arXiv subjects

Junhao Xia

Publications and source records attributed to Junhao Xia.

4 recordsLinked to original sources

arXiv2026

Agent Harness Distillation: Inference-Time Harness Extraction and Exploitation in Autonomous Multi-Agent Systems

Autonomous multi-agent systems (AMAS) built on large language models (LLMs), such as Hermes, increasingly rely on inference-time harnesses to coordinate reasoning and action. Constructing these harnesses requires substantial engineering effort and computational resources, as they are iteratively optimized over a combinatorial search space while co-evolving with the underlying LLM. Inference-time harnesses therefore constitute valuable intellectual property (IP). Although prior work has investigated IP leakage in static multi-agent systems with pre-configured architectures, it remains unclear whether similar risks arise in AMAS, where harness behavior emerges dynamically during inference. To address this gap, we introduce Agent Harness Distillation (AHD), a framework for studying the security risks arising from inference-time harness extraction in AMAS. We formalize harness extraction as a new security problem and develop an evaluation framework for quantifying such risks. AHD extracts inference-time harness capabilities from a target agent through black-box interactions and consists of two stages. In the pre-distillation stage, AHD infers inference-time harness behaviors from the responses of the target agent and constructs an initial harness. In the post-distillation stage, AHD iteratively refines the initial harness to align with the behavioral patterns of the target agent. Experiments on real-world AMAS across multiple backbone LLMs demonstrate the effectiveness of AHD and reveal substantial IP leakage risks. We further propose a deception-based defense that reduces harness extraction effectiveness while preserving the utility of the protected agent. Our findings uncover a previously underexplored security threat to AMAS.

cs.CR↗

arXiv2026

UrbanAlign: Post-hoc Semantic Calibration for VLM-Human Preference Alignment

Vision-language models (VLMs) can describe urban scenes in rich detail, yet consistently fail to produce reliable human preference labels in domain-specific tasks such as safety assessment and aesthetic evaluation. The standard fix, fine-tuning or RLHF, requires large-scale annotations and model retraining. We ask a different question: can a frozen VLM be aligned with human preferences without modifying any weights? Our key insight is that VLMs are strong concept extractors but poor decision calibrators. We propose a three-stage post-hoc pipeline that exploits this asymmetry: (i) interpretable evaluation dimensions are automatically mined from consensus exemplars; (ii) an Observer-Debater-Judge chain extracts robust concept scores from the frozen VLM; and (iii) locally-weighted ridge regression on a hybrid manifold calibrates these scores to human ratings. Applied as UrbanAlign on Place Pulse 2.0, the framework reaches 72.2% accuracy (kappa=0.45) across six perception categories, outperforming all baselines by +11.0 pp and zero-shot VLM by +15.5 pp, with full interpretability and zero weight modification.

cs.CV↗

arXiv2025

SDQ-LLM: Sigma-Delta Quantization for 1-bit LLMs of any size

Large language models (LLMs) face significant computational and memory challenges, making extremely low-bit quantization crucial for their efficient deployment. In this work, we introduce SDQ-LLM: Sigma-Delta Quantization for 1-bit LLMs of any size, a novel framework that enables extremely low-bit quantization of LLMs while preserving their linguistic reasoning capabilities. A distinctive feature of SDQ-LLM is the continuous adjustability of the Over-Sampling Ratio (OSR), enabling dynamic adaptation to memory or VRAM constraints by selecting fractional OSR (e.g. 2.5 times) for an optimal trade-off between model size and accuracy. SDQ-LLM uses upsampling combined with Sigma-Delta Quantizer to binarize or ternarize LLMs weights, encoding high-precision parameters into 1-bit or 1.58-bit representations, replacing the multiplication operations within linear layers with addition. This approach significantly enhances inference efficiency under extremely low-bit quantization. To further reduce the loss of quantization precision, we incorporate Hadamard-based weight smoothing prior to quantization, improving the stability and robustness of the weight representations. Furthermore, to fully leverage the continuity of the OSR and reduce precision loss, recognizing the correlation between quantization sensitivity and weight variance, we propose a fine-grained, layer- and linear-wise OSR allocation strategy, MultiOSR. This strategy distributes OSR both across layers and within each layer, based on weight variance and parameter scale. Finally, extensive experiments on OPT and LLaMA model families demonstrate that SDQ-LLM achieves a more efficient and high-precision performance even under highly aggressive low-OSR settings. Our code is available at https://github.com/Dreamlittlecat/LLM-Quant-Factory.

cs.LG↗

arXiv2025

DAPE: Dual-Stage Parameter-Efficient Fine-Tuning for Consistent Video Editing with Diffusion Models

Video generation based on diffusion models presents a challenging multimodal task, with video editing emerging as a pivotal direction in this field. Recent video editing approaches primarily fall into two categories: training-required and training-free methods. While training-based methods incur high computational costs, training-free alternatives often yield suboptimal performance. To address these limitations, we propose DAPE, a high-quality yet cost-effective two-stage parameter-efficient fine-tuning (PEFT) framework for video editing. In the first stage, we design an efficient norm-tuning method to enhance temporal consistency in generated videos. The second stage introduces a vision-friendly adapter to improve visual quality. Additionally, we identify critical shortcomings in existing benchmarks, including limited category diversity, imbalanced object distribution, and inconsistent frame counts. To mitigate these issues, we curate a large dataset benchmark comprising 232 videos with rich annotations and 6 editing prompts, enabling objective and comprehensive evaluation of advanced methods. Extensive experiments on existing datasets (BalanceCC, LOVEU-TGVE, RAVE) and our proposed benchmark demonstrate that DAPE significantly improves temporal coherence and text-video alignment while outperforming previous state-of-the-art approaches.

cs.CV↗