Speech-driven 3D motion generation has garnered increasing research attention. However, it faces significant challenges in achieving style controllability, primarily due to the scarcity of motion style annotations. To address this, we propose a novel diffusion-based framework for co-speech holistic motion generation that enables example-based style control from videos. Our approach integrates hierarchical speech encoding with rhythm-aware denoising to produce natural and synchronized gestures and expressions. For effective style guidance, we introduce a contrastive style encoder that captures discriminative style representations from reference clips without explicit labeling, enabling generalization to motion styles unseen during training. Furthermore, we design a neural mapper that aligns 2D and 3D gesture features in a shared embedding space, facilitating direct style extraction from in-the-wild videos and seamless transfer to 3D motion. Extensive experiments and user studies show that our proposed approach achieves state-of-the-art performance in both qualitative and quantitative evaluations, offering a flexible solution for controllable motion generation.

Multi-modal neural architecture search (MNAS) is an effective approach to obtain task-adaptive multi-modal classification models. Deep neural networks, as currently main-stream feature extractors, can provide hierarchical features for each modality. Existing MNAS methods face difficulty in exploiting such hierarchical features due to their different form coexistence such as tensorial multi-scale features and vectorized penultimate features. Moreover, existing methods always focus on the evolution of fusion operators or vectorized features of all modalities, constraining search space. In this paper, a novel two-stage method called multi-modal multi-scale evolutionary neural architecture search (MM-ENAS) is proposed. The first stage unifies the representation form of hierarchical features by the proposed evolutionary statistics strategy. The second stage identifies the optimal combination of basic fusion operations for all unified hierarchical features by the evolutionary algorithm. MM-ENAS increases search space by simultaneously searching for feature statistical extraction methods, basic fusion operators and feature representation set consisting of tensorial multi-scale features and vectorized penultimate features. Experimental results on three multi-modal tasks demonstrate that the proposed method achieves competitive performance in terms of accuracy, search time, and number of parameters compared to existing representative MNAS methods. Additionally, the method exhibits fast adaptation to various multi-modal tasks.

Multi-task learning deals with multiple related tasks simultaneously by sharing knowledge. In a typical deep multi-task learning model, all tasks use the same feature space and share the latent knowledge. If the tasks are weakly correlated or some features are negatively correlated, sharing all knowledge often leads to negative knowledge transfer among. To overcome this issue, this paper proposes a Fisher sparse multi-task learning method. It can obtain a sparse sharing representation for each task. In such a way, tasks share features on a sparse subspace. Our method can ensure that the knowledge transferred among tasks is beneficial. Specifically, we first propose a sparse deep multi-task learning model, and then introduce Fisher sparse module into traditional deep multi-task learning to learn the sparse variables of task. By alternately updating the neural network parameters and sparse variables, a sparse sharing representation can be learned for each task. In addition, in order to reduce the computational overhead, an heuristic method is used to estimate the Fisher information of neural network parameters. Experimental results show that, comparing with other methods, our proposed method can improve the performance for all tasks, and has high sparsity in multi-task learning.

Multi-task learning uses knowledge transfer among tasks to improve the generalization performance of all tasks. For deep multi-task learning, knowledge transfer is often implemented via sharing all hidden features of tasks. A major shortcoming is that it can lead to negative knowledge transfer across tasks when task correlation is weak. To overcome it, this paper proposes an evolutionary method to learn sparse sharing representations adaptively. By embedding the neural network optimization into evolutionary multitasking, our proposed method finds an optimal combination of tasks and sharing features. It can identify negative correlation and redundant features and then remove them from the hidden feature set. Thus, an optimal sparse sharing subnetwork can be produced for each task. Experiment results show that the proposed method achieve better learning performance with a smaller inference model than other related methods.