The Preconditioned Conjugate Gradient (PCG) algorithm is a widely-used iterative method for solving sparse linear systems. The sparse matrix-vector multiplication (SpMV) operation involved in the PCG algorithm dominates the computing cost of each iteration. To accelerate the PCG computation on multi-GPU systems, the input sparse matrix has to be partitioned on different GPUs. In addition, an appropriate sparse format must be used for each partition. In this paper, we propose a new multi-GPU PCG implementation that relies on horizontally partitioning the input matrix into multiple block-rows so that different sparse formats can be used at a low granularity. We then use a mapping algorithm based on the Minimum Completion Time (MCT) policy to assign different block-rows to the GPUs available in the system. Our experimental results with real-world large sparse matrices reveal a noticeable performance improvement, especially for the sparse systems characterized by irregular sparsity structure.

Recently, deep neural networks have been adopted in many application domains showing impressive performances that sometimes even exceed those of humans. However, the considerable size of deep learning models and their computational cost limits their use in mobile and embedded systems. Pruning is an important technique that has been used in the literature to compress a trained model by removing some of its elements (weights, neurons, filters, etc.) that do not notably contribute to its performance. In this paper, we propose a new block pruning method for residual networks that removes holes layers of the network based on the Multi-Armed Bandits (MAB) framework. The proposed method takes advantage of the existence of multiple trainable paths in residual models to identify and prune the less important residual blocks for inference. To evaluate our approach, we conducted extensive experiments using three different MAB algorithms, namely ε-greedy, Win Stay Lose Shift (WSLS), and the Upper Confidence Bound (UCB). The obtained results on different datasets show that our block pruning method can considerably reduce the model size and FLOPs with little to no impact on overall accuracy. We conducted a comparative analysis of our method with three existing techniques to demonstrate that our approach is highly competitive in terms of accuracy, while significantly reducing both model size and FLOPs. Moreover, our experimental results demonstrate that when our block pruning method is combined with a filter-level pruning technique, it leads to even greater reductions in model size and FLOPs while maintaining a relatively high accuracy.

Deep learning models have demonstrated remarkable success in various applications, yet their vulnerability to adversarial attacks remains a significant concern. These attacks can mislead models, imperceptibly to human eyes, creating a critical challenge in ensuring robustness. Despite recent advancements in adversarial attacks that contribute to enhancing model robustness, many existing techniques yield higher perturbation rates, lower textual similarity or lower success rates, with some, like population-based methods, incurring an increased query count. In response to that, this paper introduces two innovative methods: a k-means-based ranking approach and an iterative context-aware search algorithm complemented by a rollback method, to enhance the quality of generated adversarial samples. Our approaches showcase superiority over numerous state-of-the-art techniques by successfully compromising deep learning models with fewer modifications and achieving higher success rates, presenting a significant advancement in adversarial attack generation. This work contributes to the ongoing efforts to fortify deep learning models against adversarial attacks.

The Sparse Matrix-Vector Multiplication (SpMV) kernel dominates the computing cost in numerous scientific applications. Many implementations based on different sparse formats were proposed to improve this kernel on the recent GPU architectures. However, it has been widely observed that there is no “best-for-all” sparse format for the SpMV kernel on GPU. Indeed, serious performance degradation of an order of magnitude can be observed without a careful selection of the sparse format to use. To address this problem, we propose in this article BestSF (Best Sparse Format), a new learning-based sparse meta-format that automatically selects the most appropriate sparse format for a given input matrix. To do so, BestSF relies on a cost-sensitive classification system trained using Weighted Support Vector Machines (WSVMs) to predict the best sparse format for each input sparse matrix. Our experimental results on two different NVIDIA GPU architectures using a large number of real-world sparse matrices show that BestSF achieved a noticeable overall performance improvement over using a single sparse format. While BestSF is trained to select the best sparse format in terms of performance (GFLOPS), our further experimental investigations revealed that using BestSF also led, in most of the test cases, to the best energy efficiency (MFLOPS/W). To prove its practical effectiveness, we also evaluate the performance and energy efficiency improvement achieved when using BestSF as a building block in a GPU-based Preconditioned Conjugate Gradient (PCG) iterative solver.

Sparse Matrix-Vector Multiplication (SpMV) kernel dominates the computing cost in numerous scientific applications. Many implementations based on different sparse formats were proposed recently for optimizing this kernel on the GPU side. Since the performance of the SpMV varies significantly according to the sparsity characteristics of the input matrix and the hardware features, developing an accurate performance model for this kernel is a challenging task. The traditional approach of building such models by analytical modeling is difficult in practice and requires a thorough understanding of the interaction between the GPU hardware and the sparse code. In this paper, we propose to use a machine learning approach to predict the performance of the SpMV kernel using several sparse formats (COO, CSR, ELL, and HYB) on GPU. We used two popular machine learning algorithms, Support Vector Regression (SVR) and Multilayer Perceptron neural network (MLP). Our experimental results on two different GPUs (Fermi GTX 512 and Maxwell GTX 980 Ti) show that the SVR models deliver the best accuracy with average prediction error ranging between 7% and 14%.

Sparse Matrix-Vector Multiplication (SpMV) kernel dominates the computing cost in numerous scientific applications. Many implementations based on different sparse formats were proposed recently for this kernel on the GPU side. Since the performance of these sparse formats varies significantly according to the sparsity characteristics of the input matrix and the hardware specifications, no one of them can be considered as the best one to use for every sparse matrix. In this paper, we address the problem of selecting the best representation for a given sparse matrix on GPU by using a machine learning approach. First, we present some interesting and easy to compute features for characterizing the sparse matrices on GPU. Second, we use a multiclass Support Vector Machine (SVM) classifier to select the best format for each input matrix. We consider in this paper four popular formats (COO, CSR, ELL, and HYB), but our work can be extended to support more sparse representations. Experimental results on two different GPUs (Fermi GTX 580 and Maxwell GTX 980 Ti) show that we achieved more than 98% of the performance possible with a perfect selection.