11 April 2025 in Engineering Applications of Artificial Intelligence
18 August 2025 in IEEE Transactions on Consumer Electronics
In today’s competitive market, smarter and higher-performing consumer electronics are increasingly attractive to the public. These products not only need to perform basic functions but also offer personalized services to enhance user experience and drive sales growth. To solve this problem, this paper proposes a multimodal emotion recognition model based on Share Token Tensorized Attention Fusion (STF), which improves the intelligence of electronic products by providing in-depth understanding of the user’s emotional state.The STF method constrains the modal feature fusion by introducing the share token and combines with the tensorized attention for feature fusion, which significantly improves the accuracy of emotion recognition, improves the robustness of the model, and reduces the learning cost compared to the Transformer-based multimodal emotion recognition model. The STF-based multimodal emotion recognition model can ensure model performance and faster response speed in noisy environments, and it is easier to meet the consumers’ pursuit of consumer electronics experience.
26 November 2025 in IEEE Transactions on Parallel and Distributed Systems
Large language models in deep learning have numerous parameters, requiring significant storage space and computational resources. Compression techniques are highly effective in addressing these challenges. With the development of hardware like Graphics Processing Unit (GPU), Tensor Core can accelerate low-precision matrix multiplication but achieve acceleration for sparse matrices is challenging. Due to its sparsity, the utilization of Tensor Cores is relatively low. To address this, we propose the based on Tensor Core Compressed Sparse Row format (TC-CSR), which facilitates data loading on GPUs and matrix operations on Tensor Cores. Based on this format, we designed block Sparse Matrix-Matrix Multiplication (SpMM) and Sampled Dense-Dense Matrix Multiplication (SDDMM) kernels, which are common operations in deep learning. Utilizing these designs, we achieved a $\mathbf {1.41\times }$ speedup on Sputnik in scenarios of moderate sparsity and a $\mathbf {1.38\times }$ speedup with large-scale highly sparse matrices. Benefit from our design, we achieved a $\mathbf {1.75\times }$ speedup in end-to-end inference with sparse Transformers and save memory.
1 July 2026 in IEEE Transactions on Computers
9 December 2025 in Concurrency and Computation: Practice and Experience