Jun He

Audio-driven human gesture synthesis is a crucial task with broad applications in virtual avatars, human-computer interaction, and creative content generation. Despite notable progress, existing methods often produce coarse gestures, lack expressiveness, and fail to fully align with audio semantics. To address these challenges, we propose ExGes, a novel retrieval-enhanced diffusion framework with three key designs: (1) a Motion Base Construction, which builds a gesture library from the training dataset; (2) a Motion Retrieval Module, employing contrastive learning and momentum distillation for retrieving fine-grained reference poses; and (3) a Precise Control Module, integrating partial masking and stochastic masking to enable flexible and fine-grained control. Experimental evaluations on BEAT2 demonstrate that ExGes reduces Fréchet Gesture Distance by 4.55%and improves motion diversity by 5.3% over EMAGE, with user studies revealing a 71.3% preference for its naturalness and semantic relevance.

This study presents SecureFormer, a lightweight transformer-based model designed for network-based intrusion detection in resource-constrained environments. The increasing complexity of IoT networks necessitates intrusion detection systems that can capture complex traffic patterns while remaining efficient enough for edge deployment. To address this challenge, the architecture combines optimised embedding dimensions, compact self-attention design, and a domain-specific tokenisation strategy that converts network flow features into sequence tokens, enabling effective modelling of network behaviour without reliance on large-scale Transformer architectures. The model was evaluated on the flow-based variant of two complex IoT-focused network intrusion detection benchmark datasets, CICIoT2023 and the Edge-IIoTset, with a reduced set of 15 network- and environment-agnostic features selected via feature importance analysis. Experimental results show that SecureFormer achieves a weighted F1 score of 0.99 on CICIoT2023 and 0.96 on Edge-IIoTset while maintaining a footprint of fewer than one million parameters and low inference overhead. The study demonstrates that compact transformer architectures can effectively capture high-level dependencies in structured network traffic, even under aggressive feature reduction. This capability validates their practical viability for real-time intrusion detection in IoT environments.

Drift analysis is one of the main tools for analyzing the time complexity of evolutionary algorithms. However, it requires manual construction of drift functions to bound hitting time for each specific algorithm and problem. To address this limitation, linear drift functions were introduced for elitist evolutionary algorithms. But calculating good linear bound coefficients remains a problem. This paper proposes a new method called drift analysis of hitting probability to compute these coefficients. Each coefficient is interpreted as a bound on the hitting probability of a fitness level, transforming the task of estimating hitting time into estimating hitting probability. A new drift analysis method is then developed to estimate hitting probability, where paths are introduced to handle multimodal fitness landscapes. Explicit expressions are constructed to compute hitting probability, significantly simplifying the estimation process. An advantage of the proposed method is its ability to estimate both the lower and upper bounds of hitting time and to compare the performance of two algorithms in terms of hitting time. To demonstrate this application, two algorithms for the knapsack problem, each incorporating feasibility rules and greedy repair respectively, are compared. The analysis indicates that neither constraint handling technique consistently outperforms the other.