The rapid evolution of artificial intelligence (AI) in IoT applications necessitates the execution of inference tasks on edge devices. However, the deployment of computation-intensive neural networks on resource-constrained edge systems presents a significant challenge. This brief presents the RV-WINO processor, the first silicon implementation of a RISC-V processor based on the Winograd algorithm for convolution and general matrix multiplication (GEMM) acceleration. The processor incorporates a Winograd module, which significantly reduces multiplication operations during convolutions, leading to a substantial decrease in energy consumption. In addition, the processor includes a matrix multiplication module that reuses the multipliers of the Winograd module, accelerating fully connected and dot product operations in neural networks. The RV-WINO processor fabricated in a 55-nm CMOS process achieves the peak computational performance of 0.95 and 2.39 GOPS in INT32 and INT8 modes, with its peak energy efficiency reaching 112 and 237 GOPS/W. In convolutional neural network (CNN) inference tasks, the execution time is reduced by over 80% compared with the baseline processor.

With the expansion of mobile devices and cloud computing, massive spatial trajectory data is generated and outsourced to the cloud for storage and analysis, enabling location-based mobile computing services. However, due to the sensitivity of the trajectory data, sharing it in plaintext could lead to privacy risks, especially in operations like contact queries. Thus, achieving secure and efficient contact queries based on the trajectory data in the cloud is a significant challenge. In this paper, we propose a privacy-preserving contact query processing over trajectory data in mobile cloud computing. The projection-based secure trajectory encoding is designed to convert trajectories into secure codes such that the comparison between the distance of two moving objects and the contact distance threshold is transformed into a problem of secure code matching. Adopting the secure code matching method, a baseline privacy-preserving contact query processing is proposed. To improve the query accuracy and efficiency, an amplification factor, an HTG-index and a filter table are designed for query processing optimization, based on which an enhanced privacy-preserving contact query processing is proposed. The game stimulation-based security analysis and experimental results show that the proposed query scheme is secure and performs well in query accuracy and efficiency.

During infectious disease outbreaks, tracking contacted objects is important for suppressing the spread of the virus and using trajectories of moving objects to discover contacted objects is one of effective approaches. Existing algorithms focus on individual contact event discovery but lack the ability to obtain the k-degree contact events. In this paper, we propose efficient k-degree contact pattern mining algorithms that are capable of mining k-degree contact events. The definition of k-degree contact event is first formulated. Based on the definition, a sliding window-based baseline k-degree contact pattern mining algorithm (KCPMA) is presented. To improve the mining efficiency, the sample-point checking strategy and R-tree index are adopted and the optimized mining algorithm (KCPMA+) is proposed. Comprehensive experiments on real datasets demonstrate that the proposed algorithms are effective and efficient in mining k-degree contact events.

Audio-driven talking face has attracted broad interest from academia and industry recently. However, data acquisition and labeling in audio-driven talking face are labor-intensive and costly. The lack of data resource results in poor synthesis effect. To alleviate this issue, we propose to use TTS (Text-To-Speech) for data augmentation to improve few-shot ability of the talking face system. The misalignment problem brought by the TTS audio is solved with the introduction of soft-DTW, which is first adopted in the talking face task. Moreover, features extracted by HuBERT are explored to utilize underlying information of audio, and found to be superior over other features. The proposed method achieves 17%, 14%, 38% dominance on MSE score, DTW score and user study preference repectively over the baseline model, which shows the effectiveness of improving few-shot learning for talking face system with TTS augmentation.

Ultra-deep neural network(UDNN) tends to yield higher-quality model but its training process is often difficult to handle. Scarce GPU DRAM capacity is the primary bottleneck that limits the depth of neural network and the range of trainable minibatch size. In this paper, we present a scheme that dedicates to make the utmost use of finite GPU memory resource to speed up the training process for UDNN. Firstly, a performance-model guided dynamic swap out/in strategy between GPU and host memory is carefully orchestrated to tackle the out-of-memory problem without introducing performance penalty. Then, a hyperparameter (minibatch size, learning rate) tuning policy is designed to explore the optimal configuration after applying the swap strategy from the perspectives of training time and final accuracy simultaneously. Finally, we verify the effectiveness of our scheme in both single and distributed GPU mode.