arXiv subjects

Zhilin Liu

Publications and source records attributed to Zhilin Liu.

6 recordsLinked to original sources

arXiv2026

DAREBench: Deployment-Aware and Reliable Evaluation of Models as Agents

As large language models evolve from question-answering systems into general-purpose agents, evaluation must move beyond static answer correctness to assess multimodal perception, multi-step execution, tool use, and artifact delivery. However, existing benchmarks are often tied to specific task types, execution environments, or scoring protocols, limiting their comparability, interpretability, and reliability for deployment decisions. We introduce DAREBench (Deployment-Aware and Reliable Evaluation of Models as Agents), a benchmark designed to capture workload variation and support reliable agent evaluation. Built on a shared OpenClaw execution environment, DAREBench organizes 233 tasks selected and adapted from 22 source benchmarks into a $2\times3$ workload matrix defined by input modality and execution form, and evaluates them under a unified contract-based protocol with evidence-based score auditing. We evaluate 23 commercial API models and 12 locally deployed open-weight models over 7,587 model--task runs, reporting accuracy and token consumption alongside reference costs for API models. Results show that no single model dominates all workload groups, text and multimodal tasks exhibit distinct accuracy--cost trade-offs, and local open-weight models are competitive in several groups but still trail frontier commercial models overall. These findings suggest that agent deployment and model selection should consider workload profiles, deployment mode, and accuracy--cost trade-offs rather than rely on a single aggregate score.

cs.AI↗

arXiv2026

Mastering PokeGym: Graph-Guided Multimodal Evolution at Test Time

While artificial intelligence has mastered structured games like chess and Go, vision-language agents still struggle in visually-driven 3D games without access to game states. Existing game environments typically evaluate a fixed agent configuration, rather than an agent's ability to improve its configuration across consecutive episodes of the same task---a paradigm known as test-time learning (TTL). Furthermore, current TTL methods typically optimize single modalities---such as text prompts or actions---in isolation, ignoring the synergy between perception, reasoning, and control. To bridge these gaps, we first introduce \textbf{PokeGym}, a long-horizon benchmark built upon the 3D open-world game Pok\'emon Legends: Z-A, where agents act from visual observations without access to game states, designed to evaluate an agent's ability to learn and adapt across consecutive episodes of the task. To tackle this challenging environment, we propose Graph-Guided Evolutionary Multimodal Agent Configuration (\textbf{G-EvoMAC}), a graph-guided framework that jointly optimizes visual perception, strategy, and action set synergistically. Extensive experiments show that G-EvoMAC achieves a 60.18\% average success rate on PokeGym, outperforming the strongest baseline by over 11 percentage points, validating the power of cross-modal co-evolution.

cs.CV↗

arXiv2026

Coding with Eyes: Visual Feedback Unlocks Reliable GUI Code Generating and Debugging

Recent advances in Large Language Model (LLM)-based agents have shown remarkable progress in code generation. However, current agent methods mainly rely on text-output-based feedback (e.g. command-line outputs) for multi-round debugging and struggle in graphical user interface (GUI) that involve visual information. This is mainly due to two limitations: 1) GUI programs are event-driven, yet existing methods cannot simulate user interactions to trigger GUI element logic 2) GUI programs possess visual attributes, making it difficult for text-based approaches to assess whether the rendered interface meets user needs. To systematically address these challenges, we first introduce InteractGUI Bench, a novel benchmark comprising 984 commonly used real-world desktop GUI application tasks designed for fine-grained evaluation of both interaction logic and visual structure. Furthermore, we propose VF-Coder, a vision-feedback-based multi-agent system for debugging GUI code. By perceiving visual information and directly interacting with program interfaces, VF-Coder can identify potential logic and layout issues in a human-like manner. On InteractGUI Bench, our VF-Coder approach increases the success rate of Gemini-3-Flash from 21.68% to 28.29% and raises the visual score from 0.4284 to 0.5584, indicating the effectiveness of visual feedback in GUI debugging.

cs.SE↗

arXiv2025

Semantic segmentation with reward

In real-world scenarios, pixel-level labeling is not always available. Sometimes, we need a semantic segmentation network, and even a visual encoder can have a high compatibility, and can be trained using various types of feedback beyond traditional labels, such as feedback that indicates the quality of the parsing results. To tackle this issue, we proposed RSS (Reward in Semantic Segmentation), the first practical application of reward-based reinforcement learning on pure semantic segmentation offered in two granular levels (pixel-level and image-level). RSS incorporates various novel technologies, such as progressive scale rewards (PSR) and pair-wise spatial difference (PSD), to ensure that the reward facilitates the convergence of the semantic segmentation network, especially under image-level rewards. Experiments and visualizations on benchmark datasets demonstrate that the proposed RSS can successfully ensure the convergence of the semantic segmentation network on two levels of rewards. Additionally, the RSS, which utilizes an image-level reward, outperforms existing weakly supervised methods that also rely solely on image-level signals during training.

cs.CV↗

arXiv2025

Wireless Control over Edge Networks: Joint User Association and Communication-Computation Co-Design

This paper studies a wireless networked control system with multiple base stations (BSs) cooperatively coordinating the wireless control of a number of subsystems each consisting of a plant, a sensor, and an actuator. In this system, each sensor first offloads the sensing data to its associated BS, which then employs mobile edge computing (MEC) to process the data and sends the command signals back to the actuator for remote control. We consider the time-division-multiple-access (TDMA) service protocol among different BSs to facilitate the cascaded communication and computation process, in which different BSs implement the uplink data collection and downlink command broadcasting over orthogonal time slots. We also employ the massive multiple-input multiple-output (MIMO) at BSs, based on which each BS serves its associated sensors or actuators over the same time-frequency resources via spatial multiplexing. Under this setup, we jointly design the association between BSs and sensors/actuators as well as the joint communication and computation resource allocation, with the objective of minimizing the closed-loop control latency of the multiple subsystems while ensuring their control stability. The optimization takes into account the transmission uncertainty caused by both the hyper reliable and low-latency communications (HRLLC) and the inter-user interference , as well as the communication and computation resource constraints at distributed nodes. To solve the challenging non-convex joint optimization problem, we develop an efficient algorithm by employing the techniques of alternating optimization and successive convex approximation (SCA). Numerical results show that the proposed joint BS-sensor/actuator association and resource allocation design significantly outperforms other heuristic schemes and frequency-division-multiple-access (FDMA) counterpart.

cs.IT↗

arXiv2020

High temperature strength retention of Cu/Nb nanolaminates through dynamic strain ageing

The mechanical properties of Cu/Nb metallic nanolaminates with different layer thickness (7, 16, 34 and 63 nm) were studied by means of micropillar compression tests from room temperature to 400 C. Both strain-rate jump and constant strain rate tests were carried out and they showed evidence of dynamic strain ageing in the nanolaminates with 7, 16 and 34 nm layer thickness deformed at 200 C. Dynamic strain ageing was accompanied by a reduction of the strain rate sensitivity to 0, high strength retention at 200 C and the development of shear localization of the deformation at low strains (5%-6%) that took place along the Nb layers in the nanolaminates. Atom probe tomography of the deformed specimens revealed the presence of O in solid solution in the Nb layers but not in the Cu layers. Thus, diffusion of O atoms to the mobile dislocations in Nb was found to be the origin of the dynamic strain ageing in the Cu/Nb nanolaminates around 200 C. This mechanism was not found at higher temperatures (400 C) because deformation was mainly controlled by stress-assisted diffusion in the Cu layers. This discovery shows a novel strategy to enhance the strength retention at high temperature of metallic nanolaminates through dynamic strain ageing of one the phases.

cond-mat.mtrl-sci↗