Zeyu Zhang

Zeyu Zhang, Shuyan Wan, Sijie Wang et al.

Signed Graph Neural Networks (SGNNs) are vital for analyzing complex patterns in real-world signed graphs containing positive and negative links. However, three key challenges hinder current SGNN-based signed graph representation learning: sparsity in signed graphs leaves latent structures undiscovered, unbalanced triangles pose representation difficulties for SGNN models, and real-world signed graph datasets often lack supplementary information like node labels and features. These constraints limit the potential of SGNN-based representation learning. We address these issues with data augmentation techniques. Despite many graph data augmentation methods existing for unsigned graphs, none are tailored for signed graphs. Our paper introduces the novel Signed Graph Augmentation framework (SGA), comprising three main components. First, we employ the SGNN model to encode the signed graph, extracting latent structural information for candidate augmentation structures. Second, we evaluate these candidate samples (edges) and select the most beneficial ones for modifying the original training set. Third, we propose a novel augmentation perspective that assigns varying training difficulty to training samples, enabling the design of a new training strategy. Extensive experiments on six real-world datasets (Bitcoin-alpha, Bitcoin-otc, Epinions, Slashdot, Wiki-elec, and Wiki-RfA) demonstrate that SGA significantly improves performance across multiple benchmarks. Our method outperforms baselines by up to 22.2% in AUC for SGCN on Wiki-RfA, 33.3% in F1-binary, 48.8% in F1-micro, and 36.3% in F1-macro for GAT on Bitcoin-alpha in link sign prediction.

Xidong Wu, Shangqian Gao, Zeyu Zhang et al.

Current techniques for deep neural network (DNN) pruning often involve intricate multi-step processes that require domain-specific expertise, making their widespread adoption challenging. To address the limitation, the Only-Train-Once (OTO) and OTOv2 are proposed to eliminate the need for additional fine-tuning steps by directly training and compressing a general DNN from scratch. Nevertheless, the static design of optimizers (in OTO) can lead to convergence issues of local optima. In this paper, we proposed the Auto-Train-Once (ATO), an innovative network pruning algorithm designed to automatically reduce the computational and storage costs of DNNs. During the model training phase, our approach not only trains the target model but also leverages a controller network as an architecture generator to guide the learning of target model weights. Furthermore, we developed a novel stochastic gradient algorithm that enhances the coordination between model training and controller network training, thereby improving pruning performance. We provide a comprehensive convergence analysis as well as extensive experiments, and the results show that our approach achieves state-of-the-art performance across various model architectures (including ResNet18, ResNet34, ResNet50, ResNet56, and MobileNetv2) on standard benchmark datasets (CIFAR-10, CIFAR-100, and ImageNet).

Shangqian Gao, Ting Hua, Reza Shirkavand et al.

Large Language Models (LLMs) have demonstrated remarkable abilities in tackling a wide range of complex tasks. However, their huge computational and memory costs raise significant challenges in deploying these models on resource-constrained devices or efficiently serving them. Prior approaches have attempted to alleviate these problems by permanently removing less important model structures, yet these methods often result in substantial performance degradation due to the permanent deletion of model parameters. In this work, we tried to mitigate this issue by reducing the number of active parameters without permanently removing them. Specifically, we introduce a differentiable dynamic pruning method that pushes dense models to maintain a fixed number of active parameters by converting their MLP layers into a Mixture of Experts (MoE) architecture. Our method, even without fine-tuning, consistently outperforms previous structural pruning techniques across diverse model families, including Phi-2, LLaMA-2, LLaMA-3, and Qwen-2.5.

Zeyu Zhang, Haiying Shen, Shay Vargaftik et al.

Disaggregated Large Language Model (LLM) inference has gained popularity as it separates the computation-intensive prefill stage from the memory-intensive decode stage, avoiding the prefill-decode interference and improving resource utilization. However, transmitting Key-Value (KV) data between the two stages can be a bottleneck, especially for long prompts. Additionally, the computation time overhead for prefill and decode is key for optimizing Job Completion Time (JCT), and KV data size can become prohibitive for long prompts and sequences. Existing KV quantization methods can alleviate the transmission bottleneck and reduce memory requirements, but they introduce significant dequantization overhead, exacerbating the computation time. We propose Homomorphic Acceleration via Compression of the KV cache (HACK) for disaggregated LLM inference. HACK eliminates the heavy KV dequantization step, and directly performs computations on quantized KV data to approximate and reduce the cost of the expensive matrix-multiplication step. Extensive trace-driven experiments show that HACK reduces JCT by up to 70.9% compared to disaggregated LLM inference baseline and by up to 52.3% compared to state-of-the-art KV quantization methods.