Search

Channel Pruning is one of the most widespread techniques used to compress deep neural networks while maintaining their performances. Currently, a typical pruning algorithm leverages neural architecture search to directly find networks with a configurable width, the key step of which is to identify representative subnet for various pruning ratios by training a supernet. However, current methods mainly follow a serial training strategy to optimize supernet, which is very time-consuming. In this work, we introduce PSE-Net, a novel parallel-subnets estimator for efficient channel pruning. Specifically, we propose a parallel-subnets training algorithm that simulate the forward-backward pass of multiple subnets by droping extraneous features on batch dimension, thus various subnets could be trained in one round. Our proposed algorithm facilitates the efficiency of supernet training and equips the network with the ability to interpolate the accuracy of unsampled subnets, enabling PSE-Net to effectively evaluate and rank the subnets. Over the trained supernet, we develop a prior-distributed-based sampling algorithm to boost the performance of classical evolutionary search. Such algorithm utilizes the prior information of supernet training phase to assist in the search of optimal subnets while tackling the challenge of discovering samples that satisfy resource constraints due to the long-tail distribution of network configuration. Extensive experiments demonstrate PSE-Net outperforms previous state-of-the-art channel pruning methods on the ImageNet dataset while retaining superior supernet training efficiency. For example, under 300M FLOPs constraint, our pruned MobileNetV2 achieves 75.2% Top-1 accuracy on ImageNet dataset, exceeding the original MobileNetV2 by 2.6 units while only cost 30%/16% times than BCNet/AutoAlim., Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (Copyright © 2024 Elsevier Ltd. All rights reserved.)

Tennis is a sport that is enjoyed by young and old alike. Participants can decide the style of play and control the speed and spin of the ball according to their ability, so the physical demands on participants are not high and the amount of exercise can be effectively controlled. With the increasing demand for quality of life, the population of tennis is growing rapidly. People have taken up tennis as a form of recreation. For professional players, the serve, as the foundation of the introduction to tennis, has been a challenge. A good serve can effectively control the situation on the court and even turn defeat into victory. Therefore, this paper focuses on capturing tennis trajectories using multidimensional wavelet segmentation in an attempt to provide an effective reference for tennis. The results show that our method has a good trajectory capturing effect and can correct and guide the tennis serving action well. [ABSTRACT FROM AUTHOR]

In wireless sensor and actor networks (WSANs), a set of static sensor nodes and a set of (mobile) actor nodes form a network that performs distributed sensing and actuation tasks. In [1], Abbasi et al. presented DARA, a Distributed Actor Recovery Algorithm, which restores the connectivity of the interactor network by efficiently relocating some mobile actors when failure of an actor happens. To restore 1 and 2-connectivity of the network, two algorithms are developed in [1]. Their basic idea is to find the smallest set of actors that needs to be repositioned to restore the required level of connectivity, with the objective to minimize the movement overhead of relocation. Here, we show that the algorithms proposed in [1] will not work smoothly in all scenarios as claimed and give counterexamples for some algorithms and theorems proposed in [1]. We then present a general actor relocation problem and propose methods that will work correctly for several subsets of the problems. Specifically, our method does result in an optimum movement strategy with minimum movement overhead for the problems studied in [1]. [ABSTRACT FROM AUTHOR]

Data aggregation is a key functionality in wireless sensor networks (WSNs). This paper focuses on data aggregation scheduling problem to minimize the delay (or latency). We propose an efficient distributed algorithm that produces a collision-free schedule for data aggregation in WSNs. We theoretically prove that the delay of the aggregation schedule generated by our algorithm is at most 16R+\Delta -14 time slots. Here, R is the network radius and \Delta is the maximum node degree in the communication graph of the original network. Our algorithm significantly improves the previously known best data aggregation algorithm with an upper bound of delay of 24D+6\Delta +16 time slots, where D is the network diameter (note that D can be as large as 2R). We conduct extensive simulations to study the practical performances of our proposed data aggregation algorithm. Our simulation results corroborate our theoretical results and show that our algorithms perform better in practice. We prove that the overall lower bound of delay for data aggregation under any interference model is max \\log n, R\, where n is the network size. We provide an example to show that the lower bound is (approximately) tight under the protocol interference model when r_I=r, where r_I is the interference range and r is the transmission range. We also derive the lower bound of delay under the protocol interference model when r