Xuyang Cao

Xuyang Cao, Guoxin Wang, Sheng Shi et al.

Audio-driven portrait animation has made significant advances with diffusion-based models, improving video quality and lipsync accuracy. However, the increasing complexity of these models has led to inefficiencies in training and inference, as well as constraints on video length and inter-frame continuity. In this paper, we propose JoyVASA, a diffusion-based method for generating facial dynamics and head motion in audio-driven facial animation. Specifically, in the first stage, we introduce a decoupled facial representation framework that separates dynamic facial expressions from static 3D facial representations. This decoupling allows the system to generate longer videos by combining any static 3D facial representation with dynamic motion sequences. Then, in the second stage, a diffusion transformer is trained to generate motion sequences directly from audio cues, independent of character identity. Finally, a generator trained in the first stage uses the 3D facial representation and the generated motion sequences as inputs to render high-quality animations. With the decoupled facial representation and the identity-independent motion generation process, JoyVASA extends beyond human portraits to animate animal faces seamlessly. The model is trained on a hybrid dataset of private Chinese and public English data, enabling multilingual support. Experimental results validate the effectiveness of our approach. Future work will focus on improving real-time performance and refining expression control, further expanding the applications in portrait animation. The code is available at: https://github.com/jdh-algo/JoyVASA.

Xuyang Cao, Qianying Liu, Chuan Xiao et al.

In multilingual pretraining, the test loss of a pretrained model is heavily influenced by the proportion of each language in the pretraining data, namely the \textit{language mixture ratios}. Multilingual scaling laws can predict the test loss under different language mixture ratios and can therefore be used to estimate the optimal ratios. However, the current approaches to multilingual scaling laws do not measure the \textit{cross-lingual transfer} effect, resulting in suboptimal mixture ratios. In this paper, we consider multilingual pretraining as a cooperative game in which each language acts as a player that jointly contributes to pretraining, gaining the resulting reduction in test loss as the payoff. Consequently, from the perspective of cooperative game theory, we quantify the cross-lingual transfer from each language by its contribution in the game, and propose a game-theoretic multilingual scaling law called \textit{ShapleyLaw}. Our experiments show that ShapleyLaw outperforms baseline methods in model performance prediction and language mixture optimization.

Sheng Shi, Xuyang Cao, Jun Zhao et al.

In audio-driven video generation, creating Mandarin videos presents significant challenges. Collecting comprehensive Mandarin datasets is difficult, and the complex lip movements in Mandarin further complicate model training compared to English. In this study, we collected 29 hours of Mandarin speech video from JD Health International Inc. employees, resulting in the jdh-Hallo dataset. This dataset includes a diverse range of ages and speaking styles, encompassing both conversational and specialized medical topics. To adapt the JoyHallo model for Mandarin, we employed the Chinese wav2vec2 model for audio feature embedding. A semi-decoupled structure is proposed to capture inter-feature relationships among lip, expression, and pose features. This integration not only improves information utilization efficiency but also accelerates inference speed by 14.3%. Notably, JoyHallo maintains its strong ability to generate English videos, demonstrating excellent cross-language generation capabilities. The code and models are available at https://jdh-algo.github.io/JoyHallo.

Guoxin Wang, Jun Zhao, Xinyi Liu et al.

Medical imaging provides critical evidence for clinical diagnosis, treatment planning, and surgical decisions, yet most existing imaging models are narrowly focused and require multiple specialized networks, limiting their generalization. Although large-scale language and multimodal models exhibit strong reasoning and multi-task capabilities, real-world clinical applications demand precise visual grounding, multimodal integration, and chain-of-thought reasoning. We introduce Citrus-V, a multimodal medical foundation model that combines image analysis with textual reasoning. The model integrates detection, segmentation, and multimodal chain-of-thought reasoning, enabling pixel-level lesion localization, structured report generation, and physician-like diagnostic inference in a single framework. We propose a novel multimodal training approach and release a curated open-source data suite covering reasoning, detection, segmentation, and document understanding tasks. Evaluations demonstrate that Citrus-V outperforms existing open-source medical models and expert-level imaging systems across multiple benchmarks, delivering a unified pipeline from visual grounding to clinical reasoning and supporting precise lesion quantification, automated reporting, and reliable second opinions.

Guoxin Wang, Xuyang Cao, Shan An et al.

Deep learning approaches, together with neuroimaging techniques, play an important role in psychiatric disorders classification. Previous studies on psychiatric disorders diagnosis mainly focus on using functional connectivity matrices of resting-state functional magnetic resonance imaging (rs-fMRI) as input, which still needs to fully utilize the rich temporal information of the time series of rs-fMRI data. In this work, we proposed a multi-dimension-embedding-aware modality fusion transformer (MFFormer) for schizophrenia and bipolar disorder classification using rs-fMRI and T1 weighted structural MRI (T1w sMRI). Concretely, to fully utilize the temporal information of rs-fMRI and spatial information of sMRI, we constructed a deep learning architecture that takes as input 2D time series of rs-fMRI and 3D volumes T1w. Furthermore, to promote intra-modality attention and information fusion across different modalities, a fusion transformer module (FTM) is designed through extensive self-attention of hybrid feature maps of multi-modality. In addition, a dimension-up and dimension-down strategy is suggested to properly align feature maps of multi-dimensional from different modalities. Experimental results on our private and public OpenfMRI datasets show that our proposed MFFormer performs better than that using a single modality or multi-modality MRI on schizophrenia and bipolar disorder diagnosis.

Xuyang Cao, Oliver Michel, Kyle Jamieson

Video conferencing over 5G is increasingly prevalent, yet its Quality of Experience (QoE) often degrades under limited radio resources. This has two causes: 5G networks must serve many users, while interactive traffic requires careful handling. Motivated by the insight that different subflows within an interactive session have a disproportionate effect on QoE, we present the design and implementation of StreamGuard, a practical 5G architecture for subflow-level, QoE-aware prioritization. StreamGuard forms a closed control loop with three components: (1) a monitor in the Radio Access Network (RAN) that uses deep packet inspection to infer QoE and RAN state, (2) a controller that selects prioritization actions to balance QoE and fairness, and (3) a marking module that applies these decisions by marking packets to steer subflows into appropriate priority queues. StreamGuard further shapes application behaviors via mechanisms including selective subflow dropping and probe-based rate control, to align application behavior with radio constraints. Implemented in a real 5G testbed, StreamGuard achieves a superior QoE-fairness tradeoff compared to vanilla 5G and prior state-of-the-art approaches, improving QoE by up to 70% at comparable background throughput or preserving up to 2x higher background throughput at similar QoE.

Tao Yuan, Haoli Bai, Yinfei Pan et al.

With the increasing size of large language models, layer pruning has gained increased attention as a hardware-friendly approach for model compression. However, existing layer pruning methods struggle to simultaneously address key practical deployment challenges, including performance degradation, high training costs, and limited acceleration. To overcome these limitations, we propose \name, a task-\underline{E}ffective, training-\underline{E}conomical and inference-\underline{E}fficient layer pruning framework. \namespace introduces two key innovations: (1) a differentiable mask optimization method using a Gumbel-TopK sampler, enabling efficient and precise pruning mask search; and (2) an entropy-aware adaptive knowledge distillation strategy that enhances task performance. Extensive experiments over diverse model architectures and benchmarks demonstrate the superiority of our method over state-of-the-art approaches. Notably, \namespace achieves 96\% accuracy, a mere 0.8\% drop from the original model (96.8\%) on MATH-500 when pruning 25\% layers of Qwen3-32B, outperforming existing SOTA (95\%), with a 1.33$\times$ inference speedup by consuming merely 0.5B tokens (0.5\% of the post-training data volume).