检索结果-内蒙古大学图书馆

CONVERGENCE OF BACKPRIOPAG ATION WITH MOMENTUM FOR NETWORK A RCHITECTURES WITH SKIP CONNECTIONS

学校读者我要写书评

暂无评论

Journal of Computational Mathematics 2021年第1期39卷 147-158页

作者： Chirag Agarwal Joe Klobusicky Dan Schonfeld Department of Electrical and Cormputer Engineering University of Illinois at ChicagoChicagoIL 60607USA Department of Mathematical Sciences Rensselaer Polytechnic InstituteTroyNY 12180USA Department of Electrical and Computer Engineering University of Illinois at ChicagoChicagoIL 60607USA

We study a class of deep neural networks with architectures that form a directed acyclic graph(DAG).For backpropagation defined by gradient descent with adaptive momentum,we show weights converge for a large class of nonlinear activation functions.'The proof generalizes the results of Wu et al.(2008)who showed convergence for a feed-forward network with one hidden *** an example of the effectiveness of DAG architectures,we describe an example of compression through an AutoEncoder,and compare against sequential feed-forward networks under several metrics.

关键词： Backpropagation with momentum Autoencoders Directed acyclic graphs.

Optimized Control of IoT-Monitored Microgrid Systems Using Genetic Algorithm

学校读者我要写书评

暂无评论

Optimized Control of IoT-Monitored Microgrid Systems Using G...

2023 International Conference on Power Energy, Environment and Intelligent Control, PEEIC 2023

作者： Vijayalakshmi, N. Yuvaraj, S. Baskar, V. Vijaya Krishnaswamy, R. Sasikala, K. Srinivasan, C. Government Polytechnic College Deputed from Annamalai University Department of Electrical and Electronics Engineering Tamilnadu Coimbatore India Srm Institute of Science and Technology Faculty of Engineering and Technology Department of Electronics and Communication Engineering Tamil Nadu Chennai India School of Electrical and Electronics Engineering Sathyabama Institute of Science and Technology Tamil Nadu Chennai India University College of Engineering Department of Electronics and Communication Engineering Tamil Nadu Ariyalur India Department of Electrical and Electronics Engineering Tamil Nadu Chennai India Saveetha School of Engineering Saveetha Institute of Medical and Technical Sciences Saveetha University Department of Computer Science and Engineering Tamil Nadu Chennai India

ISBN: (纸本)9798350357769

Microgrid (MG) systems have become a good way to make power generation and distribution more efficient, especially when Distributed Energy Resources (DERs) are introduced. The inverter is a critical part of MGs since it regulates the voltage output to the grid. A unique method is proposed to tune the settings of a Proportional-Integral-Derivative (PID) controller for the inverter using Genetic Algorithm (GA). The main aim is to use GAs to improve the inverter control in an MG that is monitored by the Internet of Things (IoT). GA's are very useful for resolving difficult optimization issues. Optimization of voltage regulation, stability, and power transmission to the grid may be achieved by using GA-based tuning of PID controller settings. From the simulation results, the GA-based methodology provides better control performance than conventional tuning approaches, providing exact voltage regulation and improving the MG's stability and dependability. © 2023 IEEE.

关键词： Genetic algorithm and Optimization technique Microgrid Renewable energy Voltage control

A Provably Effective Method for Pruning Experts in Fine-tuned Sparse Mixture-of-Experts

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Chowdhury, Mohammed Nowaz Rabbani Wang, Meng Maghraoui, Kaoutar El Wang, Naigang Chen, Pin-Yu Carothers, Christopher Department of Electrical Computer and Systems Engineering Rensselaer Polytechnic Institute NY United States IBM Research Yorktown HeightsNY United States Department of Computer Science Rensselaer Polytechnic Institute NY United States

The sparsely gated mixture of experts (MoE) architecture sends different inputs to different subnetworks (experts), through trainable routers. MoE reduces the training computation significantly for large models, but its deployment can be still memory/computation expensive for some downstream tasks. Model pruning is a popular approach to reduce inference computation, but its application in MoE architecture is largely unexplored. To the best of our knowledge, this paper provides the first provably efficient technique for pruning experts in fine-tuned MoE models. We theoretically prove that prioritizing the pruning of the experts with a smaller change of the router’s l2 norm from the pre-trained model guarantees the preservation of test accuracy, while significantly reducing the model size and the computational requirements. Although our theoretical analysis is centered on binary classification tasks on simplified MoE architecture, our expert pruning method is verified on large vision MoE models such as V-MoE and E3-MoE fine-tuned on benchmark datasets such as CIFAR-10, CIFAR-100, and ImageNet. Copyright © 2024, The Authors. All rights reserved.

关键词： Large datasets

Hands-On Learning: Teaching Parallel and Distributed Computing through Unplugged Activities in Undergraduate CS Courses 24

学校读者我要写书评

暂无评论

Hands-On Learning: Teaching Parallel and Distributed Computi...

Proceedings of the SC '24 Workshops of the International Conference on High Performance Computing, Network, Storage, and Analysis

作者： Anurag Dasgupta Venkat Margapuri Simon Shamoun Shubbhi Taneja Matthew Toups Computer Science and Engineering Technology Valdosta State University Valdosta GA Department of Computing Sciences Villanova University Villanova PA Department of Computer Science Hofstra University Hempstead NY Department of Computer Science Worcester Polytechnic Institute Worcester Massachusetts Computer Science Department Tulane University New Orleans LA

ISBN: (纸本)9798350355543

The authors present and evaluate an unplugged activity to introduce parallel computing concepts to undergraduate students. Students in five CS classrooms used a deck of playing cards in small groups to consider how parallelization can improve performance and how improvement decreases with increased parallelization. Before and after the activity, students took a short survey about their solution and their ideas about parallelism. The authors carried out this activity in seven courses at five institutions in the 2023-2024 academic year. The results showed that students had an increased appreciation for parallelization and this type of activity.

关键词： CS

Reinforcement Learning of CPG-regulated Locomotion Controller for a Soft Snake Robot

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Liu, Xuan Onal, Cagdas D. Fu, Jie The Robotics Engineering Department Worcester Polytechnic Institute WorcesterMA United States The Department of Electrical and Computer Engineering University of Florida GainesvilleFL United States

Intelligent control of soft robots is challenging due to the nonlinear and difficult-to-model dynamics. One promising model-free approach for soft robot control is reinforcement learning (RL). However, model-free RL methods tend to be computationally expensive and data-inefficient and may not yield natural and smooth locomotion patterns for soft robots. In this work, we develop a bio-inspired design of a learning-based goal-tracking controller for a soft snake robot. The controller is composed of two modules: An RL module for learning goal-tracking behaviors given the unmodeled and stochastic dynamics of the robot, and a central pattern generator (CPG) with the Matsuoka oscillators for generating stable and diverse locomotion patterns. We theoretically investigate the maneuverability of Matsuoka CPG’s oscillation bias, frequency, and amplitude for steering control, velocity control, and sim-to-real adaptation of the soft snake robot. Based on this analysis, we proposed a composition of RL and CPG modules such that the RL module regulates the tonic inputs to the CPG system given state feedback from the robot, and the output of the CPG module is then transformed into pressure inputs to pneumatic actuators of the soft snake robot. This design allows the RL agent to naturally learn to entrain the desired locomotion patterns determined by the CPG maneuverability. We validated the optimality and robustness of the control design in both simulation and real experiments, and performed extensive comparisons with state-of-art RL methods to demonstrate the benefit of our bio-inspired control design. © 2022, CC BY-NC-SA.

关键词： Reinforcement learning

Impact of the Antenna on the Sub-Terahertz Indoor Channel Characteristics: An Experimental Approach

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Sen, Priyangshu Badran, Sherif Petrov, Vitaly Singh, Arjun Jornet, Josep M. Department of Engineering SUNY Polytechnic Institute UticaNY United States Department of Electrical and Computer Engineering Northeastern University BostonMA United States

Terahertz-band (100 GHz-10 THz) communication is a promising radio technology envisioned to enable ultrahigh data rate, reliable and low-latency wireless connectivity in next-generation wireless systems. However, the low transmission power of THz transmitters, the need for high gain directional antennas, and the complex interaction of THz radiation with common objects along the propagation path make crucial the understanding of the THz channel. In this paper, we conduct an extensive channel measurement campaign in an indoor setting (i.e., a conference room) through a channel sounder with 0.1 ns time resolution and 20 GHz bandwidth at 140 GHz. Particularly, the impact of different antenna directivities (and, thus, beam widths) on the channel characteristics is extensively studied. The experimentally obtained dataset is processed to develop the path loss model and, subsequently, derive key channel metrics such as the path loss exponent, delay spread, and K-factor. The results highlight the multi-faceted impact of the antenna gain on the channel and, by extension, the wireless system and, thus, show that an antenna-agnostic channel model cannot capture the propagation characteristics of the THz channel. © 2024, CC BY.

关键词： Directive antennas

Robust 3D Point Clouds Classification based on Declarative Defenders

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Li, Kaidong Zhang, Tianxiao Zhong, Chuncong Zhang, Ziming Wang, Guanghui Department of Electrical Engineering & Computer Science University of Kansas 1520 West 15th St. LawrenceKS66045 United States Department of Electrical & Computer Engineering Worcester Polytechnic Institute 100 Institute Rd. WorcesterMA01609 United States Department of Computer Science Toronto Metropolitan University 350 Victoria Street TorontoONM5B 2K3 Canada

3D point cloud classification requires distinct models from 2D image classification due to the divergent characteristics of the respective input data. While 3D point clouds are unstructured and sparse, 2D images are structured and dense. Bridging the domain gap between these two data types is a non-trivial challenge to enable model interchangeability. Recent research using Lattice Point Classifier (LPC) highlights the feasibility of cross-domain applicability. However, the lattice projection operation in LPC generates 2D images with disconnected projected pixels. In this paper, we explore three distinct algorithms for mapping 3D point clouds into 2D images. Through extensive experiments, we thoroughly examine and analyze their performance and defense mechanisms. Leveraging current large foundation models, we scrutinize the feature disparities between regular 2D images and projected 2D images. The proposed approaches demonstrate superior accuracy and robustness against adversarial attacks. The generative model-based mapping algorithms yield regular 2D images, further minimizing the domain gap from regular 2D classification tasks. The source code is available at https://***/KaidongLi/***. © 2024, CC BY.

关键词： Generative adversarial networks

Robust Model Predictive Control for Onboard Interleaved Multilevel Boost DC-DC Converter in DC Shipboard Microgrids

学校读者我要写书评

暂无评论

IEEE Transactions on Transportation Electrification 2025年

作者： Li, Zhan Chen, Junyu Li, Kerui Xu, Zhao Liu, Yonglu Wang, Minghao The Hong Kong Polytechnic University Department of Electrical and Electronic Engineering Hong Kong Hong Kong The City University of Hong Kong Department of Electrical Engineering Hong Kong Hong Kong Shenzhen Research Institute Department of Electrical and Electronic Engineering China The Hong Kong Polytechnic University Research Institute of Smart Energy Hong Kong Hong Kong Central South University School of Automation Changsha China The Department of Electrical and Computer Engineering China University ofMacau Advanced Research Institute in Hengqin Zhuhai China

DC shipboard microgrids (SMGs) are a low-carbon and feasible solution to host various distributed energy resources (DERs) and shipboard loads. However, the intermittency of DERs and constant power load (CPL) characteristic of the shipboard loads will compromise the stability of DC SMGs. To accommodate DERs and address the stability issue in space-stringent DC SMGs simultaneously, a compound large-signal stabilization scheme is proposed in this paper to operate the interleaved multilevel boost DC-DC converter (IMBDC) in DC SMGs. It combines the high-order nonlinear disturbance observer (NDO) technique and the offset-free model predictive control (MPC) algorithm. Specifically, the uncertain variations of CPLs and circuit parameters are modeled as disturbances. High-order NDOs are used to estimate these disturbances accurately within a finite time. Depending on the estimation and system model, the MPC will regulate the system states against large-signal disturbances. Simulation results demonstrate that the proposed algorithm outperforms the traditional linear and nonlinear control schemes with better stability and dynamic performances. Both the real-time simulation and hardware experiment results are provided to further validate the applicability and effectiveness of the proposed control strategy. © 2015 IEEE.

关键词： Predictive control systems

Nonlinear Optimal Regulation of SIR Epidemic Model via Finite-Horizon SDRE

学校读者我要写书评

暂无评论

Nonlinear Optimal Regulation of SIR Epidemic Model via Finit...

International Conference on computer and Information Technology (ICCIT)

作者： Sudipta Paul Samiul Islam Desineni Subbaram Naidu Md. Abu Hanif Pramanik Department of Electrical Computer & Systems Engineering Rensselaer Polytechnic Institute New York USA Department of Electrical & Computer Engineering Rajshahi University of Engineering & Technology Bangladesh Department of Electrical Engineering University of Minnesota Duluth USA

A Susceptible-Infectious-Recovered (SIR) model is a popular and fundamental epidemiological model often used to assess the efficacy of disease prevention and control measures. SIR disease model, with the implementation of various optimal control methods, is investigated to determine the optimal ways to lessen the effects of an infectious disease. In this research work, a non-linear, closed-loop, finite-horizon, optimal regulation with state-dependent Riccati equation (FH-SDRE) is implemented in an epidemic model called SIR model with vaccination to minimize the infected and susceptible population. Given the nonlinear dynamics of the SIR epidemic model, initially, this method converts the nonlinear equations into a state-dependent coefficient (SDC) matrix, which holds the nonlinearities of the dynamical system. After that, SDRE is solved at every small time instant over a specified finite time period. The core idea here is to utilize an approximate analytical approach that transforms the differential Riccati equation (DRE) into a differential Lyapunov equation (DLE) during every time period. Then, the real-time solution of the DLE at each of these time steps will result in the control law required to achieve the desired system performance. The effectiveness of the FH-SDRE regulation for the SIR model is demonstrated by minimizing the infected and susceptible population within a desired 30-day time period.

关键词：