Existing explainability approaches for convolutional neural networks (CNNs) are mainly applied after training (post-hoc) which is generally unreliable. Ante-hoc explainers trained simultaneously with the CNN are more ...
Existing explainability approaches for convolutional neural networks (CNNs) are mainly applied after training (post-hoc) which is generally unreliable. Ante-hoc explainers trained simultaneously with the CNN are more reliable. However, current ante-hoc explanation methods mainly generate inexplicit concept-based explanations tailored to specific tasks. To address these limitations, we propose a task-agnostic ante-hoc framework that can generate interpretation maps to visually explain any CNN. Our framework simultaneously trains the CNN and a weighting network - an explanation generation module. The generated maps are self-explanatory, eliminating the need for manual identification of concepts. We demonstrate that our method can interpret classification, facial landmark detection, and image captioning tasks. We show that our framework is explicit, faithful, and stable through experiments. To the best of our knowledge, this is the first ante-hoc CNN explanation strategy that produces visual explanations generic to CNNs for different tasks.
A search and rescue (SAR) robot, named LUKER-I, is introduced to replace people to access to the accident scene when the underground gas and dust explosion happen under coal mine. The body of robot LUKER-I adopts trac...
详细信息
ISBN:
(纸本)9781424447947
A search and rescue (SAR) robot, named LUKER-I, is introduced to replace people to access to the accident scene when the underground gas and dust explosion happen under coal mine. The body of robot LUKER-I adopts tracked differential driven means, and has a fin arm tracked institutions. Robot can cross the barriers such as stairs, ditches, steep slopes and convex units based on different swing angle assistant. The control system uses two layers architecture, namely the bottom control network and IPC. IPC is mainly in charge of data acquisition, compression, and transmission of the voice and images of the scene. While the bottom layer uses CAN bus network to achieve the maintainability, expanded performance, and real-time control based on modular and distributed control. The working principle and design means of the robot are described detail in the paper. A modular method is used to carry out a detailed design.
Nowadays having the most energy efficiency is desirable in its own right from both economical and environmental points of *** power management is a system level solution for reducing the consumed energy with putting o...
详细信息
Nowadays having the most energy efficiency is desirable in its own right from both economical and environmental points of *** power management is a system level solution for reducing the consumed energy with putting off unused parts of the system and putting them on in an efficient *** Emotional Learning Algorithm has been introduced to show the effect of emotions as well known stimuli in the quick and almost satisfying decision making in *** remarkable properties of emotional learning,low computational complexity and fast training,and its simplicity in multi objective problems has made it a powerful methodology in real time control and decision systems,where the gradient based methods and evolutionary algorithms are hard to be used due to their high computational *** the emotional approach has been successfully used to obtain multiple objectives in prediction problems of real world *** first we introduce methods of dynamic power management and then a new method based on BELBIC would be *** simulation results show that this method has a high efficiency in various systems.
Dynamic voltage restorer (DVR)s are known as effective and economic means to compensate for voltage sags and swells. This paper proposes a novel method to generate reference waves synchronized with the grid whether or...
详细信息
Dynamic voltage restorer (DVR)s are known as effective and economic means to compensate for voltage sags and swells. This paper proposes a novel method to generate reference waves synchronized with the grid whether or not the line voltages are distorted by a fault. The proposed reference wave generator (RWG) looks for the normal voltage waveforms in the grid from the corrupted line voltages and synthesizes the expected positive sequence reference waves for DVRs. There is no transient problem on the generated reference waves when the line voltages are distorted by faults. Experimental results verified the proposed theory very well.
Achieving energy-efficient dynamic walking has become one of the main subjects of research on robotic bipedal locomotion. Approaches based on passive-dynamic walkers can accomplish bipedal locomotion. However, passive...
详细信息
Achieving energy-efficient dynamic walking has become one of the main subjects of research on robotic bipedal locomotion. Approaches based on passive-dynamic walkers can accomplish bipedal locomotion. However, passive dynamic walking has only been studied with the legs, and the effect of an upper body has not been clarified. This paper investigates the effect of an upper body on the efficiency and stability of dynamic bipedal locomotion based on observations. We first investigated a suitable upper body, which was a simple 1-link torso with a bisecting hip mechanism that would not destroy natural dynamics of the biped model. Second, we analyzed the robot's driving mechanism and chose underactuated virtual passive dynamic walking as the method for generating an efficient dynamic gait. We confirmed that efficient dynamic walking was possible with a specific resistance of 0.01 and investigated the effect of the physical parameters of the upper body through numerical simulations.
The complex ocean environment leads to low contrast in underwater images and poor target visibility. In addition, the detected objects have multi-scale properties and clustering issues, making marine object detection ...
详细信息
A new parallel grasper mechanism having 3-DOF spherical motions is proposed. The mechanism consists of one parallel four-bar chain and four serial sub-chains that are connected in parallel to the four-bar chain. Parti...
详细信息
A new parallel grasper mechanism having 3-DOF spherical motions is proposed. The mechanism consists of one parallel four-bar chain and four serial sub-chains that are connected in parallel to the four-bar chain. Particularly, all joint axes of the mechanism are placed to have a common intersection point securing that all joint motion spaces of the mechanism are confined within the 3-DOF spherical motion spaces. Thus, the mechanism can not only grasp irregular shaped objects by changing configuration of its four-bar chain, but also exhibit 3 DOF spherical motions after grasping. The position analysis and kinematic analysis of the mechanism are performed. Also, its kinematic characteristics are investigated through simulation in aspects of three kinematic design indices. Lastly, a prototype was developed and tested through motion tracking and grasping experiments.
This paper outlines the initial steps and basic framework for developing foundation/infrastructure robots/robotics based on foundation models and parallel intelligence,as well as the potential applications of new art...
详细信息
This paper outlines the initial steps and basic framework for developing foundation/infrastructure robots/robotics based on foundation models and parallel intelligence,as well as the potential applications of new artificial intelligence(AI)techniques such as AlphaGO,ChatGPT,and Sora.
This paper discusses the control of the compensation voltages in dynamic voltage restorers (DVR). On analyzing the power circuit of a DVR system, control limitations and control targets are presented for the compensat...
详细信息
This paper discusses the control of the compensation voltages in dynamic voltage restorers (DVR). On analyzing the power circuit of a DVR system, control limitations and control targets are presented for the compensation voltage control. Based on the preceded power stage analysis, a novel controller for the compensation voltages in DVRs is proposed by a feedforward control scheme. This paper also discusses the time delay problems in the control system of DVRs. Digital control systems inevitably have control delay due to its the sampling of signal, the switching frequency of an inverter, and the signal delay due to sensor itself. The total signal delay is considerable in real implementation and it endanger the system stability. This paper analyzes the relationship among the control delay, closed loop damping factor, and the output filter parameters of DVR inverters. Based on the control system analysis, a guideline is proposed to design the appropriate output filter parameters and the inverter switching frequency for DVRs. The proposed theory is verified by an experimental DVR system, which is controlled in full digital manner.
This paper studies the resilient consensus problem with time-varying adversary attacks and asynchronous events. In the synchronous or asynchronous update case, all the nodes or only a fraction of nodes update their st...
详细信息
ISBN:
(数字)9781728113982
ISBN:
(纸本)9781728113999
This paper studies the resilient consensus problem with time-varying adversary attacks and asynchronous events. In the synchronous or asynchronous update case, all the nodes or only a fraction of nodes update their states at each time step, respectively. For the synchronous update case, the nodes under attack are time-varying but do not change after some finite time;for the asynchronous update case, the nodes under attack are time-invariant. For both cases, a resilient consensus can be reached exponentially fast under a network topology condition inspired by the notion of the scrambling matrix and an explicit expression for the convergence rate is provided. Numerical examples are given to validate the theoretical results and to compare the performance with an existing algorithm.
暂无评论