亚洲男人的天堂2018av,欧美草比,久久久久久免费视频精选,国色天香在线看免费,久久久久亚洲av成人片仓井空

Convolutional Neural Networks (CNNs) demonstrate great performance in various applications but have high computational complexity. Quantization is applied to reduce the latency and storage cost of CNNs. Among the quantization methods, Binary and Ternary Weight Networks (BWNs and TWNs) have a unique advantage over 8-bit and 4-bit quantization. They replace the multiplication operations in CNNs with additions, which are favoured on In-Memory-Computing (IMC) devices. IMC acceleration for BWNs has been widely studied. However, though TWNs have higher accuracy and better sparsity, IMC acceleration for TWNs has limited research. TWNs on existing IMC devices are inefficient because the sparsity is not well utilized, and the addition operation is not efficient. In this paper, we propose FAT as a novel IMC accelerator for TWNs. First, we propose a Sparse Addition Control Unit, which utilizes the sparsity of TWNs to skip the null operations on zero weights. Second, we propose a fast addition scheme based on the memory Sense Amplifier to avoid the time overhead of both carry propagation and writing back the carry to the memory cells. Third, we further propose a Combined-Stationary data mapping to reduce the data movement of both activations and weights and increase the parallelism of memory columns. Simulation results show that for addition operations at the Sense Amplifier level, FAT achieves 2.00X speedup, 1.22X power efficiency and 1.22X area efficiency compared with State-Of-The-Art IMC accelerator ParaPIM. FAT achieves 10.02X speedup and 12.19X energy efficiency compared with ParaPIM on networks with 80% sparsity

Training the deep convolutional neural network for computer vision problems is slow and inefficient, especially when it is large and distributed across multiple devices. The inefficiency is caused by the backpropagation algorithm's forward locking, backward locking, and update locking problems. Existing solutions for acceleration either can only handle one locking problem or lead to severe accuracy loss or memory inefficiency. Moreover, none of them consider the straggler problem among devices. In this paper, we propose Layer-wise Staleness and a novel efficient training algorithm, Diversely Stale Parameters (DSP), to address these challenges. We also analyze the convergence of DSP with two popular gradient-based methods and prove that both of them are guaranteed to converge to critical points for non-convex problems. Finally, extensive experimental results on training deep learning models demonstrate that our proposed DSP algorithm can achieve significant training speedup with stronger robustness than compared methods.

Deep Reinforcement Learning (DRL) solutions are becoming pervasive at the edge of the network as they enable autonomous decision-making in a dynamic environment. However, to be able to adapt to the ever-changing environment, the DRL solution implemented on an embedded device has to continue to occasionally take exploratory actions even after initial convergence. In other words, the device has to occasionally take random actions and update the value function, i.e., re-train the Artificial Neural Network (ANN), to ensure its performance remains optimal. Unfortunately, embedded devices often lack processing power and energy required to train the ANN. The energy aspect is particularly challenging when the edge device is powered only by a means of Energy Harvesting (EH). To overcome this problem, we propose a two-part algorithm in which the DRL process is trained at the sink. Then the weights of the fully trained underlying ANN are periodically transferred to the EH-powered embedded device taking actions. Using an EH-powered sensor, real-world measurements dataset, and optimizing for Age of Information (AoI) metric, we demonstrate that such a DRL solution can operate without any degradation in the performance, with only a few ANN updates per day.

Artificial intelligence (AI) is gaining success and importance in many different tasks. The growing pervasiveness and complexity of AI systems push researchers towards developing dedicated hardware accelerators. Spiking Neural Networks (SNN) represent a promising solution in this sense since they implement models that are more suitable for a reliable hardware design. Moreover, from a neuroscience perspective, they better emulate a human brain. This work presents the development of a hardware accelerator for an SNN, with off-line training, applied to an image recognition task, using the MNIST as the target dataset. Many techniques are used to minimize the area and to maximize the performance, such as the replacement of the multiplication operation with simple bit shifts and the minimization of the time spent on inactive spikes, useless for the update of neurons' internal state. The design targets a Xilinx Artix-7 FPGA, using in total around the 40% of the available hardware resources and reducing the classification time by three orders of magnitude, with a small 4.5% impact on the accuracy, if compared to its software, full precision counterpart.

The deep neural network suffers from many fundamental issues in machine learning. For example, it often gets trapped into a local minimum in training, and its prediction uncertainty is hard to be assessed. To address these issues, we propose the so-called kernel-expanded stochastic neural network (K-StoNet) model, which incorporates support vector regression (SVR) as the first hidden layer and reformulates the neural network as a latent variable model. The former maps the input vector into an infinite dimensional feature space via a radial basis function (RBF) kernel, ensuring absence of local minima on its training loss surface. The latter breaks the high-dimensional nonconvex neural network training problem into a series of low-dimensional convex optimization problems, and enables its prediction uncertainty easily assessed. The K-StoNet can be easily trained using the imputation-regularized optimization (IRO) algorithm. Compared to traditional deep neural networks, K-StoNet possesses a theoretical guarantee to asymptotically converge to the global optimum and enables the prediction uncertainty easily assessed. The performances of the new model in training, prediction and uncertainty quantification are illustrated by simulated and real data examples.

Federated learning (FL) offers a decentralized learning environment so that a group of clients can collaborate to train a global model at the server, while keeping their training data confidential. This paper studies how to launch over-the-air jamming attacks to disrupt the FL process when it is executed over a wireless network. As a wireless example, FL is applied to learn how to classify wireless signals collected by clients (spectrum sensors) at different locations (such as in cooperative sensing). An adversary can jam the transmissions for the local model updates from clients to the server (uplink attack), or the transmissions for the global model updates the server to clients (downlink attack), or both. Given a budget imposed on the number of clients that can be attacked per FL round, clients for the (uplink/downlink) attack are selected according to their local model accuracies that would be expected without an attack or ranked via spectrum observations. This novel attack is extended to general settings by accounting different processing speeds and attack success probabilities for clients. Compared to benchmark attack schemes, this attack approach degrades the FL performance significantly, thereby revealing new vulnerabilities of FL to jamming attacks in wireless networks.

State-of-the-art language models (LMs) represented by long-short term memory recurrent neural networks (LSTM-RNNs) and Transformers are becoming increasingly complex and expensive for practical applications. Low-bit neural network quantization provides a powerful solution to dramatically reduce their model size. Current quantization methods are based on uniform precision and fail to account for the varying performance sensitivity at different parts of LMs to quantization errors. To this end, novel mixed precision neural network LM quantization methods are proposed in this paper. The optimal local precision choices for LSTM-RNN and Transformer based neural LMs are automatically learned using three techniques. The first two approaches are based on quantization sensitivity metrics in the form of either the KL-divergence measured between full precision and quantized LMs, or Hessian trace weighted quantization perturbation that can be approximated efficiently using matrix free techniques. The third approach is based on mixed precision neural architecture search. In order to overcome the difficulty in using gradient descent methods to directly estimate discrete quantized weights, alternating direction methods of multipliers (ADMM) are used to efficiently train quantized LMs. Experiments were conducted on state-of-the-art LF-MMI CNN-TDNN systems featuring speed perturbation, i-Vector and learning hidden unit contribution (LHUC) based speaker adaptation on two tasks: Switchboard telephone speech and AMI meeting transcription. The proposed mixed precision quantization techniques achieved "lossless" quantization on both tasks, by producing model size compression ratios of up to approximately 16 times over the full precision LSTM and Transformer baseline LMs, while incurring no statistically significant word error rate increase.

State-of-the-art neural language models represented by Transformers are becoming increasingly complex and expensive for practical applications. Low-bit deep neural network quantization techniques provides a powerful solution to dramatically reduce their model size. Current low-bit quantization methods are based on uniform precision and fail to account for the varying performance sensitivity at different parts of the system to quantization errors. To this end, novel mixed precision DNN quantization methods are proposed in this paper. The optimal local precision settings are automatically learned using two techniques. The first is based on a quantization sensitivity metric in the form of Hessian trace weighted quantization perturbation. The second is based on mixed precision Transformer architecture search. Alternating direction methods of multipliers (ADMM) are used to efficiently train mixed precision quantized DNN systems. Experiments conducted on Penn Treebank (PTB) and a Switchboard corpus trained LF-MMI TDNN system suggest the proposed mixed precision Transformer quantization techniques achieved model size compression ratios of up to 16 times over the full precision baseline with no recognition performance degradation. When being used to compress a larger full precision Transformer LM with more layers, overall word error rate (WER) reductions up to 1.7% absolute (18% relative) were obtained.

Spiking neural networks (SNNs) can utilize spatio-temporal information and have a nature of energy efficiency which is a good alternative to deep neural networks(DNNs). The event-driven information processing makes SNNs can reduce the expensive computation of DNNs and save a lot of energy consumption. However, high training and inference latency is a limitation of the development of deeper SNNs. SNNs usually need tens or even hundreds of time steps during the training and inference process which causes not only the increase of latency but also the waste of energy consumption. To overcome this problem, we proposed a novel training method based on backpropagation (BP) for ultra-low latency(1-2 time steps) SNN with multi-threshold. In order to increase the information capacity of each spike, we introduce the multi-threshold Leaky Integrate and Fired (LIF) model. In our proposed training method, we proposed three approximated derivative for spike activity to solve the problem of the non-differentiable issue which cause difficulties for direct training SNNs based on BP. The experimental results show that our proposed method achieves an average accuracy of 99.56%, 93.08%, and 87.90% on MNIST, FashionMNIST, and CIFAR10, respectively with only 2 time steps. For the CIFAR10 dataset, our proposed method achieve 1.12% accuracy improvement over the previously reported direct trained SNNs with fewer time steps.

For deploying a deep learning model into production, it needs to be both accurate and compact to meet the latency and memory constraints. This usually results in a network that is deep (to ensure performance) and yet thin (to improve computational efficiency). In this paper, we propose an efficient method to train a deep thin network with a theoretic guarantee. Our method is motivated by model compression. It consists of three stages. In the first stage, we sufficiently widen the deep thin network and train it until convergence. In the second stage, we use this well-trained deep wide network to warm up (or initialize) the original deep thin network. This is achieved by letting the thin network imitate the immediate outputs of the wide network from layer to layer. In the last stage, we further fine tune this well initialized deep thin network. The theoretical guarantee is established by using mean field analysis, which shows the advantage of layerwise imitation over traditional training deep thin networks from scratch by backpropagation. We also conduct large-scale empirical experiments to validate our approach. By training with our method, ResNet50 can outperform ResNet101, and BERT_BASE can be comparable with BERT_LARGE, where both the latter models are trained via the standard training procedures as in the literature.