arXiv subjects

Junjie Wang

Publications and source records attributed to Junjie Wang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Route by Kinematics, Act by Observation: Kinematics-Supervised Expert Routing in MoE-Augmented VLA

While MoE augments VLA via expert specialization, router suffers from ineffective expert routing owing to the kinematic heterogeneity of actions across manipulation tasks and, even worse, the unavailability of the kinematic signals at inference time. In this work, we first observe that most semantically distinct manipulation tasks reduce to multiple kinematic archetypes. Motivated by this finding, we propose Kinematics-supervised explicit routing (KinRT), a new paradigm that shifts from implicit, observation-driven expert routing to explicit, kinematics-guided expert dispatching. Specifically, we perform kinematic clustering on action trajectories into multiple kinematically coherent groups, whose IDs serve as ground truth to supervise the training of the router; at inference time, the router dispatches experts only using visual-language observations, without any reliance on action kinematics. KinRT actually introduces an asymmetric bridging mechanism that distills the task kinematics from the action space in training into the observation space at inference. In addition, to assess KinRT's cross-platform generalization, we build an economical, Do-It-Yourself robot (DIYRobot) platform from scratch using 3D-print technology ($<$ 2,000USD). Extensive experiments demonstrate KinRT's superiority over both dense and MoE-featured VLAs by more than 23.26% on RoboTwin benchmark and 20.27% on our introduced DIYRobot platform. Our code and DIYRobot platform will be open-sourced.

cs.RO↗

arXiv2026

VCE-Skill: Enhancing Skill Self-Evolution with Version-Change Experience

Agents increasingly rely on reusable skills to encode task knowledge, tool-use procedures, and validation rules. Existing skill self-evolution methods primarily revise skills using execution trajectories collected from current tasks, leaving the evolution knowledge accumulated in public skill version histories largely untapped. Our pilot study reveals a clear complementarity between the two sources: public skill changes provide reusable evolution priors, whereas trajectories provide evidence grounded in the current task. Motivated by this, we propose VCE-Skill, which distills noisy and implementation-specific public skill changes into reusable, structured version-change experience and adaptively fuses it with trajectory-derived proposals from the base evolver, thereby exploiting external experience while retaining task-specific evidence. Extensive experiments demonstrate that VCE-Skill improves skill self-evolution, increasing mean scores by 3.20--4.98 points; transfer experiments further show that the resulting skills achieve stronger cross-model transfer performance. Our work highlights public skill version changes as a previously underexplored yet effective source of prior knowledge and advances trajectory-driven skill self-evolution.

cs.MA↗

arXiv2026

On a conjecture of Kolokolnikov on algebraic connectivity

For a graph $G$, let $α(G)$ be the second smallest eigenvalue of the Laplacian matrix of $G$, also known as the algebraic connectivity. Algebraic connectivity plays an important role in characterizing the connectivity of graphs and convergence properties of networks. Kolokolnikov conjectured that among all graphs on $n$ vertices with exactly $2n-4$ edges, $α(G)\leq 2$ and one of the maximizers is the complete bipartite graph whose two parts have sizes two and $n-2$, respectively. In this paper, we completely resolve this conjecture.

math.CO↗

arXiv2026

Calling the Bluff: Detecting Ever-Shifting Harmful Chat Dialogue via Ordered Reasoning Chain Regularization

Harmful chat dialogues are ever-shifting through type-shifting and lexical evasion, yet we find they share invariant principles, i.e., an Ordered Reasoning Chain (ORC) of recurring topics, harm language indicators, severity hierarchies, and type characteristics, which can help us capture the key information in the frequently changing lexical expressions. We propose BRACE, which encodes the ORC as four differentiable stages (Topic -> Indicator -> Severity -> Type) with intermediate supervision, serving as a structured regularizer blended with direct heads, and supported by prototype-based feature augmentation and feature path disentanglement. The evaluation results show that, across 4 domains and 5 harm categories, BRACE achieves harm-type macro F1 of 0.934 (RoBERTa-wwm-ext, 3-seed mean), with decoder backbones (Qwen3-1.7B LoRA) reaching 0.949. Ablation studies show that all components contribute to BRACE, and the structural decomposition of ORC enables BRACE to distinguish harmful types with semantic ambiguity. Disclaimer: This paper may contain content that is disturbing to some readers.

cs.CL↗

arXiv2026

Simple-OPD: Demystifying Warm-up for On-policy Distillation

On-policy distillation (OPD) trains a student on its own rollouts with token-level supervision from teacher models, but its effectiveness can depend strongly on the warm-up stage before OPD. In this paper, we demystify warm-up for OPD from both data and training perspectives. For data, we find that effective warm-up relies on teacher-compatible chain-of-thought supervision, and that even incorrect teacher rollouts can provide comparable benefits to correct ones. This suggests that warm-up primarily transfers a teacher-compatible thinking pattern rather than merely correct answers. For training, we show that low-rank adaptation (LoRA) with a near-saturation training duration better balances in-domain adaptation and out-of-distribution generalization than full-parameter SFT. Based on these findings, we propose Simple-OPD, a plug-and-play initialization method that warms up the student on teacher-generated CoT with LoRA before OPD. Experiments across diverse settings demonstrate the effectiveness and robustness of Simple-OPD.

cs.CL↗

arXiv2026

Superconducting ternary compounds Li-X-B (X=Mo, W) within the mild pressure range: First-principles predictions

Among the superconducting hydrides under high pressure, a number of studies concentrate on the ternary compounds to explore unique superconductors, which are capable of reducing the stable pressure and maintain superconductivity. In this work, to verify our proposed strategy of ternary composition lines (TCLs) to explore ternary compounds, we combined the first-principles calculations and crystal structure predictions to study the ternary compounds Li-X-B (X=Mo, W) under high pressure. After calculations along five and four TCLs in Li-W-B and Li-Mo-B, respectively, five Li-W-B compounds and four Li-Mo-B compounds were predicted. The compositions of LiWB4, Li4MoB2 and LiMo2B2 could be thermodynamically stable under high pressure, and Li2WB6 is around 0.02 eV/atom above the convex hull at 0 GPa, which has potential for synthesizing. Both of the predicted Li2WB6 P6/mmm and Li2WB4 R-3m are superconducting and their Tc are around 11 K, which are similar to the Tc of WB2 P6/mmm around 100 GPa. An anomalous increase of Tc was found in Li4MoB2 C2/m upon compression. We carried out full ternary search (FTS) to evaluate the validity of the TCLs strategy in Li-W-B system at 0 GPa. Our results are helpful for understanding the phase diagram of Li-X-B (X=Mo, W) under high pressure and the introducing of Li atoms provide candidate structures to reduce the measured stable pressure from ~100 GPa in WB2 P6/mmm to 0 GPa. Meanwhile, we preliminary validate the strategy of TCLs in structure predictions and we expect to improve this strategy in the future, shedding light on the studies of ternary compounds.

cond-mat.supr-con↗

arXiv2026

Do Uncertainty Signals Help? A Systematic Study of Uncertainty-Aware Decoding with Rollback Mechanisms

Prediction uncertainty is a widely adopted metric for quantifying model confidence, with downstream applications spanning model explanation, data selection, and prediction rollback. Despite its demonstrated utility, the potential of uncertainty quantification to enhance code generation in large language models (LLMs) remains largely underexplored, raising a critical question: to what extent can uncertainty serve as an effective signal for improving LLM-based code generation? To answer this question, we study uncertainty-aware rollback decoding, an inference-time strategy that uses uncertainty signals to identify unreliable generation regions and roll back to earlier valid prefixes without retraining the model. We evaluate this framework on seven code LLMs, five code generation benchmarks, and eight token-level uncertainty signals under a unified decoding setup. Our results show that the complete rollback framework improves over equal-budget restart across the evaluated benchmarks and model settings, with gains of up to 0.26 in pass@1 and 0.35 in AvgTestPassRate on functional code generation benchmarks, and an absolute improvement of up to 6.4\% in Patch-Aligned Safe Rate on Dsec-Python. Among the evaluated signals, information-theoretic measures such as token entropy and negative log-likelihood show the most favorable overall trend, frequently achieving the best or near-best results on standard benchmarks. A component-controlled ablation further shows that feedback-guided rollback provides the main improvement, while uncertainty localization provides an additional gain when checking, budget, rollback, and branch decay are held fixed.

cs.LG↗

arXiv2026

Smaller Models are Natural Explorers for Policy-Level Diversity in GRPO

We identify a new dimension for enhancing rollout diversity in Group Relative Policy Optimization (GRPO) for LLMs. While GRPO relies on diverse rollouts, prevailing strategies primarily increase diversity by injecting more token-level randomness, which may introduce step-wise noise and lead to incoherent trajectories. We uncover that smaller models within the same model family inherently exhibit higher policy-level diversity, indicated by their superior pass@k relative to larger counterparts as sample counts increase. Unlike token-level noise, this diversity is temporally correlated, preserves logical consistency, and provides structured exploration signals for gradient estimation. We thus propose S2L-PO (Small-to-Large Policy Optimization), a framework that leverages fixed small models as natural explorers to train larger models. To balance exploration and exploitation, we design a progressive annealing strategy that transitions from offline small-model rollouts to the large learner's own sampling. This shift elegantly avoids mid-training performance drops caused by the small model's capacity limits, achieving faster convergence and unlocking a higher performance ceiling. S2L-PO improves accuracy on diverse mathematical reasoning benchmarks (e.g., +8.8% on AIME 24 using a 1.7B explorer to guide the 8B model) while reducing rollout compute.

cs.LG↗

arXiv2026

Multi-Catheter Digitization in Brachytherapy via Few-Shot Synthetic-to-Real Learning and Structure-Aware Tracking

Accurate catheter digitization in CT-guided interstitial brachytherapy is a critical but time-consuming task, especially for complex implant configurations. We developed a data-efficient, physics-guided framework for automated multi-catheter digitization with minimal clinical annotation. The pipeline consists of two stages. First, an implant region-aware network was pretrained on synthetic CT volumes with simulated metallic signatures and then fine-tuned using only 10 clinical cases. Second, a structure-aware reconstruction module combined a direction-constrained 3D Hough transform with synchronous physics-constrained inward tracking to separate adherent catheter trajectories. The method was evaluated by patient-level five-fold cross-validation on 203 treatment fractions from 38 patients. The fine-tuned network achieved an HD95 of 0.853 +/- 0.362 mm. End-to-end evaluation yielded an F1 score of 0.891 +/- 0.178, with shaft and tip errors of 0.334 +/- 0.367 mm and 0.896 +/- 0.680 mm, respectively. In cases with severe catheter adhesion, the tracking F1 score remained 0.843 +/- 0.190. The complete workflow required approximately 11.6 s per case. These results indicate that combining few-shot synthetic-to-real learning with physics-guided structural tracking can provide robust and efficient multi-catheter digitization for time-sensitive clinical workflows.

physics.med-ph↗

arXiv2026

Graph-Aware Fuzzing for Graph Database Management Systems

Graph Database Management Systems (GDBMSs) are essential infrastructure for managing interconnected data. Existing GDBMS testing methods primarily rely on differential and metamorphic testing. The result consistency oracles of these methods constrain inputs to queries that are comparable across engines or transformations, leaving single engine runtime failures, such as crashes and memory errors, insufficiently explored. Developing dedicated fuzzers for GDBMSs faces two key challenges: (1) generating valid and structurally diverse queries under complex graph constraints, and (2) guiding exploration to capture topology dependent execution behavior. To address these challenges, we propose GRAF, a black box fuzzing framework for GDBMS query engines. First, GRAF introduces graph context aware query generation based on cascading dependency resolution. It instantiates parameterized Cypher skeletons generated by a Large Language Model (LLM) by jointly resolving labels, relationship types, properties, values, and variable scopes against the active graph state. This process produces structurally diverse queries while eliminating syntactic and semantic violations. Second, GRAF applies five graph specific mutation operators guided by execution state feedback, including execution time, result size, and system status. This feedback steers exploration away from unproductive queries and expensive traversals, while prioritizing local mutations around abnormal executions. We evaluated GRAF against three existing approaches on six widely used GDBMSs. GRAF consistently improves line coverage by 31.6% to 41.1% over the strongest baseline on each target. In 12 hour fuzzing, it triggered 25 unique bugs, compared to six from all baselines combined. Overall, GRAF discovered 34 previously unknown bugs, with 32 confirmed by developers and 23 assigned CVEs.

cs.SE↗

arXiv2026

EPIC: Abstraction and Polymorphism of In-Network Collectives on Ethernet

In-Network Collective (INC) acceleration holds immense potential for optimizing AI training and inference; however, its cross-layer nature has historically hindered investment and adoption within the open Ethernet ecosystem. To bridge this gap, we propose EPIC (Ethernet Polymorphic In-network Collective), an INC protocol specification and reference system built on the principle of "Unified Abstraction, Polymorphic Realization." EPIC introduces an abstraction compatible with standard Ethernet that aligns functional boundaries with participant roles, while offering polymorphic realizations tailored to varying hardware capabilities. We address three fundamental challenges: first, we employ a modular design that enables an evolutionary path from simple to complex implementations, allowing vendors to iterate their hardware incrementally; second, we apply formal verification methodologies to prove the correctness of all proposed polymorphic modes; and third, we develop a unified resource management model versatile enough for diverse INC scenarios. Extensive validation -- spanning model checking, packet/flow simulations, VM emulation, Tofino Testbed, and FPGA/RTL verification -- confirms EPIC's correctness, performance gain, and feasibility.

cs.DC↗

arXiv2026

Diagnosis-Driven Automatic Repair for Agentic Workflow via Symbolic Inference

Platform-orchestrated agentic workflows have become a popular paradigm for developing LLM-based applications. However, their reliability remains a major challenge due to the uncertainty of LLM outputs, complex inter-node dependencies, and heterogeneous tool interactions. Existing agentic workflow optimization and agent enhancement methods primarily rely on trajectory-level feedback. Without explicitly identifying the underlying failure root causes, their resulting repair plans are often insufficiently targeted. We propose FlowFixer, a diagnosis-driven automated repair framework for agentic workflows. FlowFixer first transforms workflow executions into unified symbolic traces and performs symbolic inference to derive executable behavioral specifications that capture node correctness, temporal dependencies, and causal relationships. Based on specification verification, it conducts failure attribution and root cause analysis, and then generates targeted repair patches. To reduce verification costs, FlowFixer further employs a multi-dimensional pre-execution assessment to filter infeasible repairs before dynamic verification. We evaluate FlowFixer on workflow failures collected from three popular development platforms: Dify, Coze and n8n. Results show that FlowFixer achieves a repair success rate of 71.3%, outperforming state-of-the-art baselines by 11.9% to 27.6%. It also improves failure attribution accuracy by 4.8% to 33.1% and root cause analysis accuracy by 15.3% to 38.8%. This work offers a new perspective on reliable diagnosis and repair of agentic workflows through symbolic modeling and inference.

cs.SE↗

arXiv2026

From Failed Trajectories to Reliable LLM Agents: Diagnosing and Repairing Harness Flaws

LLM agents increasingly rely on agent harness: the runtime infrastructure around the base model that defines execution environments, tool interfaces, context, lifecycle orchestration, observability, verification, and governance. Existing self-improving agents and automatic harness evolution methods mainly improve agents through runtime supervision, prompt optimization, workflow search, or harness modification based on final outcomes. However, they often fail to diagnose where the responsible evidence lies in failed trajectories and which harness implementation mechanism causes the unreliable behavior, resulting in broad, indirect, or poorly scoped changes. This paper proposes HarnessFix, a trace-grounded and diagnosis-driven framework for repairing agent harnesses. HarnessFix compiles raw execution traces and harness artifacts into a Harness-aware Trace Intermediate Representation (HTIR), which normalizes fragmented trajectory evidence, captures step-level data-flow and control-flow relations, and aligns runtime steps with the harness artifacts that shape their behavior. It then attributes failures to responsible steps and harness artifacts, and consolidates recurring diagnoses into repair-oriented flaw records. Finally, HarnessFix maps these records to scoped repair operators, generates patches under flaw-specific repair specifications, and accepts them through regression-aware validation. We evaluate HarnessFix on four popular benchmarks, and results show that it improves the performance over the initial harnesses by 6.3% to 18.4%, significantly outperforming human-designed and self-evolution baselines. HarnessFix highlights the value of treating failed trajectories not only as feedback signals, but also as structured evidence for diagnosing and repairing the harness mechanisms behind agent failures.

cs.SE↗

arXiv2026

Making Sense of Touch from the Child's View for Contrastive Learning

Is the sense of touch a mechanism for human babies' learning of visual concepts? If so, can we quantify its importance, and to what extent do babies rely on their sense of touch for visual learning? To approach these questions in a principled way, we propose a structured coding system for baby-centric touch events, yielding a dataset of 264k two-second clips of touch events coded according to this system. Using this dataset, we pretrain developmentally grounded models that reveal promising insights into the nature of baby learning from touch.

cs.LG↗

arXiv2026

High-order tensor neural network for iteration-free structure relaxation

Structure relaxation is important for the discovery of new materials, yet conventional ab initio optimization remains a major bottleneck in high-throughput screening workflows. Machine learning potentials have accelerated relaxation by orders of magnitude, but they still rely on iterative optimization and high-quality DFT force labels. Here, we present HotRelax, a high-order tensor message-passing neural network for one-shot, end-to-end prediction of relaxed structures. Trained directly on paired unrelaxed and relaxed structures, HotRelax requires no DFT force labels and predicts relaxed structures in a single forward pass, without iterative inference or post-processing. Across five diverse datasets spanning 3D bulk crystals, 2D layered materials and catalysts, HotRelax shows strong performance relative to state-of-the-art end-to-end relaxation models, achieving lower prediction errors on several benchmarks while maintaining a compact model size and efficient inference. Extensive DFT calculations further show that the predicted structures are close in energy to their DFT-relaxed counterparts. When integrated into catalytic workflows, HotRelax also improves the accuracy and generalization of relaxed-state energy prediction models. Together, these results support HotRelax as an efficient and widely applicable framework for end-to-end structure relaxation, with strong potential to accelerate high-throughput materials discovery.

physics.comp-ph↗

arXiv2026

A lifting theorem for generalized Turán numbers of triangles

For graphs $H$ and $F$, the generalized Turán number $\operatorname{ex}(n,H,F)$ denotes the maximum number of copies of $H$ in an $n$-vertex $F$-free graph. We prove a general lifting principle for the case $H=K_3$ and the forbidden graph is a vertex-disjoint union of several copies of a graph. The key hypothesis is a local neighborhood-forcing condition: there is a graph $R$ with $\operatorname{ex}(n,R)=o(n^2)$ such that $F\subseteq K_1\nabla R$. Under this condition, the corresponding single-forbidden-graph asymptotics, together with a construction attaining the relevant extremal triangle and edge densities simultaneously, lift to an asymptotic value for $\operatorname{ex}(n,K_3,(s+1)F)$ for every integer $s $. We also prove an exact version in terms of the maximum value of a weighted expression over all graphs of a given size that avoid the forbidden graph. As applications, we obtain exact or asymptotic results for disjoint unions of suspensions of paths and stars. We also recover known exact results for disjoint odd cycles.

math.CO↗

arXiv2026

Large Language Models for Mobile GUI Text Input Generation: An Empirical Study

Mobile apps have become essential, making quality assurance increasingly important. GUI testing is widely used for automated exploration, yet text-input components remain a major obstacle, as many UI pages require semantically appropriate text inputs before proceeding. Large Language Models have shown promise in generating context-aware text, but the effectiveness of different UI representations, feedback mechanisms, and human intervention remains unclear. This paper presents a large-scale empirical study addressing these gaps. We evaluate nine state-of-the-art LLMs across 115 real-world apps, comparing three UI-context prompting settings: extracted textual context, UI-hierarchy XML, and screenshot-based vision input. Results show extracted context and XML achieve comparable page-pass-through rates of 71.4% and 71.0%, while vision-based input reaches 65.1% but incurs substantially higher token costs. In bug-detection experiments with 37 real-world text-input bugs, LLMs generating invalid inputs detect about 51% of issues across all evaluated models. A feedback-enhanced protocol, incorporating execution outcomes into subsequent attempts, improves average PPTRs to 69.2-73.8% and raises bug-detection rates to 51.0-64.5%. Human testers further refine inputs, yielding additional gains. We integrate the process into DroidBot, augmenting its UI-exploration capabilities. We derive actionable insights on context selection, cost-effectiveness, feedback strategies, and human-LLM collaboration, advancing both knowledge and practice in Android testing.

cs.SE↗

arXiv2026

Scaling Multi-Reference Image Generation with Dynamic Reward Optimization

While personalized image generation has achieved remarkable progress, multi-reference image generation (MRIG) remains a challenging task. Most existing benchmarks fail to adequately evaluate complex MRIG scenarios, hindering further progress in this area. To better assess model performance on complex MRIG tasks, we introduce OmniRef-Bench, a benchmark that covers complex combinations of reference image types and a large number of reference images. Evaluations on OmniRef-Bench show that mainstream open-source models struggle in complex MRIG scenarios, and their performance deteriorates significantly as the number of mixed-type reference images increases. To address this issue, we propose DyRef, a two-stage training framework. In the first stage, supervised fine-tuning equips the model with the basic capability to handle complex MRIG tasks. In the second stage, we introduce Difficulty-aware Advantage Reweighting (DAR) and Discriminative Reward Scaling (DRS). DAR dynamically adjusts the optimization objective to improve performance when handling a large number of mixed-type reference images. DRS enlarges intra-group reward differences for more effective policy optimization. Experiments demonstrate that DyRef significantly improves the performance of open-source models on OmniRef-Bench and single-image editing benchmarks, demonstrating the effectiveness and generalization capability of our approach.

cs.CV↗