Can Jin, Rui Wu, Tong Che, Qixin Zhang, Hongwu Peng, Jiahui Zhao, Zhenting Wang, Wenqi Wei, Ligong Han, Zhao Zhang, Yuan Cao, Ruixiang Tang, Dimitris N. Metaxas. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Traditional verification methods in chip design are highly time-consuming and computationally demanding, especially for large scale circuits. Graph neural networks (GNNs) have gained popularity as a potential solution to improve verification efficiency. However, there lacks a joint framework that considers all chip design domain knowledge, graph theory, and GPU kernel designs. To address this challenge, we introduce GROOT, an algorithm and system co-design framework that contains chip design domain knowledge and redesigned GPU kernels, to improve verification efficiency. More specifically, we create node features utilizing the circuit node types and the polarity of the connections between the input edges to nodes in And-Inverter Graphs (AIGs). We utilize a graph partitioning algorithm to divide the large graphs into smaller sub-graphs for fast GPU processing and develop a graph edge re-growth algorithm to recover verification accuracy. We carefully profile the EDA graph workloads and observe the uniqueness of their polarized distribution of high degree (HD) nodes and low degree (LD) nodes. We redesign two GPU kernels (HD-kernel and LD-kernel), to fit the EDA graph learning workload on a single GPU. We compare the results with state-of-the-art (SOTA) methods: GAMORA, a GNN-based approach, and the traditional ABC framework. Results show that GROOT achieves a significant reduction in memory footprint (59.38 %), with high accuracy (99.96%) for a very large CSA multiplier, i.e. 1,024 bits with a batch size of 16, which consists of 134,103,040 nodes and 268,140,544 edges. We compare GROOT with GPU-based GPU Kernel designs SOTAs such as cuSPARSE, MergePath-SpMM, and GNNAdvisor. We achieve up to 1.104×, 5.796×, and 1.469× improvement in runtime, respectively.

Abstract Top-k selection algorithms are fundamental in a wide range of applications, including high-performance computing, information retrieval, big data processing, and neural network model training. In this paper, we present RTop-K, a highly efficient parallel row-wise top-k selection algorithm specifically designed for GPUs. RTop-K leverages a binary search-based approach to optimize row-wise top-k selection, providing a scalable and accelerated solution. We conduct a detailed analysis of early stopping in our algorithm, showing that it effectively maintains the testing accuracy of neural network models while substantially improving performance. Our GPU implementation of RTop-K demonstrates superior performance over state-of-the-art row-wise top-k GPU implementations, achieving an average speed-up of up to 11.49× with early stopping and 7.29× without early stopping. Moreover, RTop-K accelerates the overall training workflow of MaxK-GNNs, delivering speed-ups ranging from 11.97% to 33.29% across different models and datasets.

Graph Convolutional Network (GCN) has gained popularity as it could lower the human expert's burden in making tactical real-time decisions.As Moore's law is reaching an end, the acceleration of the conventional GCN systems is limited.One promising alternative is the Adiabatic Quantum-Flux-Parametron (AQFP) superconducting computing as it can achieve extremely high energy efficiency compared to CMOS.In this paper, we propose an AQFP-aware GCN acceleration framework via co-optimizing AQFP hardware and GCN algorithms.More specifically, we first develop a regrowth-after-partitioning algorithm to enable the AQFP hardware parallelism and accelerate the aggregation computation while maintaining accuracy.Then, we propose two distinct AQFP-based architectures tailored specifically for each of the combination and aggregation stages.Furthermore, to unlock the extreme energy efficiency, we develop a hybrid binarized/low-bit GCN hardware/software co-design that can be efficiently executed on AQFP-based devices.Leveraging the AQFP randomized behavior, we adjust the AQFP buffer design to achieve multi-bit intermediate results and explore the bit-width at the output of the combination step.

The increasing scale and complexity of integrated circuit design have led to increased challenges in Electronic Design Automation (EDA).Graph Neural Networks (GNNs), have emerged as a promising approach to assist EDA design as circuits can be naturally represented as graph.While GNNs offer a foundation for circuit analysis, they often fail

In the acceleration of deep neural network training, the graphics processing unit (GPU) has become the mainstream platform. GPUs face substantial challenges on Graph Neural Networks (GNNs), such as workload imbalance and memory access irregularities, leading to underutilized hardware. Existing solutions such as PyG, DGL with cuSPARSE, and GNNAdvisor frameworks partially address these challenges. However, the memory traffic involved with Sparse-Dense Matrix Matrix Multiplication (SpMM) is still significant.

Large Language Models (LLMs) employ auto-regressive decoding that requires sequential computation, with each step reliant on the previous one's output. This creates a bottleneck as each step necessitates moving the full model parameters from High-Bandwidth Memory (HBM) to the accelerator's cache. While methods such as speculative decoding have been suggested to address this issue, their implementation is impeded by the challenges associated with acquiring and maintaining a separate draft model. In this paper, we present Medusa, an efficient method that augments LLM inference by adding extra decoding heads to predict multiple subsequent tokens in parallel. Using a tree-based attention mechanism, Medusa constructs multiple candidate continuations and verifies them simultaneously in each decoding step. By leveraging parallel processing, Medusa reduces the number of decoding steps required. We present two levels of fine-tuning procedures for Medusa to meet the needs of different use cases: Medusa-1: Medusa is directly fine-tuned on top of a frozen backbone LLM, enabling lossless inference acceleration. Medusa-2: Medusa is fine-tuned together with the backbone LLM, enabling better prediction accuracy of Medusa heads and higher speedup but needing a special training recipe that preserves the model's capabilities. Moreover, we propose several extensions that improve or expand the utility of Medusa, including a self-distillation to handle situations where no training data is available and a typical acceptance scheme to boost the acceptance rate while maintaining generation quality. We evaluate Medusa on models of various sizes and training procedures. Our experiments demonstrate that Medusa-1 can achieve over 2.2$\times$ speedup without compromising generation quality, while Medusa-2 further improves the speedup to 2.3-2.8$\times$.

Generalization remains a central challenge in machine learning. In this work, we propose *Learning from Teaching* (**LoT**), a novel regularization technique for deep neural networks to enhance generalization. Inspired by the human ability to capture concise and abstract patterns, we hypothesize that generalizable correlations are expected to be easier to imitate. LoT operationalizes this concept to improve the generalization of the main model with auxiliary student learners. The student learners are trained by the main model and, in turn, provide feedback to help the main model capture more generalizable and imitable correlations. Our experimental results across several domains, including Computer Vision, Natural Language Processing, and methodologies like Reinforcement Learning, demonstrate that the introduction of LoT brings significant benefits compared to training models on the original dataset. The results suggest the effectiveness and efficiency of LoT in identifying generalizable information at the right scales while discarding spurious data correlations, thus making LoT a valuable addition to current machine learning. Code is available at https://github.com/jincan333/LoT.

Over-parameterization of deep neural networks (DNNs) has shown high prediction accuracy for many applications. Although effective, the large number of parameters hinders its popularity on resource-limited devices and has an outsize environmental impact. Sparse training (using a fixed number of nonzero weights in each iteration) could significantly mitigate the training costs by reducing the model size. However, existing sparse training methods mainly use either random-based or greedy-based drop-and-grow strategies, resulting in local minimal and low accuracy. In this work, to assist explainable sparse training, we propose important weights Exploitation and coverage Exploration to characterize Dynamic Sparse Training (DST-EE), and provide quantitative analysis of these two metrics. We further design an acquisition function and provide the theoretical guarantees for the proposed method and clarify its convergence property. Experimental results show that sparse models (up to 98% sparsity) obtained by our proposed method outperform the SOTA sparse training methods on a wide variety of deep learning tasks. On VGG-19 / CIFAR-100, ResNet-50 / CIFAR-10, ResNet-50 / CIFAR-100, our method has even higher accuracy than dense models. On ResNet-50 / ImageNet, the proposed method has up to 8.2% accuracy improvement compared to SOTA sparse training methods.

Two-party computation (2PC) is promising to enable privacy-preserving deep learning (DL). However, the 2PC-based privacy-preserving DL implementation comes with high comparison protocol overhead from the non-linear operators. This work presents PASNet, a novel systematic framework that enables low latency, high energy efficiency & accuracy, and security-guaranteed 2PC-DL by integrating the hardware latency of the cryptographic building block into the neural architecture search loss function. We develop a cryptographic hardware scheduler and the corresponding performance model for Field Programmable Gate Arrays (FPGA) as a case study. The experimental results demonstrate that our light-weighted model PASNet-A and heavily-weighted model PASNet-B achieve 63 ms and 228 ms latency on private inference on ImageNet, which are 147 and 40 times faster than the SOTA CryptGPU system, and achieve 70.54% & 78.79% accuracy and more than 1000 times higher energy efficiency. The pretrained PASNet models and test code can be found on Github1.

Graph Convolutional Networks (GCNs) are pivotal in extracting latent information from graph data across various domains, yet their acceleration on mainstream GPUs is challenged by workload imbalance and memory access irregularity. To address these challenges, we present Accel-GCN, a GPU accelerator architecture for GCNs. The design of Accel-GCN encompasses: (i) a lightweight degree sorting stage to group nodes with similar degree; (ii) a block-level partition strategy that dynamically adjusts warp workload sizes, enhancing shared memory locality and workload balance, and reducing metadata overhead compared to designs like GNNAdvisor; (iii) a combined warp strategy that improves memory coalescing and computational parallelism in the column dimension of dense matrices. Utilizing these principles, we formulate a kernel for SpMM in GCNs that employs block-level partitioning and combined warp strategy. This approach augments performance and multi-level memory efficiency and optimizes memory bandwidth by exploiting memory coalescing and alignment. Evaluation of Accel-GCN across 18 benchmark graphs reveals that it outperforms cuSPARSE, GNNAdvisor, and graph-BLAST by factors of 1.17×, 1.86×, and 2.94× respectively. The results underscore Accel-GCN as an effective solution for enhancing GCN computational efficiency. The implementation can be found on Github*.

The growth of the Machine-Learning-As-A-Service (MLaaS) market has highlighted clients’ data privacy and security issues. Private inference (PI) techniques using cryptographic primitives offer a solution but often have high computation and communication costs, particularly with non-linear operators like ReLU. Many attempts to reduce ReLU operations exist, but they may need heuristic threshold selection or cause substantial accuracy loss. This work introduces AutoReP, a gradient-based approach to lessen non-linear operators and alleviate these issues. It automates the selection of ReLU and polynomial functions to speed up PI applications and introduces distribution-aware polynomial approximation (DaPa) to maintain model expressivity while accurately approximating ReLUs. Our experimental results demonstrate significant accuracy improvements of 6.12% (94.31%, 12.9K ReLU budget, CIFAR-10), 8.39% (74.92%, 12.9K ReLU budget, CIFAR-100), and 9.45% (63.69%, 55K ReLU budget, Tiny-ImageNet) over current state-of-the-art methods, e.g., SNL. Morever, AutoReP is applied to EfficientNet-B2 on ImageNet dataset, and achieved 75.55% accuracy with 176.1 × ReLU budget reduction. The codes are shared on Github1.

The growing prevalence of Machine Learning as a Service (MLaaS) enables a wide range of applications but simultaneously raises numerous security and privacy concerns. A key issue involves the potential privacy exposure of involved parties, such as the customer’s input data and the vendor’s model. Consequently, two-party computing (2PC) has emerged as a promising solution to safeguard the privacy of different parties during deep neural network (DNN) inference. However, the state-of-the-art (SOTA) 2PC-DNN techniques are tailored explicitly to traditional instruction set architecture (ISA) systems like CPUs and CPU+GPU. This reliance on ISA systems significantly constrains their energy efficiency, as these architectures typically employ 32- or 64-bit instruction sets. In contrast, the possibilities of harnessing dynamic and adaptive quantization to build high-performance 2PC-DNNs remain largely unexplored due to the lack of compatible algorithms and hardware accelerators.

The growth of Graph Convolution Network (GCN) model sizes has revolutionized numerous applications, surpassing human performance in areas such as personal healthcare and financial systems. The deployment of GCNs in the cloud raises privacy concerns due to potential adversarial attacks on client data. To address security concerns, Privacy-Preserving Machine Learning (PPML) using Homomorphic Encryption (HE) secures sensitive client data. However, it introduces substantial computational overhead in practical applications. To tackle those challenges, we present LinGCN, a framework designed to reduce multiplication depth and optimize the performance of HE based GCN inference. LinGCN is structured around three key elements: (1) A differentiable structural linearization algorithm, complemented by a parameterized discrete indicator function, co-trained with model weights to meet the optimization goal. This strategy promotes fine-grained node-level non-linear location selection, resulting in a model with minimized multiplication depth. (2) A compact node-wise polynomial replacement policy with a second-order trainable activation function, steered towards superior convergence by a two-level distillation approach from an all-ReLU based teacher model. (3) an enhanced HE solution that enables finer-grained operator fusion for node-wise activation functions, further reducing multiplication level consumption in HE-based inference. Our experiments on the NTU-XVIEW skeleton joint dataset reveal that LinGCN excels in latency, accuracy, and scalability for homomorphically encrypted inference, outperforming solutions such as CryptoGCN. Remarkably, LinGCN achieves a 14.2× latency speedup relative to CryptoGCN, while preserving an inference accuracy of ~75\% and notably reducing multiplication depth. Additionally, LinGCN proves scalable for larger models, delivering a substantial 85.78\% accuracy with 6371s latency, a 10.47\% accuracy improvement over CryptoGCN.

Transformers are considered one of the most important deep learning models since 2018, in part because it establishes state-of-the-art (SOTA) records and could potentially replace existing Deep Neural Networks (DNNs). Despite the remarkable triumphs, the prolonged turnaround time of Transformer models is a widely recognized roadblock. The variety of sequence lengths imposes additional computing overhead where inputs need to be zero-padded to the maximum sentence length in the batch to accommodate the parallel computing platforms. This paper targets the field-programmable gate array (FPGA) and proposes a coherent sequence length adaptive algorithm-hardware co-design for Transformer acceleration. Particularly, we develop a hardware-friendly sparse attention operator and a length-aware hardware resource scheduling algorithm. The proposed sparse attention operator brings the complexity of attention-based models down to linear complexity and alleviates the off-chip memory traffic. The proposed length-aware resource hardware scheduling algorithm dynamically allocates the hardware resources to fill up the pipeline slots and eliminates bubbles for NLP tasks. Experiments show that our design has very small accuracy loss and has 80.2 × and 2.6 × speedup compared to CPU and GPU implementation, and 4 × higher energy efficiency than state-of-the-art GPU accelerator optimized via CUBLAS GEMM.

Graph Neural Networks (GCNs) have attracted wide attention and are applied to the real world. However, due to the ever-growing graph data with significant irregularities, off-chip communication with poor data locality has become the major bottleneck hurdling the development of GCNs. Fortunately, recent works demonstrate Resistive Random Access Memory (ReRAM) has the potential to perform inherently parallel in-situ computation of Matrix-Vector Multiplication (MVM) in the analog regime fundamentally breaking the communication bottleneck.Inspired by this observation, we propose a novel ReRAM-based GCN acceleration co-design (i.e. algorithm-hardware) framework, CoDG-ReRAM, that can deliver real-time GCN inference with high accuracy. On the algorithm side, we propose a novel model optimization pipeline that simultaneously and efficiently sparsifies and regularizes both graph and parameter matrices in GCNs and creates ReRAM-friendly models. On the hardware side, we take advantage of the software optimization results to provide a more systematic mapping scheme and in-crease computation efficiency to have an energy-efficient ReRAM-based GCN acceleration with low latency. Experimental results show that the proposed work improves performance and energy efficiency by 4× and 5.1 × respectively over SOTA ReRAM-based accelerators of GCNs with a negligible accuracy loss.

As real-world graphs expand in size, larger GNN models with billions of parameters are deployed. High parameter count in such models makes training and inference on graphs expensive and challenging. To reduce the computational and memory costs of GNNs, optimization methods such as pruning the redundant nodes and edges in input graphs have been commonly adopted. However, model compression, which directly targets the sparsification of model layers, has been mostly limited to traditional Deep Neural Networks (DNNs) used for tasks such as image classification and object detection. In this paper, we utilize two state-of-the-art model compression methods (1) train and prune and (2) sparse training for the sparsification of weight layers in GNNs. We evaluate and compare the efficiency of both methods in terms of accuracy, training sparsity, and training FLOPs on real-world graphs. Our experimental results show that on the ia-email, wiki-talk, and stackoverflow datasets for link prediction, sparse training with much lower training FLOPs achieves a comparable accuracy with the train and prune method. On the brain dataset for node classification, sparse training uses a lower number FLOPs Oess than 1/7 FLOPs of train and prune method) and preserves a much better accuracy performance under extreme model sparsity. Our model sparsification code is publicly available on GitHubl1.

Being able to learn from complex data with phase information is imperative for many signal processing applications. Today’s real-valued deep neural networks (DNNs) have shown efficiency in latent information analysis but fall short when applied to the complex domain. Deep complex networks (DCN), in contrast, can learn from complex data, but have high computational costs; therefore, they cannot satisfy the instant decision-making requirements of many deployable systems dealing with short observations or short signal bursts. Recent, Binarized Complex Neural Network (BCNN), which integrates DCNs with binarized neural networks (BNN), shows great potential in classifying complex data in real-time. In this paper, we propose a structural pruning based accelerator of BCNN, which is able to provide more than 5000 frames/s inference throughput on edge devices. The high performance comes from both the algorithm and hardware sides. On the algorithm side, we conduct structural pruning to the original BCNN models and obtain 20 × pruning rates with negligible accuracy loss; on the hardware side, we propose a novel 2D convolution operation accelerator for the binary complex neural network. Experimental results show that the proposed design works with over 90% utilization and is able to achieve the inference throughput of 5882 frames/s and 4938 frames/s for complex NIN-Net and ResNet-18 using CIFAR-10 dataset and Alveo U280 Board.

Although Transformer-based deep learning models have been widely used in many natural language processing (NLP) tasks as well as computer vision, they suffer from gigantic model size and long latency. Network pruning can reduce the computational cost and model size. However, existing works mainly focus on irregular(sparse) pruning, which often causes irregular computations and extra indices per remained weight. In this work, we propose a Tensor-core inspired hierarchical model compression method to push the performance limit on modern GPUs. We present two modes of the two-step process. In the first mode, we use the Tensor-core aware block-based weight pruning method to exploit model sparsity in a coarse-grained manner and then use low-rank [33] decomposition to further reduce the weight storage in a fine-grained manner.In the second mode, we first use irregular pruning to achieve a highly sparse model and then apply the Tensor-core aware weight constraint on the sparse model to decompose the sparse matrix to several smaller but Tensor-core friendly sub-matrices. Experiments on Transformer, BERTBASE models show the proposed method outperforms the state-of-the-art.

Recently, Transformers gradually gain popularity and perform outstanding for many Natural Language Processing (NLP) tasks. However, Transformers suffer from heavy computation and memory footprint, making it difficult to deploy on embedded devices. The field-programmable gate array (FPGA) is widely used to accelerate deep learning algorithms for its advantages. However, the trained Transformer models are too large to accommodate to an FPGA fabric. To accommodate Transformer onto FPGA and achieve efficient execution, we propose an acceleration framework coupling the balanced model compression at the algorithm level and FPGA-implementation optimization at the hardware level. At algorithm level, we adopt a block-balanced pruning and propose an efficient sparse matrix storage format for this pruning technique, named Compressed Block Row (CBR). At the hardware level, we design an accelerator for sparse model. And we also abstract a performance analytic model to evaluate the performance of accelerator. Experiments show that our CBR format perform better than general formats and can significantly save storage space. And our accelerator can achieve $38\times$ and $1.93\times$ speedup compared to other works on CPU and GPU respectively.

Molecular similarity search has been widely used in drug discovery to identify structurally similar compounds from large molecular databases rapidly. With the increasing size of chemical libraries, there is growing interest in the efficient acceleration of large-scale similarity search. Existing works mainly focus on CPU and GPU to accelerate the computation of the Tanimoto coefficient in measuring the pairwise similarity between different molecular fingerprints. In this paper, we propose and optimize an FPGA-based accelerator design on exhaustive and approximate search algorithms. On exhaustive search using BitBound & folding, we analyze the similarity cutoff and folding level relationship with search speedup and accuracy, and propose a scalable on-the-fly query engine on FPGAs to reduce the resource utilization and pipeline interval. We achieve a 450 million compounds-per-second processing throughput for a single query engine. On approximate search using hierarchical navigable small world (HNSW), a popular algorithm with high recall and query speed. We propose an FPGA-based graph traversal engine to utilize a high throughput register array based priority queue and fine-grained distance calculation engine to increase the processing capability. Experimental results show that the proposed FPGA-based HNSW implementation has a 103385 query per second (QPS) on the Chembl database with 0.92 recall and achieves a 35x speedup than the existing CPU implementation on average. To the best of our knowledge, our FPGA-based implementation is the first attempt to accelerate molecular similarity search algorithms on FPGA and has the highest performance among existing approaches.

State-of-the-art Transformer-based models, with gigantic parameters, are difficult to be accommodated on resource constrained embedded devices. Moreover, with the development of technology, more and more embedded devices are available to run a Transformer model. For a Transformer model with different constraints (tight or loose), it can be deployed onto devices with different computing power. However, in previous work, designers did not choose the best device among multiple devices. Instead, they just used an existing device to deploy model, which was not necessarily the best fit and may lead to underutilization of resources. To address the deployment challenge of Transformer and the problem to select the best device, we propose an algorithm$\leftrightarrows$hardware closed-loop acceleration framework. Given a dataset, a model, latency constraint LC and accuracy constraint AC, our framework can provide a best device satisfying both constraints. In order to generate a compressed model with high sparsity ratio, we propose a novel pruning technique, hierarchical pruning (HP). We optimize the sparse matrix storage format for HP matrix to further reduce memory usage for FPGA implementation. We design a accelerator that takes advantage of HP to solve the problem of concurrent random access. Experiments on Transformer and TinyBert model show that our framework can find different devices for various LC and AC, covering from low-end devices to high-end devices. Our HP can achieve higher sparsity ratio and is more flexible than other sparsity pattern. Our framework can achieve 37 x, 1.9 x, 1.7x speedup compared to CPU, GPU and FPGA, respectively.