Siyuan Zhao

Youth esports now extends far beyond professional competition, encompassing ranked online play, school clubs, amateur events, livestreaming, and digitally mediated peer communities. These settings expose adolescents to energy drinks and other stimulant-containing products in contexts where alertness, concentration, and prolonged play are valued. Energy drinks merit particular attention because their prominence in gaming culture combines caffeine exposure with, for many products, substantial added sugar. Potential consequences include disrupted sleep, acute stimulant effects, and adverse nutritional and metabolic profiles. A further concern is behavioral: repeated consumption for gaming-related fatigue or concentration may foster expectations that psychoactive products are useful tools for performance. Research on nonmedical prescription-stimulant use among adolescents demonstrates a comparable performance-oriented motive, while longitudinal studies link energy-drink use with later tobacco, alcohol, and other substance involvement. These associations remain non-causal and have not been established specifically in recreational esports. This Perspective frames mass-participation youth esports as an emerging setting for early public-health intervention. It proposes action across marketing and sponsorship, schools and community programs, beverage and product policy, families and healthcare, and longitudinal surveillance, while emphasizing the need to establish prevalence, temporal ordering, and high-risk patterns directly in nonprofessional adolescent participants across different social and regulatory settings.

Infrared and visible image fusion plays a critical role in multimodal perception systems, particularly under challenging conditions such as low illumination, occlusion, or complex backgrounds. However, existing approaches often struggle with global feature modelling, cross-modal dependency learning, and preserving structural details in the fused images. In this paper, we propose a novel adversarial fusion framework driven by a state-space modelling paradigm to address these limitations. In the feature extraction phase, a computationally efficient state-space model is utilized to capture global semantic context from both infrared and visible inputs. A cross-modality state-space architecture is then introduced in the fusion phase to model long-range dependencies between heterogeneous features effectively. Finally, a multi-class discriminator, trained under an adversarial learning scheme, enhances the structural fidelity and detail consistency of the fused output. Extensive experiments conducted on publicly available infrared–visible fusion datasets demonstrate that the proposed method achieves superior performance in terms of information retention, contrast enhancement, and visual realism. The results confirm the robustness and generalizability of our framework for complex scene understanding and downstream tasks such as object detection under adverse conditions.

When addressing constrained multi-objective optimization problems (CMOPs), the key challenge lies in achieving a balance between the objective functions and the constraint conditions. However, existing evolutionary algorithms exhibit certain limitations when tackling CMOPs with complex feasible regions. To address this issue, this paper proposes a constrained multi-objective evolutionary algorithm based on a dual-population cooperative correlation (CMOEA-DCC). Under the CMOEA-DDC framework, the system maintains two independently evolving populations: the driving population and the conventional population. These two populations share information through a collaborative interaction mechanism, where the driving population focuses on objective optimization, while the conventional population balances both objectives and constraints. To further enhance the performance of the algorithm, a shift-based density estimation (SDE) method is introduced to maintain the diversity of solutions in the driving population, while a multi-criteria evaluation metric is adopted to improve the feasibility quality of solutions in the normal population. CMOEA-DDC was compared with seven representative constrained multi-objective evolutionary algorithms (CMOEAs) across various test problems and real-world application scenarios. Through an in-depth analysis of a series of experimental results, it can be concluded that CMOEA-DDC significantly outperforms the other competing algorithms in terms of performance.

Federated learning offers a powerful approach for training models across decentralized datasets, enabling the creation of machine learning models that respect data privacy. However, federated learning faces significant challenges due to its vulnerability to adversarial attacks, especially when clients have diverse and potentially malicious data distributions. These challenges can lead to severe degradation in the global model’s performance and generalization. In this paper, we present a novel federated image adversarial training framework that combines client-specific adversarial example generation with a robust trimmed aggregation technique. By creating adversarial examples tailored to each client’s local data, our method strengthens individual model defenses against adversarial attacks. Meanwhile, the trimmed aggregation strategy ensures the global model’s robustness by mitigating the impact of harmful or low-quality updates during the model aggregation process. This framework effectively addresses both the issue of data heterogeneity and adversarial threats in federated learning settings. Our experimental results on standard image classification datasets show that the proposed approach significantly enhances model robustness, surpassing existing methods in defending against various adversarial attacks while maintaining high classification accuracy. This framework holds strong promise for real-world applications, particularly in privacy-sensitive domains where both security and model reliability are essential.

The core issue in handling constrained multi-objective optimization problems (CMOP) is how to maintain a balance between objectives and constraints. However, existing constrained multi-objective evolutionary algorithms (CMOEAs) often fail to achieve the desired performance when confronted with complex feasible regions. Building upon this theoretical foundation, a two-stage archive-based constrained multi-objective evolutionary algorithm (CMOEA-TA) based on genetic algorithms (GA) is proposed. In CMOEA-TA, First stage: The archive appropriately relaxes constraints based on the proportion of feasible solutions and constraint violations, compelling the population to explore more search space. Second stage: Sharing valuable information between the archive and the population, while embedding constraint dominance principles to enhance the feasibility of solutions. In addition an angle-based selection strategy was used to select more valuable solutions to increase the diversity of the population. To verify its effectiveness, CMOEA-TA was tested on 54 CMOPs in 4 benchmark suites and 7 state-of-the-art algorithms were compared. The experimental results show that it is far superior to seven competitors in inverse generation distance (IGD) and hypervolume (HV) metrics.

Speech-driven gesture generation using transformer-based generative models represents a rapidly advancing area within virtual human creation. However, existing models face significant challenges due to their quadratic time and space complexities, limiting scalability and efficiency. To address these limitations, we introduce DiM-Gestor, an innovative end-to-end generative model leveraging the Mamba-2 architecture. DiM-Gestor features a dual-component framework: (1) a fuzzy feature extractor and (2) a speech-to-gesture mapping module, both built on the Mamba-2. The fuzzy feature extractor, integrated with a Chinese Pre-trained Model and Mamba-2, autonomously extracts implicit, continuous speech features. These features are synthesized into a unified latent representation and then processed by the speech-to-gesture mapping module. This module employs an Adaptive Layer Normalization (AdaLN)-enhanced Mamba-2 mechanism to uniformly apply transformations across all sequence tokens. This enables precise modeling of the nuanced interplay between speech features and gesture dynamics. We utilize a diffusion model to train and infer diverse gesture outputs. Extensive subjective and objective evaluations conducted on the newly released Chinese Co-Speech Gestures dataset corroborate the efficacy of our proposed model. Compared with Transformer-based architecture, the assessments reveal that our approach delivers competitive results and significantly reduces memory usage, approximately 2.4 times, and enhances inference speeds by 2 to 4 times. Additionally, we released the CCG dataset, a Chinese Co-Speech Gestures dataset, comprising 15.97 hours (six styles across five scenarios) of 3D full-body skeleton gesture motion performed by professional Chinese TV broadcasters.

The generation of co-speech gestures for digital humans is an emerging area in the field of virtual human creation. Prior research has progressed by using acoustic and semantic information as input and adopting a classification method to identify the person’s ID and emotion for driving co-speech gesture generation. However, this endeavor still faces significant challenges. These challenges go beyond the intricate interplay among co-speech gestures, speech acoustic, and semantics; they also encompass the complexities associated with personality, emotion, and other obscure but important factors. This paper introduces “DiT-Gestures”, a speech-conditional diffusion-based and non-autoregressive transformer-based generative model with the WavLM pre-trained model and a dynamic mask attention network (DMAN). It can produce individual and stylized full-body co-speech gestures by only using raw speech audio, eliminating the need for complex multimodal processing and manual annotation. Firstly, considering that speech audio contains acoustic and semantic features and conveys personality traits, emotions, and more subtle information related to accompanying gestures, we pioneer the adaptation of WavLM, a large-scale pre-trained model, to extract the style from raw audio information. Secondly, we replace the causal mask by introducing a learnable dynamic mask for better local modeling in the neighborhood of the target frames. Extensive subjective evaluation experiments are conducted on the Trinity, ZEGGS, and BEAT datasets to confirm WavLM’s and the model’s ability to synthesize natural co-speech gestures with various styles.