arXiv subjects

Puhan Luo

Publications and source records attributed to Puhan Luo.

5 recordsLinked to original sources

arXiv2026

What Guides the Agent? Adjudicating Unauthorized Behavior via Localizing Behavior-Guiding Instructions

LLM agents integrated with external resources gain complex task capabilities, yet the unified natural-language context channel makes them vulnerable to injection attacks: untrusted external data may be dynamically parsed as behavior-guiding instructions during LLM inference, thereby subverting the agent's decision. Existing defenses focus on static detection or isolation of malicious content at the input/output level, remains insufficient for detecting such dynamic inducements that arise during model reasoning. We propose Attnlocate, a runtime framework for fine-grained localization of context spans that genuinely influence tool-calling decisions, i.e., behavior-guiding instructions. Attnlocate casts this localization problem as an object detection task, aiming to detect the distinctive activation traces induced by behavior-guiding instructions within the attention matrix. Specifically, we design a multi-head, multi-layer attention aggregation scheme to construct a token-level feature space tailored for object detection. Then, a 1-D U-Net equipped with an anchor-free detection head is deployed to detect these spans. Finally, based on the authority of the provider from which the detected behavior-guiding spans originate, Attnlocate dynamically adjudicates malicious invocation attempts. We evaluate Attnlocate across ten agent configurations from five LLM families, covering scenarios involving indirect prompt injection and tool poisoning. Attnlocate achieves a mean IoU of 0.743, an average AUROC of 0.956, and a 0.934 true-positive rate at 0.067 false-positive rate. It also transfers effectively across unseen models and supports authority policy adaptation without retraining.

cs.CR↗

arXiv2026

Coverage-Driven Adaptive Keyframe Selection for Video Understanding

Recent advances in large vision-language models (LVLMs) have enabled long-video understanding and analysis. However, processing the large number of frames in a video incurs substantial computational overhead. Existing methods reduce LVLM inference costs by scoring frame-query relevance before inference and selecting keyframes accordingly. Nevertheless, the distribution of relevant frames varies across queries, and these methods often need to score hundreds or thousands of frames. To address this limitation, we propose CSES, a training-free semantic keyframe selector that adaptively determines the numbers of frames to score and keyframes to select. CSES estimates the prominence of the frame-query relevance profile to guide active acquisition and adapt the temporal coverage of each input. It then formulates keyframe selection as a coverage problem that jointly accounts for semantic relevance, temporal redundancy, and visual redundancy. Active acquisition and keyframe selection terminate based on coverage saturation. The selection objective is monotone and submodular, enabling greedy optimization with a standard approximation guarantee. Experiments with four LVLMs on two benchmarks show that our method preserves accuracy while scoring $4$-$13\times$ fewer frames and selecting $18.4\%$-$20.5\%$ fewer input keyframes than existing baselines. CSES further achieves a $3.1$-$5.4\times$ speedup in frame selection over baselines.

cs.CV↗

arXiv2026

ComPrivDet: Efficient Privacy Object Detection in Compressed Domains Through Inference Reuse

As the Internet of Things (IoT) becomes deeply embedded in daily life, users are increasingly concerned about privacy leakage, especially from video data. Since frame-by-frame protection in large-scale video analytics (e.g., smart communities) introduces significant latency, a more efficient solution is to selectively protect frames containing privacy objects (e.g., faces). Existing object detectors require fully decoded videos or per-frame processing in compressed videos, leading to decoding overhead or reduced accuracy. Therefore, we propose ComPrivDet, an efficient method for detecting privacy objects in compressed video by reusing I-frame inference results. By identifying the presence of new objects through compressed-domain cues, ComPrivDet either skips P- and B-frame detections or efficiently refines them with a lightweight detector. ComPrivDet maintains 99.75% accuracy in private face detection and 96.83% in private license plate detection while skipping over 80% of inferences. It averages 9.84% higher accuracy with 75.95% lower latency than existing compressed-domain detection methods.

cs.CV↗

arXiv2026

GCA-BULF: A Bottom-Up Framework for Short-Term Load Forecasting Using Grouped Critical Appliances

With the rise of time-of-use and tiered electricity pricing, energy consumers are encouraged to adopt peak-shifting strategies by automatically controlling high-power appliances. These help lower energy costs while enhancing the power grid's stability. To support such energy management with high resilience and responsiveness, reliable short-term load forecasting (STLF) plays a critical role. STLF predicts electricity consumption over time horizons ranging from minutes to days, using historical data, temporal patterns, and contextual factors. Traditional top-down forecasting methods struggle to capture the complex consumption patterns of diverse and mixed appliance loads. Although bottom-up methods improve forecasting accuracy by integrating appliance-level data, monitoring all appliances is costly, and many do not meaningfully impact total load prediction. Therefore, we propose GCA-BULF, a bottom-up short-term load forecasting framework based on grouped critical appliances, supported by three key designs. First, the Critical Appliance Filtering module ranks appliances according to their power consumption, switching frequency, and usage pattern periodicity, and identifies critical ones through iterative load decomposition. Next, the Related Appliance Grouping module clusters these appliances based on spatial and temporal correlations for group-level forecasting. Finally, the Collaborative Load Forecasting module refines the total load prediction by combining multiple group-level forecasts. We evaluate GCA-BULF on residential and office building load forecasting tasks. Experimental results reveal that GCA-BULF improves hourly total load forecasting by 20.85%-57.88% compared to existing top-down methods and by 33.03%-92.48% compared to bottom-up methods.

cs.LG↗

arXiv2024

A-VL: Adaptive Attention for Large Vision-Language Models

The Large Vision-Language Model (LVLM) integrates computer vision and natural language processing techniques, offering substantial application potential. However, these models demand extensive resources during inference. Adaptive attention techniques can dynamically reduce computational redundancy and thus improve efficiency. Although current adaptive attention methods significantly reduce the memory requirements of Transformer-based language models, they are not tailored for LVLMs. We observe that LVLMs generate responses from both remote image tokens and local text tokens, and different modalities have different attention patterns. This observation inspires us to manage the attention for each modality separately. Specifically, for visual input, we store the cache of potentially useful information but only compute the most critical parts. For language input, we care more about local information. Based on our observation and analysis of vision-language attention patterns, we develop A-VL, a plug-and-play adaptive attention tailored for LVLM inference. Extensive evaluations on three vision-language tasks and five datasets show the effectiveness of our designs. Our approach A-VL outperforms existing adaptive attention methods in reducing memory usage and computational load without compromising performance.

cs.AI↗