This paper analyzes the characteristics of emotion state and group behavior in the evacuation process.During the emergency evacuation,emotion state and group behavior are interacting with each other,and indivisible.Th...This paper analyzes the characteristics of emotion state and group behavior in the evacuation process.During the emergency evacuation,emotion state and group behavior are interacting with each other,and indivisible.The emotion spread model with the effect of group behavior,and the leader-follower model with the effect of emotion state are proposed.On this basis,exit choice strategies with the effect of emotion state and group behavior are proposed.Fusing emotion spread model,leader-follower model,and exit choice strategies into a cellular automata(CA)-based pedestrian simulation model,we simulate the evacuation process in a multi-exit case.Simulation results indicate that panic emotion and group behavior are two negative influence factors for pedestrian evacuation.Compared with panic emotion or group behavior only,pedestrian evacuation efficiency with the effects of both is lower.展开更多
Multi-exit architecture allows early-stop inference to reduce computational cost,which can be used in resource-constrained circumstances.Recent works combine the multi-exit architecture with self-distillation to simul...Multi-exit architecture allows early-stop inference to reduce computational cost,which can be used in resource-constrained circumstances.Recent works combine the multi-exit architecture with self-distillation to simultaneously achieve high efficiency and decent performance at different network depths.However,existing methods mainly transfer knowledge from deep exits or a single ensemble to guide all exits,without considering that inappropriate learning gaps between students and teachers may degrade the model performance,especially in shallow exits.To address this issue,we propose Multi-exit self-distillation with Appropriate TEachers(MATE)to provide diverse and appropriate teacher knowledge for each exit.In MATE,multiple ensemble teachers are obtained from all exits with different trainable weights.Each exit subsequently receives knowledge from all teachers,while focusing mainly on its primary teacher to keep an appropriate gap for efficient knowledge transfer.In this way,MATE achieves diversity in knowledge distillation while ensuring learning efficiency.Experimental results on CIFAR-100,TinyImageNet,and three fine-grained datasets demonstrate that MATE consistently outperforms state-of-the-art multi-exit self-distillation methods with various network architectures.展开更多
基金Project supported by the National Key Research and Development Program of China(Grant No.2017YFC0803903)the National Natural Science Foundation of China(Grant No.62003182)。
文摘This paper analyzes the characteristics of emotion state and group behavior in the evacuation process.During the emergency evacuation,emotion state and group behavior are interacting with each other,and indivisible.The emotion spread model with the effect of group behavior,and the leader-follower model with the effect of emotion state are proposed.On this basis,exit choice strategies with the effect of emotion state and group behavior are proposed.Fusing emotion spread model,leader-follower model,and exit choice strategies into a cellular automata(CA)-based pedestrian simulation model,we simulate the evacuation process in a multi-exit case.Simulation results indicate that panic emotion and group behavior are two negative influence factors for pedestrian evacuation.Compared with panic emotion or group behavior only,pedestrian evacuation efficiency with the effects of both is lower.
基金supported by the National Natural Science Foundation of China(No.U1866602)the Starry Night Science Fund of Zhejiang University Shanghai Institute for Advanced Study,China(No.SN-ZJU-SIAS-001)。
文摘Multi-exit architecture allows early-stop inference to reduce computational cost,which can be used in resource-constrained circumstances.Recent works combine the multi-exit architecture with self-distillation to simultaneously achieve high efficiency and decent performance at different network depths.However,existing methods mainly transfer knowledge from deep exits or a single ensemble to guide all exits,without considering that inappropriate learning gaps between students and teachers may degrade the model performance,especially in shallow exits.To address this issue,we propose Multi-exit self-distillation with Appropriate TEachers(MATE)to provide diverse and appropriate teacher knowledge for each exit.In MATE,multiple ensemble teachers are obtained from all exits with different trainable weights.Each exit subsequently receives knowledge from all teachers,while focusing mainly on its primary teacher to keep an appropriate gap for efficient knowledge transfer.In this way,MATE achieves diversity in knowledge distillation while ensuring learning efficiency.Experimental results on CIFAR-100,TinyImageNet,and three fine-grained datasets demonstrate that MATE consistently outperforms state-of-the-art multi-exit self-distillation methods with various network architectures.
文摘[研究目的]解决多源数据融合过程中参与者贡献与收益的匹配问题、参与者选择缺乏灵活性以及联邦系统的动态适应性不足问题,提升多源数据融合的公平性和合理性。[研究方法]提出一种基于动态自适应联邦学习的多源数据融合框架(Federated Learning and Dynamic Improvement,FLDI),并设计预算分配机制、参与者选择机制以及参与者动态进出机制确保多源数据融合过程的安全、公平和可持续。分别在分类任务的专利、论文以及媒体数据集和预测任务的MNIST、FMNIST和CIFAR-10数据集上展开性能测试,并在不同场景中评估框架性能。[研究结果/结论]FLDI在面对复杂场景时,其准确率相较于FedAvg和FedProx提升了3%~4%;在干净数据集场景下,FLDI在分类任务的平均准确率达到67.01%,在预测任务的平均准确率达到81.56%;进行增强实验后,FLDI在分类任务的平均准确率上升了4.54%,在预测任务的平均准确率上升了3.31%;FLDI框架在分类任务和预测任务中较之FedAvg和FedProx更具性能优势。
