Fast and accurate pre-routing timing prediction is essential in the chip design flow. However, existing machine learning (ML)assisted pre-routing timing methods often overlook the impact of power delivery networks (PDNs), which contribute to IR drop and routing congestion. This limitation can make these methods less practical for realworld circuit design flows. To address this, we propose two specialized encoders-an IR drop-aware encoder and a routing congestion-aware encoder-that effectively capture PDN effects through multimodal fusion of netlist, layout, and PDN data. To mitigate the challenges of imbalanced multimodal fusion, we further develop a Pareto optimization approach to ensure balanced utilization of all modalities, enhancing timing prediction accuracy. Comprehensive experiments on large-scale open-source designs using TSMC’s 16 nm technology node validate the superiority of our model over state-of-the-art pre-routing timing prediction methods.

Perception-optimized Learned Image Compression (LIC) methods have recently made significant progress. They surpass both traditional image compression algorithms and non-perceptually optimized LIC methods in image sharpness, detail representation, and subjective perception, even at similar or lower bitrates. However, LIC methods optimized for perception often generate false details and textures in reconstructions, leading to underperformance in objective metrics like PSNR and MS-SSIM, which limits their applicability. In this paper, we introduce a comprehensive loss metric based on saliency detection that aids in achieving exceptional perceptual quality while minimizing distortions. By applying this metric in training, we develop the SaLIC model, i.e., Saliency-Enhanced Learned Image Compression. User study results indicate that, at similar or lower bitrates, SaLIC exhibits better human perceptual quality compared to HiFiC and VVC; quantitative results show that the PSNR of SaLIC significantly outperforms HiFiC (by 1-2dB), and the MS-SSIM of SaLIC even surpasses VVC, achieving a balance between perception and distortion.

Abstract Wearable human action recognition (HAR) has practical applications in daily life. However, traditional HAR methods solely focus on identifying user movements, lacking interactivity and user engagement. This paper proposes a novel immersive HAR method called MovPosVR. Virtual reality (VR) technology is employed to create realistic scenes and enhance the user experience. To improve the accuracy of user action recognition in immersive HAR, a multi‐scale spatio‐temporal attention network (MSSTANet) is proposed. The network combines the convolutional residual squeeze and excitation (CRSE) module with the multi‐branch convolution and long short‐term memory (MCLSTM) module to extract spatio‐temporal features and automatically select relevant features from action signals. Additionally, a multi‐head attention with shared linear mechanism (MHASLM) module is designed to facilitate information interaction, further enhancing feature extraction and improving accuracy. The MSSTANet network achieves superior performance, with accuracy rates of 99.33% and 98.83% on the publicly available WISDM and PAMPA2 datasets, respectively, surpassing state‐of‐the‐art networks. Our method showcases the potential to display user actions and position information in a virtual world, enriching user experiences and interactions across diverse application scenarios.

To achieve good performance in face recognition, a large scale training dataset is usually required. A simple yet effective way to improve the recognition performance is to use a dataset as large as possible by combining multiple datasets in the training. However, it is problematic and troublesome to naively combine different datasets due to two major issues. First, the same person can possibly appear in different datasets, leading to an identity overlapping issue between different datasets. Naively treating the same person as different classes in different datasets during training will affect back-propagation and generate nonrepresentative embeddings. On the other hand, manually cleaning labels may take formidable human efforts, especially when there are millions of images and thousands of identities. Second, different datasets are collected in different situations and thus will lead to different domain distributions. Naively combining datasets will make it difficult to learn domain invariant embeddings across different datasets. In this paper, we propose DAIL: Dataset-Aware and Invariant Learning to resolve the above-mentioned issues. To solve the first issue of identity overlapping, we propose a dataset-aware loss for multi-dataset training by reducing the penalty when the same person appears in multiple datasets. This can be readily achieved with a modified softmax loss with a dataset-aware term. To solve the second issue, domain adaptation with gradient reversal layers is employed for dataset invariant learning. The proposed approach not only achieves the state-of-the-art results on several commonly used face recognition validation sets, including LFW, CFP-FP, and AgeDB-30, but also shows great benefit for practical use.