Federated learning (FL) has achieved favorable progress in addressing the data silo problem without compromising clients' data privacy. In real-world scenarios, there are numerous low-capacity clients, i.e., devices with limited resources like computational power, storage and bandwidth, holding unique and valuable data. Yet the conventional model-homogeneous paradigm is unsuitable due to its uniform model demands on all clients. To effectively utilize the data from clients of various capacities for learning a well-performing global model, researchers have proposed submodel extraction-based partial training methods allowing clients to locally train heterogeneous submodels of different sizes. However, existing partial training methods are inadequate in terms of parameter space coverage efficiency and convergence stability, which adversely affects convergence rate and the final performance. In this work, we introduce FedEcover, a model-heterogeneous framework to learn a fast and stable converging global model in challenging scenarios with dual heterogeneity of data and client capacity. Specifically, our framework incorporates an efficient submodel extraction scheme applying a random sampling without replacement strategy and a step-size decay mechanism in the global aggregation process, to enable the global model fully leveraging the heterogeneous data distributed across capacity-heterogeneous clients. Experimental results on multiple models and datasets demonstrate that our framework outperforms existing submodel extraction-based partial training methods and model-homogeneous FedAvg in both convergence rate and converged performance of the global model.

Artificial intelligence (AI) based anticancer drug recommendation systems have emerged as powerful tools for precision dosing. Although existing methods have advanced in terms of predictive accuracy, they encounter three significant obstacles, including the "black-box" problem resulting in unexplainable reasoning, the computational difficulty for graph-based structures, and the combinatorial explosion during multi-step reasoning. To tackle these issues, we introduce a novel Macro-Micro agent Drug sensitivity inference (MarMirDrug). Specifically, our methodology enhances interpretability via knowledge graphs (KG). To reduce computational overhead, our MarMirDrug also transforms the graph structures of KG into low-dimensional embeddings. To manage the combinatorial explosion of graph paths that occurs with increasing inference steps, we incorporate a macro-micro dual-agent reinforcement learning algorithm, and combines reinforcement learning with KG-based reasoning to infer path reasoning. In computational experimental outcomes, the efficiency of our embedding model surpasses that of the baseline model, achieving an average improvement of 117.65% for hit score. Also, our dual-agent framework exhibits superior performance and interpretability in drug response prediction, with AUC values outperforming conventional baseline methods. In summary, our approach integrates interpretability, computational efficiency, and predictive accuracy, providing novel contributions to precision medicine.

3D Morphable Model (3DMM) is a statistical tool widely employed in reconstructing 3D face shape. Existing methods are aimed at predicting 3DMM shape parameters with a single encoder but suffer from unclear distinction of different attributes. To address this problem, Two-Pathway Encoder-Decoder Network (2PEDN) is proposed to regress the identity and expression components via global and local pathways. Specifically, each 2D face image is cropped into global face and local details as the inputs for the corresponding pathways. 2PEDN is trained to predict 3D face shape components with two sets of loss functions designed to supervise 3D face reconstruction error and face identification error. To reduce the conflict between abundant facial details and saving computer storage space, a magnitudes converter is devised. Experiments demonstrate that the proposed method outperforms several 3D face recontruction methods.

We propose a novel joint framework for 3D face reconstruction (3DFR) that integrates facial attribute estimation (FAE) as an auxiliary task. One of the essential problems of 3DFR is to extract semantic facial features (e.g., Big Nose, High Cheekbones, and Asian) from in-the-wild 2D images, which is inherently involved with FAE. These two tasks, though heterogeneous, are highly relevant to each other. To achieve this, we leverage a Convolutional Neural Network to extract shared facial representations for both shape decoder and attribute classifier. We further develop an in-batch hybrid-task training scheme that enables our model to learn from heterogeneous facial datasets jointly within a mini-batch. Thanks to the joint loss that provides supervision from both 3DFR and FAE domains, our model learns the correlations between 3D shapes and facial attributes, which benefit both feature extraction and shape inference. Quantitative evaluation and qualitative visualization results confirm the effectiveness and robustness of our joint framework.