arXiv subjects

Xiaohui Zhou

Publications and source records attributed to Xiaohui Zhou.

9 recordsLinked to original sources

arXiv2026

IMPACT: Influence Modeling for Open-Set Time Series Anomaly Detection

Open-set anomaly detection (OSAD) is an emerging paradigm designed to utilize limited labeled data from anomaly classes seen in training to identify both seen and unseen anomalies during testing. Current approaches rely on simple augmentation methods to generate pseudo anomalies that replicate unseen anomalies. Despite being promising in image data, these methods are found to be ineffective in time series data due to the failure to preserve its sequential nature, resulting in trivial or unrealistic anomaly patterns. They are further plagued when the training data is contaminated with unlabeled anomalies. This work introduces $\textbf{IMPACT}$, a novel framework that leverages $\underline{\textbf{i}}$nfluence $\underline{\textbf{m}}$odeling for o$\underline{\textbf{p}}$en-set time series $\underline{\textbf{a}}$nomaly dete$\underline{\textbf{ct}}$ion, to tackle these challenges. The key insight is to $\textbf{i)}$ learn an influence function that can accurately estimate the impact of individual training samples on the modeling, and then $\textbf{ii)}$ leverage these influence scores to generate semantically divergent yet realistic unseen anomalies for time series while repurposing high-influential samples as supervised anomalies for anomaly decontamination. Extensive experiments show that IMPACT significantly outperforms existing state-of-the-art methods, showing superior accuracy under varying OSAD settings and contamination rates. Code is available at https://github.com/mala-lab/IMPACT.

cs.LG↗

arXiv2026

A Benchmark and Multi-Agent System for Instruction-driven Cinematic Video Compilation

The surging demand for adapting long-form cinematic content into short videos has motivated the need for versatile automatic video compilation systems. However, existing compilation methods are limited to predefined tasks, and the community lacks a comprehensive benchmark to evaluate the cinematic compilation. To address this, we introduce CineBench, the first benchmark for instruction-driven cinematic video compilation, featuring diverse user instructions and high-quality ground-truth compilations annotated by professional editors. To overcome contextual collapse and temporal fragmentation, we present CineAgents, a multi-agent system that reformulates cinematic video compilation into ``design-and-compose'' paradigm. CineAgents performs script reverse-engineering to construct a hierarchical narrative memory to provide multi-level context and employs an iterative narrative planning process that refines a creative blueprint into a final compiled script. Extensive experiments demonstrate that CineAgents significantly outperforms existing methods, generating compilations with superior narrative coherence and logical coherence.

cs.CV↗

arXiv2026

Micro-Macro Retrieval: Reducing Long-Form Hallucination in Large Language Models

Large Language Models (LLMs) achieve impressive performance across many tasks but remain prone to hallucination, especially in long-form generation where redundant retrieved contexts and lengthy reasoning chains amplify factual errors. Recent studies highlight a critical phenomenon: the closer key information appears to the model outputs, the higher the factual accuracy. However, existing retrieval-augmented language models (RALMs) lack effective mechanisms to ensure this proximity - external evidence is injected into reasoning via multi-turn retrieval, but this cannot ensure key information stays close to the outputs. We propose Micro-Macro Retrieval (M2R), a novel retrieve-while-generate framework to fill this gap. At the macro level, M2R retrieves coarse-grained evidence from external sources; at the micro level, it extracts essential results from a key information repository built during reasoning and reuses them while generating answers. This design directly addresses the key-information-to-output proximity bottleneck, effectively reducing hallucination in long-form tasks. M2R is trained with a curriculum learning-based reinforcement learning strategy using customized rule-based rewards, enabling stable acquisition of retrieval and grounding skills. Extensive experiments across different benchmarks demonstrate the effectiveness of M2R, especially in lengthy-context settings.

cs.CL↗

arXiv2026

Data Driven Air Entrainment Velocity Parameterization by Breaking Waves

Wave breaking injects turbulence and bubbles into the upper ocean, modulating air-sea exchange of momentum, heat, gases, and sea-spray aerosols. These fluxes depend nonlinearly on sea state but remain poorly represented in coupled atmosphere-wave-ocean models, where air-entrainment velocity is often parameterized using wind speed or significant wave height alone. We develop a global machine-learning parameterization of Va trained on a 43-year WAVEWATCH III simulation that resolves the breaker-front distribution and associated energetics. A multilayer perceptron with seven physically motivated predictors (wind speed, wave height, wave age, steepness, direction, and depth) reproduces spectral-reference Va with high skill. The model reduces longstanding biases in bulk formulas, notably overestimation in swell-dominated low latitudes and underestimation in storm tracks. Applied globally, it improves bubble-mediated CO2 transfer velocity and sea-salt aerosol emission, reducing errors by an order of magnitude. Validation against independent HiWinGS observations supports robust deep-water performance.

physics.ao-ph↗

arXiv2025

Ambiguity Awareness Optimization: Towards Semantic Disambiguation for Direct Preference Optimization

Direct Preference Optimization (DPO) is a widely used reinforcement learning from human feedback (RLHF) method across various domains. Recent research has increasingly focused on the role of token importance in improving DPO effectiveness. It is observed that identical or semantically similar content (defined as ambiguous content) frequently appears within the preference pairs. We hypothesize that the presence of ambiguous content during DPO training may introduce ambiguity, thereby limiting further improvements in alignment. Through mathematical analysis and proof-of-concept experiments, we reveal that ambiguous content may potentially introduce ambiguities, thereby degrading performance. To address this issue, we introduce Ambiguity Awareness Optimization (AAO), a simple yet effective approach that automatically re-weights ambiguous content to reduce ambiguities by calculating semantic similarity from preference pairs. Through extensive experiments, we demonstrate that AAO consistently and significantly surpasses state-of-the-art approaches in performance, without markedly increasing response length, across multiple model scales and widely adopted benchmark datasets, including AlpacaEval 2, MT-Bench, and Arena-Hard. Specifically, AAO outperforms DPO by up to 8.9 points on AlpacaEval 2 and achieves an improvement of by up to 15.0 points on Arena-Hard.

cs.CL↗

arXiv2025

AIMMerging: Adaptive Iterative Model Merging Using Training Trajectories for Language Model Continual Learning

Continual learning (CL) is essential for deploying large language models (LLMs) in dynamic real-world environments without the need for costly retraining. Recent model merging-based methods have attracted significant attention, but they still struggle to effectively manage the trade-off between learning new knowledge and preventing forgetting, a challenge largely stemming from suboptimal number of merges and merging frequency. In this paper, we introduce Adaptive Iterative Model Merging (AimMerging), a novel CL framework that utilizes learning and forgetting signals from the training trajectory to dynamically monitor the model's training status. Guided by dynamic monitoring, the training trajectory-guided merge controller adaptively determines the timing and frequency of iterative fusion, while the rehearsal-based knowledge fusion module computes the merging weights and executes the fusion. Comprehensive experiments on three CL benchmarks with various model sizes (from 770M to 13B) demonstrate that AimMerging achieves significant performance improvements over existing state-of-the-art methods, with an average relative improvement of 80% and 59% on FWT and BWT, respectively. The source code is provided for reproducibility.

cs.CL↗

arXiv2018

Underlay Drone Cell for Temporary Events: Impact of Drone Height and Aerial Channel Environments

Providing seamless connection to a large number of devices is one of the biggest challenges for the Internet of Things (IoT) networks. Using a drone as an aerial base station (ABS) to provide coverage to devices or users on ground is envisaged as a promising solution for IoT networks. In this paper, we consider a communication network with an underlay ABS to provide coverage for a temporary event, such as a sporting event or a concert in a stadium. Using stochastic geometry, we propose a general analytical framework to compute the uplink and downlink coverage probabilities for both the aerial and the terrestrial cellular system. Our framework is valid for any aerial channel model for which the probabilistic functions of line-of-sight (LOS) and non-line-of-sight (NLOS) links are specified. The accuracy of the analytical results is verified by Monte Carlo simulations considering two commonly adopted aerial channel models. Our results show the non-trivial impact of the different aerial channel environments (i.e., suburban, urban, dense urban and high-rise urban) on the uplink and downlink coverage probabilities and provide design guidelines for best ABS deployment height.

cs.IT↗

arXiv2018

Uplink Coverage Performance of an Underlay Drone Cell for Temporary Events

Using a drone as an aerial base station (ABS) to provide coverage to users on the ground is envisaged as a promising solution for beyond fifth generation (beyond-5G) wireless networks. While the literature to date has examined downlink cellular networks with ABSs, we consider an uplink cellular network with an ABS. Specifically, we analyze the use of an underlay ABS to provide coverage for a temporary event, such as a sporting event or a concert in a stadium. Using stochastic geometry, we derive the analytical expressions for the uplink coverage probability of the terrestrial base station (TBS) and the ABS. The results are expressed in terms of (i) the Laplace transforms of the interference power distribution at the TBS and the ABS and (ii) the distance distribution between the ABS and an independently and uniformly distributed (i.u.d.) ABS-supported user equipment and between the ABS and an i.u.d. TBS-supported user equipment. The accuracy of the analytical results is verified by Monte Carlo simulations. Our results show that varying the ABS height leads to a trade-off between the uplink coverage probability of the TBS and the ABS. In addition, assuming a quality of service of 90% at the TBS, an uplink coverage probability of the ABS of over 85% can be achieved, with the ABS deployed at or below its optimal height of typically between 250-500 m for the considered setup.

cs.IT↗

arXiv2017

Power Beacon-Assisted Millimeter Wave Ad Hoc Networks

Deployment of low cost power beacons (PBs) is a promising solution for dedicated wireless power transfer (WPT) in future wireless networks. In this paper, we present a tractable model for PB-assisted millimeter wave (mmWave) wireless ad hoc networks, where each transmitter (TX) harvests energy from all PBs and then uses the harvested energy to transmit information to its desired receiver. Our model accounts for realistic aspects of WPT and mmWave transmissions, such as power circuit activation threshold, allowed maximum harvested power, maximum transmit power, beamforming and blockage. Using stochastic geometry, we obtain the Laplace transform of the aggregate received power at the TX to calculate the power coverage probability. We approximate and discretize the transmit power of each TX into a finite number of discrete power levels in log scale to compute the channel and total coverage probability. We compare our analytical predictions to simulations and observe good accuracy. The proposed model allows insights into effect of system parameters, such as transmit power of PBs, PB density, main lobe beam-width and power circuit activation threshold on the overall coverage probability. The results confirm that it is feasible and safe to power TXs in a mmWave ad hoc network using PBs.

cs.IT↗