General matrix-matrix multiplication (GEMM) is a fundamental operation in both deep learning and scientific computing. To accelerate these workloads, GPUs with a large number of streaming multiprocessors (SMs) are widely used. However, as modern GPUs scale in core count and adopt larger tile sizes, the wave quantization problem induced by partially filled waves results in growing hardware underutilization and substantially degraded performance. Existing solutions to this problem often suffer from low execution efficiency or introduce additional synchronization overhead.

Operator fusion, a key technique to improve data locality and alleviate GPU memory bandwidth pressure, often fails to extend to the fusion of multiple compute-intensive operators due to saturated computation throughput. However, the dynamicity of tensor dimension sizes could potentially lead to these operators becoming memory-bound, necessitating the generation of fused kernels — a task hindered by limited search spaces for fusion strategies, redundant memory access, and prolonged tuning time, leading to sub-optimal performance and inefficient deployment. We introduce MCFuser, a pioneering framework designed to overcome these obstacles by generating high-performance fused kernels for what we define as memory-bound compute-intensive (MBCI) operator chains. Leveraging high-level tiling expressions to delineate a comprehensive search space, coupled with Directed Acyclic Graph (DAG) analysis to eliminate redundant memory accesses, MCFuser streamlines kernel optimization. By implementing guidelines to prune the search space and incorporating an analytical performance model with a heuristic search, MCFuser not only significantly accelerates the tuning process but also demonstrates superior performance. Benchmarked against leading compilers like Ansor on NVIDIA A100 and RTX3080 GPUs, MCFuser achieves up to a 5.9x speedup in kernel performance and outpaces other baselines while reducing tuning time by over $\mathbf{7 0}$-fold, showcasing its agility.

With the rapid evolution of the Internet, Internet of Things (IoT), and geographic information systems (GIS), spatio-temporal Big Data (STBD) is experiencing exponential growth, marking the onset of the STBD era. Recent studies have concentrated on developing algorithms and techniques for the collection, management, storage, processing, analysis, and visualization of STBD. Researchers have made significant advancements by enhancing STBD handling techniques, creating novel systems, and integrating spatio-temporal support into existing systems. However, these studies often neglect resource management and system optimization, crucial factors for enhancing the efficiency of STBD processing and applications. Additionally, the transition of STBD to the innovative Cloud-Edge-End unified computing system needs to be noticed. In this survey, we comprehensively explore the entire ecosystem of STBD analytics systems. We delineate the STBD analytics ecosystem and categorize the technologies used to process GIS data into five modules: STBD, computation resources, processing platform, resource management, and applications. Specifically, we subdivide STBD and its applications into geoscience-oriented and human-social activity-oriented. Within the processing platform module, we further categorize it into the data management layer (DBMS-GIS), data processing layer (BigData-GIS), data analysis layer (AI-GIS), and cloud native layer (Cloud-GIS). The resource management module and each layer in the processing platform are classified into three categories: task-oriented, resource-oriented, and cloud-based. Finally, we propose research agendas for potential future developments.

Transformer-based models have made significant advancements across various domains, largely due to the self-attention mechanism’s ability to capture contextual relationships in input sequences. However, processing long sequences remains computationally expensive for Transformer models, primarily due to theO(n2) complexity associated with self-attention. To address this, sparse attention has been proposed to reduce the quadratic dependency to linear. Nevertheless, deploying the sparse transformer efficiently encounters two major obstacles: 1) Existing system optimizations are less effective for the sparse transformer due to the algorithm’s approximation properties leading to fragmented attention, and 2) the variability of input sequences results in computation and memory access inefficiencies. We present Raptor-T, a cutting-edge transformer framework designed for handling long and variable-length sequences. Raptor-T harnesses the power of the sparse transformer to reduce resource requirements for processing long sequences while also implementing system-level optimizations to accelerate inference performance. To address the fragmented attention issue, Raptor-T employs fused and memory-efficient Multi-Head Attention. Additionally, we introduce an asynchronous data processing method to mitigate GPU-blocking operations caused by sparse attention. Furthermore, Raptor-T minimizes padding for variable-length inputs, effectively reducing the overhead associated with padding and achieving balanced computation on GPUs. In evaluation, we compare Raptor-T’s performance against state-of-the-art frameworks on an NVIDIA A100 GPU. The experimental results demonstrate that Raptor-T outperforms FlashAttention-2 and FasterTransformer, achieving an impressive average end-to-end performance improvement of 3.41X and 3.71X, respectively.

Recently, Temporal Graph Neural Networks (TGNNs), as an extension of Graph Neural Networks, have demonstrated remarkable effectiveness in handling dynamic graph data. Distributed TGNN training requires efficiently tackling temporal dependency, which often leads to excessive cross-device communication that generates significant redundant data. However, existing systems are unable to remove the redundancy in data reuse and transfer, and suffer from severe communication overhead in a distributed setting. This work introduces Sven, a co-designed algorithm-system library aimed at accelerating TGNN training on a multi-GPU platform. Exploiting dependency patterns of TGNN models, we develop a redundancy-free graph organization to mitigate redundant data transfer. Additionally, we investigate communication imbalance issues among devices and formulate the graph partitioning problem as minimizing the maximum communication balance cost, which is proved to be an NP-hard problem. We propose an approximation algorithm called Re-FlexBiCut to tackle this problem. Furthermore, we incorporate prefetching, adaptive micro-batch pipelining, and asynchronous pipelining to present a hierarchical pipelining mechanism that mitigates the communication overhead. Sven represents the first comprehensive optimization solution for scaling memory-based TGNN training. Through extensive experiments conducted on a 64-GPU cluster, Sven demonstrates impressive speedup, ranging from 1.9x to 3.5x, compared to State-of-the-Art approaches. Additionally, Sven achieves up to 5.26x higher communication efficiency and reduces communication imbalance by up to 59.2%.

In recent years, the Mixture-of-Experts (MoE) technique has gained widespread popularity as a means to scale pretrained models to exceptionally large sizes. Dynamic activation of experts allows for conditional computation, increasing the number of parameters of neural networks, which is critical for absorbing the vast amounts of knowledge available in many deep learning areas. However, despite the existing system and algorithm optimizations, there are significant challenges to be tackled when it comes to the inefficiencies of communication and memory consumption. In this paper, we present the design and implementation of MPMoE, a high-performance library that accelerates MoE training with adaptive and memory-efficient pipeline parallelism. Inspired by that the MoE training procedure can be divided into multiple independent sub-stages. We design a pipeline parallelism method for reducing communication latency by overlapping with computation operations. Further, we analyze the memory footprint breakdown of MoE training and identify that activations and temporary buffers are the primary contributors to the overall memory footprint. Toward memory efficiency, we propose memory reuse strategies to reduce memory requirements by eliminating memory redundancies. Finally, to optimize pipeline granularity and memory reuse strategies jointly, we propose a profile-based algorithm and a performance model to determine the configurations of MPMoE at runtime. We implement MPMoE upon PyTorch and evaluate it with common MoE models in two physical clusters, including 64 NVIDIA A100 GPU cards and 16 NVIDIA V100 GPU cards. Compared with the state-of-art approach, MPMoE achieves up to 2.3× speedup while reducing more than 30% memory footprint for training large models.

Temporal Graph Neural Networks(TGNNs) extend the success of Graph Neural Networks to dynamic graphs. Distributed TGNN training requires efficiently tackling temporal dependency, which often leads to excessive cross-device communication that generates significant redundant data. However, existing systems are unable to remove the redundancy in data reuse and transfer, and suffer from severe communication overhead in a distributed setting. This paper presents Sven, an algorithm and system co-designed TGNN training library for the end-to-end performance optimization on multi-node multi-GPU systems. Exploiting dependency patterns of TGNN models and characteristics of dynamic graph datasets, we design redundancy-free data organization and load-balancing partitioning strategies that mitigate the redundant data communication and evenly partition dynamic graphs at the vertex level. Furthermore, we develop a hierarchical pipeline mechanism integrating data prefetching, micro-batch pipelining, and asynchronous pipelining to mitigate the communication overhead. As the first scaling study on the memory-based TGNNs training, experiments conducted on an HPC cluster of 64 GPUs show that Sven can achieve up to 1.7x-3.3x speedup over the state-of-art approaches and a factor of up to 5.26x communication efficiency improvement.

Recently, Mixture-of-Experts (MoE) has become one of the most popular techniques to scale pre-trained models to extraordinarily large sizes. Dynamic activation of experts allows for conditional computation, increasing the number of parameters of neural networks, which is critical for absorbing the vast amounts of knowledge available in many deep learning areas. However, despite the existing system and algorithm optimizations, there are significant challenges to be tackled when it comes to the inefficiencies of communication and memory consumption.In this paper, we present the design and implementation of MPipeMoE, a high-performance library that accelerates MoE training with adaptive and memory-efficient pipeline parallelism. Inspired by that the MoE training procedure can be divided into multiple independent sub-stages, we design adaptive pipeline parallelism with an online algorithm to configure the granularity of the pipelining. Further, we analyze the memory footprint breakdown of MoE training and identify that activations and temporary buffers are the primary contributors to the overall memory footprint. Toward memory efficiency, we propose memory reusing strategies to reduce memory requirements by eliminating memory redundancies, and develop an adaptive selection component to determine the optimal strategy that considers both hardware capacities and model characteristics at runtime. We implement MPipeMoE upon PyTorch and evaluate it with common MoE models in a physical cluster consisting of 8 NVIDIA DGX A100 servers. Compared with the state-of-art approach, MPipeMoE achieves up to 2.8× speedup and reduces memory footprint by up to 47% in training large models.

Hash coding has been widely used in the approximate nearest neighbor search for large-scale image retrieval. Given semantic annotations such as class labels and pairwise similarities of the training data, hashing methods can learn and generate effective and compact binary codes. While some newly introduced images may contain undefined semantic labels, which we call unseen images, zero-shot hashing (ZSH) techniques have been studied for retrieval. However, existing ZSH methods mainly focus on the retrieval of single-label images and cannot handle multilabel ones. In this article, for the first time, a novel transductive ZSH method is proposed for multilabel unseen image retrieval. In order to predict the labels of the unseen/target data, a visual-semantic bridge is built via instance-concept coherence ranking on the seen/source data. Then, pairwise similarity loss and focal quantization loss are constructed for training a hashing model using both the seen/source and unseen/target data. Extensive evaluations on three popular multilabel data sets demonstrate that the proposed hashing method achieves significantly better results than the comparison methods.

Hash coding has been widely used in the approximate nearest neighbor search for large-scale image retrieval. Recently, many deep hashing methods have been proposed and shown largely improved performance over traditional feature-learning methods. Most of these methods examine the pairwise similarity on the semantic-level labels, where the pairwise similarity is generally defined in a hard-assignment way. That is, the pairwise similarity is “1” if they share no less than one class label and “0” if they do not share any. However, such similarity definition cannot reflect the similarity ranking for pairwise images that hold multiple labels. In this paper, an improved deep hashing method is proposed to enhance the ability of multi-label image retrieval. We introduce a pairwise quantified similarity calculated on the normalized semantic labels. Based on this, we divide the pairwise similarity into two situations-“hard similarity” and “soft similarity,” where cross-entropy loss and mean square error loss are adapted respectively for more robust feature learning and hash coding. Experiments on four popular datasets demonstrate that the proposed method outperforms the competing methods and achieves the state-of-the-art performance in multi-label image retrieval.

Cracks are typical line structures that are of interest in many computer-vision applications. In practice, many cracks, e.g., pavement cracks, show poor continuity and low contrast, which brings great challenges to image-based crack detection by using low-level features. In this paper, we propose DeepCrack - an end-to-end trainable deep convolutional neural network for automatic crack detection by learning high-level features for crack representation. In this method, multi-scale deep convolutional features learned at hierarchical convolutional stages are fused together to capture the line structures. More detailed representations are made in larger-scale feature maps and more holistic representations are made in smaller-scale feature maps. We build DeepCrack net on the encoder-decoder architecture of SegNet, and pairwisely fuse the convolutional features generated in the encoder network and in the decoder network at the same scale. We train DeepCrack net on one crack dataset and evaluate it on three others. The experimental results demonstrate that DeepCrack achieves F-Measure over 0.87 on the three challenging datasets in average and outperforms the current state-of-the-art methods.