approach.

eess.SY↗