Adaptive network pruning approach has recently drawn significant attention due to its excellent capability to identify the importance and redundancy of layers and filters and customize a suitable pruning solution. However, it remains unsatisfactory since current adaptive pruning methods rely mostly on an additional monitor to score layer and filter importance, and thus faces high complexity and weak interpretability. To tackle these issues, we have deeply researched the weight reconstruction process in iterative prune-train process and propose a Protective Self-Adaptive Pruning (PSAP) method. First of all, PSAP can utilize its own information, weight sparsity ratio, to adaptively adjust pruning ratio of layers before each pruning step. Moreover, we propose a protective reconstruction mechanism to prevent important filters from being pruned through supervising gradients and to avoid unrecoverable information loss as well. Our PSAP is handy and explicit because it merely depends on weights and gradients of model itself, instead of requiring an additional monitor as in early works. Experiments on ImageNet and CIFAR-10 also demonstrate its superiority to current works in both accuracy and compression ratio, especially for compressing with a high ratio or pruning from scratch.
相關內容
Automated hand gesture recognition has been a focus of the AI community for decades. Traditionally, work in this domain revolved largely around scenarios assuming the availability of the flow of images of the user hands. This has partly been due to the prevalence of camera-based devices and the wide availability of image data. However, there is growing demand for gesture recognition technology that can be implemented on low-power devices using limited sensor data instead of high-dimensional inputs like hand images. In this work, we demonstrate a hand gesture recognition system and method that uses signals from capacitive sensors embedded into the etee hand controller. The controller generates real-time signals from each of the wearer five fingers. We use a machine learning technique to analyse the time series signals and identify three features that can represent 5 fingers within 500 ms. The analysis is composed of a two stage training strategy, including dimension reduction through principal component analysis and classification with K nearest neighbour. Remarkably, we found that this combination showed a level of performance which was comparable to more advanced methods such as supervised variational autoencoder. The base system can also be equipped with the capability to learn from occasional errors by providing it with an additional adaptive error correction mechanism. The results showed that the error corrector improve the classification performance in the base system without compromising its performance. The system requires no more than 1 ms of computing time per input sample, and is smaller than deep neural networks, demonstrating the feasibility of agile gesture recognition systems based on this technology.
Originally introduced as a neural network for ensemble learning, mixture of experts (MoE) has recently become a fundamental building block of highly successful modern deep neural networks for heterogeneous data analysis in several applications, including those in machine learning, statistics, bioinformatics, economics, and medicine. Despite its popularity in practice, a satisfactory level of understanding of the convergence behavior of Gaussian-gated MoE parameter estimation is far from complete. The underlying reason for this challenge is the inclusion of covariates in the Gaussian gating and expert networks, which leads to their intrinsically complex interactions via partial differential equations with respect to their parameters. We address these issues by designing novel Voronoi loss functions to accurately capture heterogeneity in the maximum likelihood estimator (MLE) for resolving parameter estimation in these models. Our results reveal distinct behaviors of the MLE under two settings: the first setting is when all the location parameters in the Gaussian gating are non-zeros while the second setting is when there exists at least one zero-valued location parameter. Notably, these behaviors can be characterized by the solvability of two different systems of polynomial equations. Finally, we conduct a simulation study to verify our theoretical results.
Many problems in robotics, such as estimating the state from noisy sensor data or aligning two LiDAR point clouds, can be posed and solved as least-squares problems. Unfortunately, vanilla nonminimal solvers for least-squares problems are notoriously sensitive to outliers. As such, various robust loss functions have been proposed to reduce the sensitivity to outliers. Examples of loss functions include pseudo-Huber, Cauchy, and Geman-McClure. Recently, these loss functions have been generalized into a single loss function that enables the best loss function to be found adaptively based on the distribution of the residuals. However, even with the generalized robust loss function, most nonminimal solvers can only be solved locally given a prior state estimate due to the nonconvexity of the problem. The first contribution of this paper is to combine graduated nonconvexity (GNC) with the generalized robust loss function to solve least-squares problems without a prior state estimate and without the need to specify a loss function. Moreover, existing loss functions, including the generalized loss function, are based on Gaussian-like distribution. However, residuals are often defined as the squared norm of a multivariate error and distributed in a Chi-like fashion. The second contribution of this paper is to apply a norm-aware adaptive robust loss function within a GNC framework. This leads to additional robustness when compared with state-of-the-art methods. Simulations and experiments demonstrate that the proposed approach is more robust and yields faster convergence times compared to other GNC formulations.
In order to satisfy their ever increasing capacity and compute requirements, many machine learning models are distributed across multiple nodes using space-efficient parallelism strategies. As a result, collective communications are often on the critical path, and hiding their latency by overlapping kernel-granular communication and computation is difficult due to the absence of independent computation. In this work, we propose fusing computation with communication using GPU-initiated networking, and leverage GPUs' massive parallelism to enable fine-grained overlap of the fused operations. We have developed a single, self-contained GPU kernel where workgroups (WGs) immediately communicate their results to remote GPUs when they complete their computation. Meanwhile, other WGs within the same kernel perform overlapping computation, maintaining high ALU utilization. Furthermore, we propose zero-copy optimizations for peer-to-peer GPU communication where the data computed by one GPU is directly written to the destination buffers within the peer GPUs, eliminating intermediate stores and extra buffering. Our approach leverages the emerging multi-node GPU system trend where GPUs are physically close to network with direct GPU-NIC interconnects. We demonstrate our approach by creating an embedding + All-to-All fused kernel which overlaps embedding operations and the dependent all-to-all collective in DLRM models. We evaluate our approach both using simulation and real hardware. Our evaluations show that our approach can effectively overlap All-to-All communication with embedding computations, subsequently reducing their combined execution time by 31% on average (up to 58%) for inter-node and by 25% (up to 35%) for intra-node configurations. Scale-out simulations indicate that our approach reduces DLRM execution time by ~10% for 128 node system.
Federated learning (FL), which addresses data privacy issues by training models on resource-constrained mobile devices in a distributed manner, has attracted significant research attention. However, the problem of optimizing FL client selection in mobile federated learning networks (MFLNs), where devices move in and out of each others' coverage and no FL server knows all the data owners, remains open. To bridge this gap, we propose a first-of-its-kind \underline{Soc}ially-aware \underline{Fed}erated \underline{C}lient \underline{S}election (SocFedCS) approach to minimize costs and train high-quality FL models. SocFedCS enriches the candidate FL client pool by enabling data owners to propagate FL task information through their local networks of trust, even as devices are moving into and out of each others' coverage. Based on Lyapunov optimization, we first transform this time-coupled problem into a step-by-step optimization problem. Then, we design a method based on alternating minimization and self-adaptive global best harmony search to solve this mixed-integer optimization problem. Extensive experiments comparing SocFedCS against five state-of-the-art approaches based on four real-world multimedia datasets demonstrate that it achieves 2.06\% higher test accuracy and 12.24\% lower cost on average than the best-performing baseline.
Different from conventional federated learning, personalized federated learning (PFL) is able to train a customized model for each individual client according to its unique requirement. The mainstream approach is to adopt a kind of weighted aggregation method to generate personalized models, in which weights are determined by the loss value or model parameters among different clients. However, such kinds of methods require clients to download others' models. It not only sheer increases communication traffic but also potentially infringes data privacy. In this paper, we propose a new PFL algorithm called \emph{FedDWA (Federated Learning with Dynamic Weight Adjustment)} to address the above problem, which leverages the parameter server (PS) to compute personalized aggregation weights based on collected models from clients. In this way, FedDWA can capture similarities between clients with much less communication overhead. More specifically, we formulate the PFL problem as an optimization problem by minimizing the distance between personalized models and guidance models, so as to customize aggregation weights for each client. Guidance models are obtained by the local one-step ahead adaptation on individual clients. Finally, we conduct extensive experiments using five real datasets and the results demonstrate that FedDWA can significantly reduce the communication traffic and achieve much higher model accuracy than the state-of-the-art approaches.
Federated Learning (FL) is a widely embraced paradigm for distilling artificial intelligence from distributed mobile data. However, the deployment of FL in mobile networks can be compromised by exposure to interference from neighboring cells or jammers. Existing interference mitigation techniques require multi-cell cooperation or at least interference channel state information, which is expensive in practice. On the other hand, power control that treats interference as noise may not be effective due to limited power budgets, and also that this mechanism can trigger countermeasures by interference sources. As a practical approach for protecting FL against interference, we propose Spectrum Breathing, which cascades stochastic-gradient pruning and spread spectrum to suppress interference without bandwidth expansion. The cost is higher learning latency by exploiting the graceful degradation of learning speed due to pruning. We synchronize the two operations such that their levels are controlled by the same parameter, Breathing Depth. To optimally control the parameter, we develop a martingale-based approach to convergence analysis of Over-the-Air FL with spectrum breathing, termed AirBreathing FL. We show a performance tradeoff between gradient-pruning and interference-induced error as regulated by the breathing depth. Given receive SIR and model size, the optimization of the tradeoff yields two schemes for controlling the breathing depth that can be either fixed or adaptive to channels and the learning process. As shown by experiments, in scenarios where traditional Over-the-Air FL fails to converge in the presence of strong interference, AirBreahing FL with either fixed or adaptive breathing depth can ensure convergence where the adaptive scheme achieves close-to-ideal performance.
Many networks in political and social research are bipartite, with edges connecting exclusively across two distinct types of nodes. A common example includes cosponsorship networks, in which legislators are connected indirectly through the bills they support. Yet most existing network models are designed for unipartite networks, where edges can arise between any pair of nodes. We show that using a unipartite network model to analyze bipartite networks, as often done in practice, can result in aggregation bias. To address this methodological problem, we develop a statistical model of bipartite networks by extending the popular mixed-membership stochastic blockmodel. Our model allows researchers to identify the groups of nodes, within each node type, that share common patterns of edge formation. The model also incorporates both node and dyad-level covariates as the predictors of the edge formation patterns. We develop an efficient computational algorithm for fitting the model, and apply it to cosponsorship data from the United States Senate. We show that senators tapped into communities defined by party lines and seniority when forming cosponsorships on bills, while the pattern of cosponsorships depends on the timing and substance of legislations. We also find evidence for norms of reciprocity, and uncover the substantial role played by policy expertise in the formation of cosponsorships between senators and legislation. An open-source software package is available for implementing the proposed methodology.
The time and effort involved in hand-designing deep neural networks is immense. This has prompted the development of Neural Architecture Search (NAS) techniques to automate this design. However, NAS algorithms tend to be slow and expensive; they need to train vast numbers of candidate networks to inform the search process. This could be alleviated if we could partially predict a network's trained accuracy from its initial state. In this work, we examine the overlap of activations between datapoints in untrained networks and motivate how this can give a measure which is usefully indicative of a network's trained performance. We incorporate this measure into a simple algorithm that allows us to search for powerful networks without any training in a matter of seconds on a single GPU, and verify its effectiveness on NAS-Bench-101, NAS-Bench-201, NATS-Bench, and Network Design Spaces. Our approach can be readily combined with more expensive search methods; we examine a simple adaptation of regularised evolutionary search. Code for reproducing our experiments is available at //github.com/BayesWatch/nas-without-training.
The growing energy and performance costs of deep learning have driven the community to reduce the size of neural networks by selectively pruning components. Similarly to their biological counterparts, sparse networks generalize just as well, if not better than, the original dense networks. Sparsity can reduce the memory footprint of regular networks to fit mobile devices, as well as shorten training time for ever growing networks. In this paper, we survey prior work on sparsity in deep learning and provide an extensive tutorial of sparsification for both inference and training. We describe approaches to remove and add elements of neural networks, different training strategies to achieve model sparsity, and mechanisms to exploit sparsity in practice. Our work distills ideas from more than 300 research papers and provides guidance to practitioners who wish to utilize sparsity today, as well as to researchers whose goal is to push the frontier forward. We include the necessary background on mathematical methods in sparsification, describe phenomena such as early structure adaptation, the intricate relations between sparsity and the training process, and show techniques for achieving acceleration on real hardware. We also define a metric of pruned parameter efficiency that could serve as a baseline for comparison of different sparse networks. We close by speculating on how sparsity can improve future workloads and outline major open problems in the field.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191