Search arXiv⌕ Search

arXiv subjects

Siyuan Liu

Publications and source records attributed to Siyuan Liu.

At least 19 recordsLinked to original sources
arXiv2026

Trial by Trace: Benchmarking AI Agent Harnesses for Reliability and Failure Attribution

LLM agents are increasingly used to solve real-world problems, and their effectiveness depends on both the underlying models and the harnesses that support their operation. However, most existing benchmarks focus on model capabilities and overall agent performance, with limited attention to the evaluation of agent harnesses. To address this gap, we introduce TraceTrial-Bench, which organizes harness reliability into seven components covering context, execution, governance, lifecycle, observability, tools, and verification. We construct 95 expert-level, mechanism-oriented tasks with executable or state-based ground truth to support semantically equivalent comparisons while preserving the native behavior of each harness. We design a Trajectory Tracing (TT) module to capture agent execution traces and introduce a two-layer evaluation framework that separately assesses agent performance and harness reliability. In addition, we design a Multi-source Evidence Acquisition (MEA) module to provide evidence from the agent context, harness, and execution environment for LLM-based evaluation, and develop a Failure Attribution (FA) module that uses explicit mechanism chains to support attribution at the run level. Across 4 harnesses, 9 models, and 95 tasks, our evaluation of 3,420 runs yields macro-averaged harness scores ranging from 59.8% to 72.0%, while 31.1% of runs achieve task success despite failing at least one harness criterion. TraceTrial-Bench provides an auditable and reproducible foundation for comparing agent harnesses, diagnosing their failures, and studying how harness design interacts with model behavior.

cs.AI↗
arXiv2026

Observer-Based Control for Multi-Agent Systems under Communication-Task Graph Mismatch and Directed Switching Topologies

This paper addresses decentralized estimation and control of heterogeneous multi-agent systems communicating over directed switching topologies. The objective is to satisfy cooperative tasks that require coordination among agents according to a time-invariant task graph. To address the potential mismatch between the communication and task graphs, each agent is equipped with a k-hop Prescribed Performance State Observer (k-hop PPSO) to estimate the states of agents up to k communication hops away using only information exchanged with its 1-hop neighbors. A novel prescribed performance funnel construction is developed to account for the switching communication topology, and a sufficient lower bound on the dwell-time is derived to guarantee the prescribed performance specifications under topology variations. Further analysis shows that, if the desired control objective is achieved with exact knowledge of the required nonlocal states and the resulting closed-loop system is uniformly set-ISS with respect to the state estimation error, then replacing these states with their estimates guarantees convergence to a neighborhood of the desired set. The size of this neighborhood can be made arbitrarily small by appropriately selecting the observer performance bounds. A simulation study demonstrates task satisfaction under directed switching topologies.

eess.SY↗
arXiv2026

Evolving in Thought Space: Training a Small Model at Test Time Unlocks Better Discoveries

Open-ended scientific discovery often requires repeatedly proposing and evaluating candidate solutions. LLM-based systems can support this process by generating and refining executable solutions from verifier feedback. Methods such as TTT-Discover use test-time training (TTT) to update the solution-generating LLM from verifier feedback, adapting its generation policy to improve subsequent proposals on the target problem. However, this becomes expensive when reliable execution requires a large model, since training must maintain gradients, optimizer states, and policy statistics while repeatedly generating long, structured outputs. It also complicates credit assignment: outcome-level verifier feedback must jointly evaluate the high-level strategy and its low-level implementation. In this work, we introduce Guidance-TTT, which separates these roles. A compact guidance model is trained at test time to propose high-level strategic changes, while a frozen execution model implements them as complete executable solutions. At each step, the system selects a promising previously discovered solution, proposes a change, executes and verifies it, and updates only the guidance model using an adaptive group-relative RL objective. This concentrates test-time learning on short strategic decisions while retaining the implementation capability of a substantially stronger model without adapting it. Without web access, Guidance-TTT produces strong solutions across four distinct domains: combinatorial optimization (Polyomino Packing), heuristic programming (AHC058), machine learning (Lasso), and GPU kernel optimization (TriMul). Across these tasks, it outperforms the best solutions reported in prior work while remaining competitive with state-of-the-art results on public online leaderboards. Code is available at https://github.com/Human-Agent-Society/reef/tree/guidance-ttt-support.

cs.AI↗
arXiv2026

DENSE: Distilling Agent Trajectories into Evidence-Grounded Shortcut Trees for Self-Refinement

Online agent deployments accumulate execution trajectories at massive scale and behavioral diversity, for which predefined annotation criteria hardly exist. Extracting useful evidence therefore demands costly manual annotation or verifier signals that fail to scale, leaving valuable evidence buried among redundant, incomplete, and failed executions. This raises a question: without post-execution rewards or correctness labels, how can reusable experience be distilled from the trajectories themselves? To address this challenge, we introduce DENSE (Distilling Evidence from Nested Subtask Executions), which organizes trajectory-derived evidence into nested shortcut trees. By consolidating redundant attempts, identifying resolved subtasks, and retaining useful steps alongside outstanding requirements, DENSE transforms noisy execution traces into structured and reusable task-solving feedback. To evaluate whether such feedback helps agents retry the same task, we design REFIT, which measures success-rate changes between the initial attempt and feedback-guided retries. Among feedback methods without external outcome supervision, DENSE achieves the highest strict pass rate across four agent models on Terminal-Bench 2.1, improving over initial attempts by 7.12-21.81 percentage points with 19.0-43.6% fewer agent tokens on retries. In addition, on hard tasks DENSE consistently outperforms self-reflection in cumulative pass rate across multiple feedback iterations on all four models, demonstrating its strong potential for continual agent self-improvement.

cs.AI↗
arXiv2026

Controllable and Photorealistic Pedestrian Risky Motion Generation for End-to-End Driving Safety Evaluation

Evaluating end-to-end autonomous driving under rare, safety-critical vehicle-pedestrian interactions requires photorealistic, sensor-level scenarios. However, trajectory-based scenario generators cannot synthesize raw visual observations, whereas video-based approaches lack controllability. To bridge this gap, we present ControlPed, a novel framework that combines trajectory-level conflict synthesis with 3D Gaussian Splatting (3DGS) to generate photorealistic, motion-controllable safety-critical scenarios. Built upon HazardPed, a dataset derived from 10,352 traffic videos comprising 422 conflict trajectories, HD maps, and 857 annotated 3D human motions, ControlPed first generates conflict trajectories, lifts them into 3D human motion sequences via text-conditioned motion diffusion, and finally renders multi-view sensor observations using animatable 3DGS avatars. Safety evaluation in 88 rendered photorealistic scenarios reveals that seven leading end-to-end driving models suffer a severe performance drop, with their mean HDScore plunging from 88.8 to 47.4, exposing major failure modes under dangerous pedestrian behaviors. The dataset and testing benchmarks will be released to facilitate safety assessment of vehicle-pedestrian interactions.

cs.RO↗
arXiv2026

Is Better Teacher Supervision Enough? Unlocking Student-side Learning in Multimodal On-Policy Distillation

On-policy distillation (OPD) improves reasoning by providing token-level supervision from a teacher on a student's own trajectories. Existing methods primarily focus on enhancing this teacher-side guidance (e.g., by enriching teacher inputs and refining teacher feedback), yet we find that limited student perception is another critical bottleneck in multimodal OPD. By providing oracle visual facts, the performance of OPD-trained students can still be substantially improved for both weak and strong teachers. To address this bottleneck, we propose S-OPD, a simple multimodal on-policy distillation framework that explicitly strengthens student perceptual learning through two objectives. Specifically, Teacher-calibrated Policy Contrast separates student policies under original and masked images with teacher-based token-level gating, strengthening the student's reliance on visual evidence during reasoning. Policy Agreement aligns student policies under original and noise-perturbed images, further improving perceptual robustness to visual noise. Notably, our method can be seamlessly plugged into existing OPD frameworks, requiring no additional data annotations, model parameters or inference operations. Extensive experiments on eight benchmarks across student scales and distillation paradigms demonstrate consistent performance improvements, with gains of up to 4.25 points on LogicVista. When combined with existing teacher-side supervision methods, our method can yield further gains. Code is available at https://github.com/Sirilaw/S-OPD.

cs.CV↗
arXiv2026

Defending Against Malicious Finetuning by Scaling Train-time Adversarial Attacks

Malicious finetuning attacks pose a major safety threat against open-weight large language models (LLMs). However, existing alignment-stage defenses provide limited protection against strong attacks that use full-parameter finetuning. To address this issue, we propose Patcher, a novel and efficient adversarial training algorithm that alternates between an attack stage and a defense stage: the attack stage simulates many-step malicious finetuning and computes the resulting parameter displacement as an "attack vector", while the defense stage aims to preserve safe behavior under such perturbations. Compared with other adversarial training methods, a key feature of Patcher is that the attack vector can be reused across multiple defense updates, substantially reducing training costs. Theoretically, we characterize how the relative gradient estimation error is influenced by the number of attack steps during training and reuse-duration of attack vectors. Empirically, experiment results show that Patcher reduces Attack Success Rate by 67.5%, 53.4% and 54.8% on three benchmarks compared to the second-best method while preserving the model's utility. Moreover, Patcher remains effective under diverse model sizes and attack scenarios, and generalizes to prompt-based jailbreak attacks. Code is available at https://github.com/haomingwen/patcher

cs.CL↗
arXiv2026

What Should a Self-Teacher See? Privileged Context Design for On-Policy Self-Distillation

More privileged information does not always make a better teacher. We study this tension in on-policy self-distillation (OPSD), where a self-teacher scores the student's own rollouts under privileged context, conventionally a complete reference solution that bundles the final answer with one particular reasoning path. Holding the student view and training fixed within each scale, we compare that default against three abstractions compiled offline, a named strategy, a method-independent framing, and a problem category, and against an answer-only control that keeps the destination but removes the path. In the primary runs on competition mathematics, the best intermediate contexts improve the in-domain peak mean over the full solution by 1.4 points at 4B and 1.6 at 8B, while storing an order of magnitude fewer hint tokens. Comparisons across three seeds also show positive mean gains for the framing and category contexts at both scales. Answer-only conditioning remains competitive in the primary runs, within 0.2 points of the full solution at these scales. The preferred context varies with student scale and task. Initial teacher-student KL does not order downstream performance. What a self-teacher should see is therefore not everything it could, but the level of abstraction its student can still act on.

cs.LG↗
arXiv2026

First Demonstration of Flip DRAM from Process, Architecture to System to Push DRAM Scaling beyond 4F2: 2F2 Self-aligned Flip Vertical Channel Transistor (FVCT) DRAM and Flip WL (FWL) 3D-DRAM

For the first time, we proposed a novel stacking technology for DRAM scaling by flipping and backside processes, making full use of DRAM wafer's backside and investigating it on both 4F2 and 3D-DRAM. For 4F2 VCT, 2F2 Flip VCT featuring self-aligned back-to-back stacked 1T1C bitcell, with various BL and WL configurations, were studied and key process modules such as self-aligned stacked vertical channel, BL and WL formations, wafer bonding and flipping, substrate thinning and low-R Co storage node (SN) were successfully developed, addressing the potential thermal, misalign and parasitic concerns in the Flip VCT process. A full DRAM DTCO framework was also established from device to mat and chip level. Compared to 4F2 VCT DRAM with the same mat size, 2F2 FVCT delivers 27.5% less parasitics, 11% better sense margin, 16.3% higher charge sharing (CS) speed and 50% less area. For 3D-DRAM, a brand-new flip WL staircase design with peripheral circuit innovations was studied and proved to have 25% density gain, 15.1% faster turn-on speed and 6.8% less CS time, proving further extendibility of flip technology on DRAM.

cond-mat.mes-hall↗
arXiv2026

VST: Verifiable Structured Transport for Auditable Agent-to-Agent Alpha Discovery

Agent-to-agent (A2A) alpha discovery is slowed by repeated feedback cycles between mining and evaluation agents, whose hand-offs, in contemporary LLM multi-agent systems, are free-form natural-language messages that carry no stable contract and cannot be replayed. We first restructure this communication as a structured agent-to-agent protocol of \emph{typed, causally addressable, unicast records}, so that the committed stream forms a causal trajectory. On that trajectory a single predictor with four typed heads forecasts the accumulated guidance the two miners would receive several cycles ahead; a transactional verify--leap controller then commits a multi-cycle speculative outcome only when it passes a four-level gate, and otherwise rolls back to the exact prior state. Structure is the enabling contribution, and its value is not accuracy. A controlled ablation shows an equal-information free-text channel reaches the same predictor hit rate. What typing provides is a state that can be schema-checked, replayed deterministically, and prevented by construction from leaking a forecast to an evaluator: auditability by construction, not an empirically stress-tested guarantee. On a CSI~1000 out-of-sample holdout, our single run is the only one among eight methods (seven baselines and ours) to hold a positive median annualized return and Sharpe at the factor level, though the median return \emph{in excess} of the benchmark stays negative for every method including ours; its development-selected top-20 portfolios reach a $0.71$ median holdout Sharpe, selected on a split inside the optimization horizon. We report these single-run results descriptively, gross of costs, and are explicit about their limits throughout; in particular we do not isolate the effect of the leap machinery from the inherited search substrate, which we leave to future work.

cs.AI↗
arXiv2026

A*-Thought-V2: Efficient Latent Reasoning via Geometric Dynamics of LLM

Chain-of-Thought (CoT) improves the reasoning ability of Large Language Models (LLMs) but incurs substantial computation and context costs. Existing methods either lose intermediate information through hard pruning or lack a principled criterion for continuous compression. We present A*-Thought-V2, a geometric dynamics of LLM guided framework that models CoT as a hidden-state trajectory and replaces hard deletion with an explicit-implicit interleaved latent architecture. After projecting question, step, and solution representations into a 3D PCA space, it measures alignment between each local transition and global question-to-solution direction. Aligned steps remain explicit text, whereas deviating steps are compressed into continuous latent tokens. Directional angles capture both local semantics and reasoning dynamics: small angles indicate direct execution and answer formation, while large angles more frequently involve checking, correction, and branch exploration; their temporal variation reveals exploration, convergence, and refinement stages. To train this architecture, we introduce stepwise embedding forcing, which pools each redundant step into a single latent embedding, and label forcing, which supervises that latent token with a soft multi-modal vocabulary distribution instead of a hard one-hot label. Experiments on Qwen3.5-9B and Qwen3.6-27B across six in-domain and out-of-domain benchmarks show that A*-Thought-V2 improves average accuracy by up to 2.6% while reducing response length by up to half, increasing Accuracy per Computation Unit by 2.29$\times$, and reducing preprocessing and training time by 94.6% and up to 80.3%, respectively. Representation analyses suggest that latent states form a compact region distinct from textual states, while higher entropy at latent-token positions reflects broader soft targets that encourage richer step-level feature learning.

cs.CL↗
arXiv2026

DriveZero: End-to-End Driving Beyond Human Demonstrations

Most end-to-end autonomous-driving systems learn by imitating human driving logs, leaving their learned behavior constrained by the quality and behavioral coverage of the recorded trajectories. This report presents DriveZero, an end-to-end system that learns driving behavior beyond human demonstrations. It decomposes driving into a perception model and an action model, pretrains each in the regime best suited to it, and combines them into one end-to-end planner. The two models call for different learning recipes: perception must understand the world, and benefits from massive and diverse visual data; action must interact with it, and requires closed-loop feedback. On the action side, we introduce DriveRL, a mixed-agent closed-loop reinforcement-learning framework. It converts real driving logs into interactive worlds, where a privileged teacher policy is trained with PPO through closed-loop rollouts. For the perception model, DriveVFM consolidates multiple frozen vision foundation models, including DINOv3, SigLIP2, SAM and Depth Anything V2, into a single backbone from raw images alone, requiring no task-specific annotations. DriveZero then unifies the two: a camera-only planner that distills the frozen DriveRL teacher through its rolled-out trajectories. The goal-conditioned teacher can moreover be queried under augmented driving intents, yielding diverse, goal-consistent supervision that logged data cannot provide. On nuPlan, DriveRL with value-guided test-time action search achieves a mean score of 93.57 across the Val14, Test14-hard, and Test14-random community splits in both non-reactive and reactive modes, exceeding the Log-Replay expert on all three splits. DriveZero achieves state-of-the-art performance on NAVSIMv1, NAVSIMv2 and the closed-loop HUGSIM benchmark without any human trajectory supervision.

cs.CV↗
arXiv2026

Bridging Modalities and Tasks: A Unified Hierarchical ViT for SAR-to-Optical Translation and Semantic Segmentation

Synthetic Aperture Radar (SAR) images have all-weather, day-and-night observation capabilities. However, compared with optical images, their speckle noise and non-intuitive scattering mechanism limit the interpretability of the images. Generative models for SAR-to-optical (S2O) conversion can improve visual interpretability, but existing methods often ignore the constraints on semantic structure, which are necessary for downstream tasks, for the sake of visual effects. We propose a unified collaborative dual-task learning framework, termed BMT (Bridging Modalities and Tasks), that jointly optimizes S2O image translation and semantic segmentation through a shared hierarchical Vision Transformer. The framework integrates: (1) a LocalViTBlock that fuses global self-attention with spatial depthwise convolution through a learnable gating mechanism; (2) an enhanced output module combining multi-scale refinement processing, color correction and anti-aliasing, which calibrates channel-level color statistics through feature fusion; (3) a ControlNet-style conditional injection mechanism that encodes SAR wavelet features and segmentation labels into a multi-scale feature pyramid and injects them at each encoder layer through zero-initialized convolution; (4) a bounded Kendall uncertainty weighting scheme that prevents either task from dominating the shared representation. We evaluate the framework under both paired and unpaired translation settings, on the public WHU-OPT-SAR paired dataset and a self-constructed unpaired ship dataset built from HRSID and DIOR, respectively. The experimental results show that the proposed method achieves competitive S2O translation quality and semantic segmentation performance. The dataset and source code have been publicly released at https://github.com/Lewisyuaner/BMT-S2O-main.

cs.CV↗
arXiv2026

EMemBench: Interactive Benchmarking of Episodic Memory for VLM Agents

We introduce EMemBench, a programmatic benchmark generator for evaluating long-term episodic memory of agents through interactive games. Rather than using a fixed set of questions, EMemBench generates questions from environment-grounded trajectories, covering both text-only and visual game environments. Each template computes verifiable ground truth from underlying game signals, with controlled answerability and balanced coverage over memory skills: single/multi-hop recall, induction, temporal, spatial, logical, and adversarial. We evaluate memory agents with strong LMs/VLMs as backbones, using in-context prompting as baselines. Across 15 text games and multiple visual seeds, results are far from saturated: induction and spatial reasoning are persistent bottlenecks, especially in visual settings. Persistent memory yields clear gains for open backbones on text games, but improvements are less consistent for VLM agents, suggesting that visually grounded episodic memory remains an open challenge. A human study further contextualizes the difficulty and interpretability of EMemBench.

cs.CL↗
arXiv2026

Coupled anisotropic weak topological states and Floquet mixed-parity altermagnetism in two-dimensional Su-Schrieffer-Heeger models

Su-Schrieffer-Heeger (SSH) topological systems and altermagnetic (AM) states are two important research areas in condensed matter physics. Realizing the coupled states between the SSH lattices and AM phanse in a single syetem remains challenging. However, the Floquet engineering change it. Our work constructs five two-dimensional-(2D-) SSH models to describe the coupled phases of SSH weak topological state and AM order by tight-binding (TB) method. The evolution of band structures, spin-splitting, and topological phase transitions under the circularly polarized light (CPL) and relative atomic displacement (RAD) in 2D SSH lattices were systematically investigated. The results reveal that the inequality of hopping parameters (t1 and t2) serves as the fundamental origin of SSH topological states and AM order. For the 2D nonmagnetic state, anisotropic weak topological states with Zak phase governed edge states are realized by unit cell selection, similar to the conventional 1D SSH model. The collinear antiferromagnetic state preserves the spin-degenerate band structure and intrinsic weak topological properties. Furthermore, the Floquet engineering introduces the AM phase of odd-parity p-wave while the RAD introduces the AM phase of even-parity d-wave. By combining the two effects, the mixed-parity (non-odd/non-even parity) AM phase could be realized, achieving the simultaneous control of the light field of spin-splitting and topological edge states. The physical mechanisms of Floquet engineering and RAD for the 2D rectangular SSH lattice are also from the inequality of t1 and t2, which can be not only fully understand by the TB methods, but also in good agreement with the first-principle calculations of 2D carbon-based materials. This work establishes an effective theoretical platform for coupling anisotropic SSH weak topological states and AM orders with multi-parities in 2D systems.

cond-mat.mtrl-sci↗
arXiv2026

Bridging Quantum Mechanics to Liquid Properties via a Universal Organic Force Field

Molecular dynamics simulations are essential tools for unraveling atomic-level insights into the structure and behavior of condensed-phase systems. However, the universal and accurate prediction of macroscopic properties based on quantum mechanical calculations remains a significant challenge, often hindered by the trade-off between computational cost and simulation accuracy. Here we present ByteFF-Pol, a polarizable force field parameterized by a graph neural network and trained exclusively on high-level quantum mechanical data. By leveraging physically-motivated force field forms and training strategies, ByteFF-Pol predicts thermodynamic and transport properties for a wide range of small-molecule liquids and electrolytes with high accuracy, surpassing current classical and machine learning force fields. This ability to make predictions without system-specific training bridges the gap between microscopic calculations and macroscopic liquid properties, enabling the exploration of previously intractable chemical spaces. This advancement enables the precise design of new electrolytes and custom-tailored solvents, establishing a robust foundation for data-driven materials discovery.

physics.comp-ph↗
arXiv2026

Self-Supervised Topologically Invariant Manifold Learning for Railway Image Quality Assessment

Existing blind image quality assessment (BIQA) methods typically rely on synthetic distortions and subjective annotations, limiting generalization in real-world domains. To address this, we propose a fully self-supervised BIQA framework based on topologically invariant manifold learning under boundary constraints, which constructs a stable quality reference without manual labels. The framework generates progressive background dilution scales via repeated random cropping around each target; exploiting the monotonic degradation of target information density across these scales, it establishes a self-constrained quality manifold. A linearized spatial moment projection eliminates geometric distortions from random cropping; then a monotonicity divergence filter prunes background-sensitive evaluators, isolating an elite pool \(\mathcal{M}_{\text{elite}}\). A robust M-estimator with a principal component stabilizer fuses the metrics into an asymptotically efficient pseudo-ground truth \(q_{\text{PGT}}\), contracting variance toward the Cramér-Rao lower bound. Extensive evaluations demonstrate that the elite evaluator pool, distilled from 11 baseline metrics, secures superior zero-shot transferability across standard synthetic and wild benchmarks (CSIQ, LIVEC, LIVE-2). Concurrently, deployments on the CQU Railway Rolling Stock Surveillance Dataset (2,797 images) yield a manifold cosine similarity \(>0.999\) and a 100.0\% survival rate under industrial extreme stresses, robustly validating its cross-paradigm decoupling and topological resilience.

cs.CV↗
arXiv2026

VFEAgent: A Multimodal Agent Framework for End-to-End Automated Finite Element Analysis

Finite Element Analysis (FEA) serves as the cornerstone of modern engineering design. However, its workflow is inherently complex and relies heavily on domain expertise. Although recent efforts have integrated Large Language Models (LLMs) into FEA, existing approaches face limitations in handling multimodal inputs and executing complex tasks. To address these limitations, we propose VFEAgent, an end-to-end multi-agent system designed to automate FEA modeling and simulation directly from input images and problem descriptions. Our methodology integrates two core components: (1) a multimodal vision-language multi-agent pipeline that employs ReAct-driven reasoning to extract structured FEA specifications from heterogeneous inputs and (2) a verification-first code synthesis framework, incorporating robust self-debugging and fallback mechanisms to ensure executability and physical validity. We systematically evaluated the system across various engineering mechanics scenarios. The results demonstrate that VFEAgent achieves a high success rate in generating complete and physically valid simulations, outperforming LLM-based baseline methods in reliability and correctness. These findings validate the feasibility of automating the complete FEA workflow, highlighting the framework's potential to liberate engineers from tedious manual analysis.

cs.AI↗