There has been significant recent progress to reduce the computational effort of static IR drop analysis using neural networks, and modeling as an image-to-image translation task. A crucial issue is lack of sufficient data from real industry designs to train these networks. In this work, we first propose a number of improvements to the state-of-the-art U-Net neural network model to achieve better IR drop prediction. First, we propose U-Net with attention gates which allows selective emphasis on relevant parts of the input data without supervision. This is desired because of the often sparse nature of the IR drop map. We also embed the U-Net model with a preprocessing convolutional block which introduces an initial perimage filter to better handle the multi-image to single-image nature of the problem. Next, to address lack of sufficient data we propose a two-phase training process which utilizes a mix of artificially-generated data and a limited number of points from real designs with custom learning and dropout rates at each phase, and a custom loss function. We also propose a data augmentation step based on image transformations to augment the training data. Based on the ICCAD 2023 contest setup, our results are on-average, 38% (64%) better in MAE and 26% (142%) in F1 score compared to the winner of the ICCAD 2023 contest (and U-Net only [3]) when only tested on the set of real designs in the testing set.

In this paper, we introduce ReBERT, a specialized large language model (LLM) based on BERT, fine-tuned specifically for grouping bits into words within gate-level netlists. By treating the netlist as a form of language, we encode bits and their fan-in cones into sequences that capture structural dependencies. A novel contribution is augmenting BERT's embedding with a tree-based embedding strategy which mirrors the hierarchical nature of circuit designs in hardware. Leveraging the powerful representational learning capabilities of LLMs, we interpret hardware circuits at a higher level of abstraction. We evaluate ReBERT on various hardware designs, demonstrating that it significantly outperforms a state-of-the-art work based on partial structural matching in recovering word-level groupings. Our improvements are on average between 12.2% to 218.1% depending on degree of corrupting the structural patterns.

Deploying large language models (LLMs) on edge devices enables real-time responses while preserving user privacy. However, constrained memory and compute resources pose significant challenges for high-quality, single-device inference. To address this, we propose HCInfer, a hierarchical coordination framework for collaborative LLM inference across edge devices. By leveraging idle neighboring devices, HCInfer alleviates performance bottlenecks typical in isolated deployments. HCInfer employs a two-level coordination strategy. At the inter-device level, it leverages idle neighboring devices to collaboratively process attention computations, significantly reducing synchronization overhead. At the intra-device level, it applies finegrained memory and compute optimizations to fully exploit local hardware capabilities. Building on this architecture, HCInfer integrates three key components: (1) Asymmetric Transformer decomposition decouples attention and FFN computation, enabling selective and parallel execution across devices. (2) Layer-wise subdeadline scheduling dynamically profiles execution latency and adapts precision or structure to meet real-time constraints (3) An Overhead Mitigation Module efficiently manages on-device resource usage to support scalability without overwhelming hardware. We evaluate HCInfer on PC, smart home, and mobile platforms using OPT-13B, Qwen2.5-14B, and Llama2-13B models. Experiments show HCInfer achieves 1.67× to 4.3× speedup in TTFT and 1.16× to 17.15× speedup in TPOT compared to existing baselines, maintaining a sub-deadline miss rate (SubDMR) of 15.3% under worst-case conditions while keeping model accuracy degradation within 8% for typical cases and up to 11% in extreme scenarios. These results demonstrate HCInfer's potential to enable efficient and responsive LLM inference in real-world edge environments.

Social networks are fundamental to virtual communities (e.g., forums, blogs) and virtual communities benefit from well-established social networks. As making friends with other members is a common way to establish social relationships and people need to decide whom they should trust when making friends, friend recommendation has received considerable attention in virtual communities. Towards this goal, we first formulate hypotheses on factors that influence trust and the probability of establishing friendships from various interaction attributes in virtual communities. Through experiments on real interaction and friendship data, we validate the proposed hypotheses and propose a novel interaction ranking-based trustworthy friend recommendation model called IntRank for recommending trustworthy friends to community members. Different from traditional friend recommendation mechanisms, IntRank is built on the foundation of carefully verified interaction attributes that influence trust and friendship probability in virtual communities. It is able to effectively recommend trustworthy friends as confirmed by the performance evaluation results.