The ability of robots to imitate human learning strategies-rapidly adapting to new tasks without large datasets-has garnered significant attention in meta-learning. Meta-reinforcement learning seeks to enhance robotic...
详细信息
ISBN:
(数字)9798331521554
ISBN:
(纸本)9798331521561
The ability of robots to imitate human learning strategies-rapidly adapting to new tasks without large datasets-has garnered significant attention in meta-learning. Meta-reinforcement learning seeks to enhance robotic agent flexibility across diverse tasks and contexts, offering promise where single-task learning often fails. Despite advancements like multi-task diffusion models and task-weighted optimization mechanisms, effectively training tasks with varying complexities simultaneously remains a major challenge. This paper introduces a novel meta-reinforcement learning method that addresses this issue by clustering the training tasks of robotic arms based on semantic and trajectory similarities, while leveraging adaptive learning rates and task-specific weights proposed by the multitask optimization techniques. Our approach, TEAM, emphasizes performance-driven semantic clustering, optimizing based on robotic task similarity, complexity, and convergence objectives. We also integrate fast adaptive and multi-task optimization of the diffusion model to enhance computational efficiency and adaptability. More specifically, we introduce a cluster-specific optimization technique, using specialized parameters for each group to allow more refined task handling. The experimental validation demonstrates the effectiveness of this scalable method in improving performance, adaptability, and efficiency in real-world, heterogeneous robotic tasks, further advancing robotic computing in meta-reinforcement learning.
In an effort to continue to help provide various and thriving experiences to engineering undergraduates and help increase retention, a mid-size university uses a high impact practice of using peer teachers in the clas...
详细信息
ISBN:
(数字)9781665475280
ISBN:
(纸本)9781665475297
In an effort to continue to help provide various and thriving experiences to engineering undergraduates and help increase retention, a mid-size university uses a high impact practice of using peer teachers in the classroom. It is a standard practice to use graduate teaching assistants in most areas of higher education, especially in engineering classes, discussions, labs or just to hold office hours and grade. However, an increasing number of universities have adopted and leveraged undergraduate teaching assistants as it demonstrates to effectively improve students' grades, retention, student self-efficacy, and provide some financial relief to academic institutions [1]. The impact of using peer teachers is especially evident in the first and second years in engineering. Students who participate in the role are third year or above demonstrate expertise, leadership, and an interest in teaching as part of their development. At a mid-size minority serving institution, an undergraduate teaching assistant (termed as teaching fellow) was developed informally in 2005 in the mechanical and chemical engineering department and expanded in 2017 to the entire College of engineering and Informational Technology. In this case study, alumni and current teaching fellows were interviewed to assess the impact of their experiences and how it influenced their educational experience in their major and current career. Several themes were discovered to include increased professional and personal skill sets, self-efficacy in engineering, motivation to participate in the program, impact on career, creation of community and improvements needed to the program. A few teaching fellows decided to continue to be a p12 teacher.
The standard approach to modeling the human brain as a complex system is with a network, where the basic unit of interaction is a pairwise link between two brain regions. While powerful, this approach is limited by th...
详细信息
The standard approach to modeling the human brain as a complex system is with a network, where the basic unit of interaction is a pairwise link between two brain regions. While powerful, this approach is limited by the inability to assess higher-order interactions involving three or more elements directly. In this work, we present a method for capturing higher-order dependencies in discrete data based on partial entropy decomposition (PED). Our approach decomposes the joint entropy of the whole system into a set of strictly non-negative partial entropy atoms that describe the redundant, unique, and synergistic interactions that compose the system's structure. We begin by showing how the PED can provide insights into the mathematical structure of both the FC network itself, as well as established measures of higher-order dependency such as the O-information. When applied to resting state fMRI data, we find robust evidence of higher-order synergies that are largely invisible to standard functional connectivity analyses. This synergistic structure is symmetrical across hemispheres, largely conserved across individual subjects, and is distinct from structural features based on redundancy that have previously dominated FC analyses. Our approach can also be localized in time, allowing a frame-by-frame analysis of how the distributions of redundancies and synergies change over the course of a recording. We find that different ensembles of regions can transiently change from being redundancy-dominated to synergy-dominated, and that the temporal pattern is structured in time. These results provide strong evidence that there exists a large space of unexplored structures in human brain data that have been largely missed by a focus on bivariate network connectivity models. This synergistic "shadow structures" is dynamic in time and, likely will illuminate new and interesting links between brain and behavior. Beyond brain-specific application, the PED provides a very general app
The efficient Kalman filter has been widely used in recent decades to obtain air navigation information in UAVs. However, for a good performance of the Kalman filter, the model that describes the system dynamics must ...
The efficient Kalman filter has been widely used in recent decades to obtain air navigation information in UAVs. However, for a good performance of the Kalman filter, the model that describes the system dynamics must not contain uncertainties. This paper presents the implementation of a robust Kalman filter to estimate the attitude, velocity, and position of UAVs. The robust filter considers uncertainties in the sensor models. A mathematical structure based on the solution of linear systems synthesizes the predictor-corrector robust estimation algorithm. The main contribution of this study is the proposed QR decomposition based on Givens rotation to solve the linear system. The simulated experiments used sensory data collected in Zürich-Switzerland and ground truth referencing attitude, velocity, and position. The offline simulation results express the effectiveness of the robust Kalman filter for this application, with a reduction of up to 18.9% in the estimation error, in relation to the standard Kalman filter. The proposal to use systolic arrays for numerical solutions has shown promise for implementation in parallel processing platforms, such as FPGAs.
Internet of Things (IoT) industrial applications have increased in internet-connected devices. As a result, accessibility, scalability, connectivity, and adaptability have become major challenges. It is possible to cr...
详细信息
ISBN:
(数字)9798331528324
ISBN:
(纸本)9798331528331
Internet of Things (IoT) industrial applications have increased in internet-connected devices. As a result, accessibility, scalability, connectivity, and adaptability have become major challenges. It is possible to create connections between loT devices over the wireless medium, but efficiently utilizing scarce spectrum is the biggest concern. The paper proposes a novel method for selecting sensing nodes with the goal of reducing energy usage and increasing sensing efficiency. In order to ensure that node trust values conform to each other without any ambiguity, cryptography encrypts all information that relates to a node in a central node (CN). Nodes participating in CSS are selected based on their performance. Based on a comparison of various algorithms using varying numbers of nodes, the proposed algorithm achieves a 10% sensing efficiency.
Zero-day attacks present a significant security threat to vehicular networks, exploiting vulnerabilities at both software and hardware levels within such systems that remain undiscovered. Mitigating these threats is e...
详细信息
ISBN:
(数字)9798331524937
ISBN:
(纸本)9798331524944
Zero-day attacks present a significant security threat to vehicular networks, exploiting vulnerabilities at both software and hardware levels within such systems that remain undiscovered. Mitigating these threats is essential to ensuring the safety and security of vehicular systems. Support Vector Machine (SVM) is a good candidate for anomaly detection of zero-day attacks within vehicular networks because it can handle highdimensional data and effectively distinguish between normal and abnormal patterns in complex and dynamic environments. A trained SVM on the normal operation data of in-vehicular network can identify flag deviations, thus making it effective in the detection of any previously unknown attack patterns, which is a common behaviour of zero-day attacks. In this paper, we introduce an anomaly detection method called “ZeroCAN” which models the behaviour of every single electronic control unit on the network with a separate SVM and a set of high-level features that capture the timing and data payload aspects of CANbus traffic. This approach achieves an anomaly detection rate of over $\mathbf{9 9 \%}$ and a false positive rate below $\mathbf{0. 0 1 \%}$ during normal operation in most cases.
programming can help K-12 students to develop their 21st-century core skills. Despite the benefits, programming is not common to be delivered in Indonesian K-12 education. There is a need to understand potential chall...
programming can help K-12 students to develop their 21st-century core skills. Despite the benefits, programming is not common to be delivered in Indonesian K-12 education. There is a need to understand potential challenges in introducing programming to K-12 students. We developed a questionnaire survey covering four identified dimensions of challenges: administrative, facilities, teachers, and students. We also asked about common programming assessments and their preferred software features for teaching programming. Forty K-12 teachers were invited to complete the survey. The responses were analyzed with thematic analysis using a bigram-based Latent Dirichlet Allocation topic modeling and descriptive statistics. Our study shows that the challenges include limited learning modules, an insufficient number of computers, limited programming skills, and limited computational thinking skills. Scratch was the most common programming language used and many programming assessments were about debugging a program or writing a small program. Visualization and animation can be helpful in teaching programming.
Multivariate time series (MTS) data, when sampled irregularly and asynchronously, often present extensive missing values. Conventional methodologies for MTS analysis tend to rely on temporal embeddings based on timest...
详细信息
Three-dimensional (3D) bulk fin-typed field effect transistors (FinFETs) have emerged as key devices that can scale down the technology node beyond 22-nm. However, the scaled devices have created new sources of fluctu...
Three-dimensional (3D) bulk fin-typed field effect transistors (FinFETs) have emerged as key devices that can scale down the technology node beyond 22-nm. However, the scaled devices have created new sources of fluctuation inherent in 3D geometry. The interface trap is one such fluctuation that is caused by the trapping and de-trapping of charge carriers and has an adverse effect on device characteristics and variability. In this work, we study impacts of random interface traps (RITs) on electrical characteristic of bulk FinFETs by using a 3D quantum-mechanically corrected device simulation. RIT position effects on short channel effects (SCEs) are examined with physical governed influence to show the major fluctuations. More than 50% reductions of the RITs-induced characteristic fluctuation of the germanium (Ge) devices are observed, compared with Si devices. The Ge ones can reduce SCE variations and exhibit high immunity to RITs.
This paper presents the development of a versatile mobile robot platform designed for precision agriculture. The robot’s proposed overview architecture and its manufacturing are presented while it further discusses t...
This paper presents the development of a versatile mobile robot platform designed for precision agriculture. The robot’s proposed overview architecture and its manufacturing are presented while it further discusses the incorporated vision-based perception modules for road segmentation in farm environments and maize stem detection. For road segmentation, the Segment Anything Model (SAM) based on Zero-Shot Segmentation was utilized. The SAM algorithm effectively extracted navigable spaces in challenging scenarios, demonstrating its robustness and adaptability. Furthermore, considerations were made for computational efficiency, motivating future implementation on low-power devices. In the maize stem detection module, a comprehensive dataset of maize stem images obtained from a local agricultural field was created. The images were processed using the YOLOv5 model, resulting in a highly accurate and efficient maize stem detection module. The validation of both perception modules highlights the successful integration of vision-based technologies into the platform. The platform’s adaptability and robustness make it a valuable tool for precision agriculture applications. By leveraging these technologies, the proposed vehicle contributes to improving crop monitoring and management, enhancing overall agricultural practices.
暂无评论