KnowRU: Knowledge Reuse via Knowledge Distillation in Multi-Agent Reinforcement Learning

Recently, deep reinforcement learning (RL) algorithms have achieved significant progress in the multi-agent domain. However, training for increasingly complex tasks would be time-consuming and resource intensive. To alleviate this problem, efficient leveraging of historical experience is essential, which is under-explored in previous studies because most existing methods fail to achieve this goal in a continuously dynamic system owing to their complicated design. In this paper, we propose a method for knowledge reuse called “KnowRU”, which can be easily deployed in the majority of multi-agent reinforcement learning (MARL) algorithms without requiring complicated hand-coded design. We employ the knowledge distillation paradigm to transfer knowledge among agents to shorten the training phase for new tasks while improving the asymptotic performance of agents. To empirically demonstrate the robustness and effectiveness of KnowRU, we perform extensive experiments on state-of-the-art MARL algorithms in collaborative and competitive scenarios. The results show that KnowRU outperforms recently reported methods and not only successfully accelerates the training phase, but also improves the training performance, emphasizing the importance of the proposed knowledge reuse for MARL.

DESIGN KNOWLEDGE REUSE IN DESIGN OF PROGRESSIVE STAMPING TOOLS: A QUALITATIVE STUDY

Progressive stamping tools are widely used in mass-production of sheet metal components and their performance is critical as the design of the tool impact the cost of the manufactured component significantly.Knowledge reuse is an important part of successful design in general, and in progressive stamping tool design in particular. In the study described in this paper, 8 tool designers from 5 different Swedish companies were interviewed about (1) at what points in the tool design process they search for previously designed tools for information and knowledge reuse, (2) under what conditions and for what reasons does this reuse take place, and (3) what types of information and knowledge are sought for at each point. The results show that reuse of information and knowledge from previously designed tools happens in many parts of the tool design process. The reasons and conditions for reuse vary depending on where in the process the designer is. High component complexity is one example of a common factor triggering reuse. Also, information about the performance of the tool is important to tool designers, as they only want to reuse information and knowledge from tools with good performance and low maintenance.

Application domain ontology-based transformation to database design

The paper analyses graph oriented ontology transformation into conceptual data model. A number of methodswere proposed to develop conceptual datamodels, but only fewdealswith knowledge reuse. In this paperwe present an approach for knowledge represented by ontology automatic transformation into conceptual data model. The graph transformation language is presented and adapted for formal transformation of ontology into conceptualmodel. Details and examples of proposed ontology transformation into conceptual data model are presented.

Application of Computer Technology in Optimal Design of Overall Structure of Special Machinery

With the transformation and upgrading of my country’s industrial structure, the level of manufacturing automation has gradually improved. According to research, the design of mechanical products is mostly completed by improvement or innovation on the basis of existing design knowledge. Knowledge reuse is a technique to ensure the maximization of design resource utilization by reusing design knowledge. This article applies knowledge reuse technology to the development and design of mechanical products. By integrating the technical logic of the functional analysis system with the development of quality functions, the transformation of customer demand information and product function design is realized, and the task of the product design plan analysis phase is completed. This paper uses the finite element analysis software ANSYS to explore a new nonlinear finite element modeling method and conducts simulation experiments. At the same time, this paper improves the genetic algorithm, which effectively improves the optimization efficiency and completes the parameter optimization under multiobjective and multistructure conditions. From the experimental results, it takes 328.64 seconds for the basic genetic algorithm to search for the optimal solution of the complex problem. The improved time is shortened to 86.31 seconds, and the efficiency is increased by 73.74%. This shows that the improved genetic algorithm has better robustness and can find the optimal solution in a shorter calculation time. For complex problems such as the optimization of the overall structure of special machinery, the improved genetic algorithm obviously helps to improve the optimization efficiency and improves the effectiveness and pertinence of product design schemes.

Identifying and categorising knowledge reuse activities in electronic repositories

Purpose Knowledge reuse using electronic repositories, while increasingly important, requires more thorough analysis. Service modularity has been recently applied in services research but has not been integrated into knowledge reuse studies. The purpose of this paper is to draw on both service modularity and knowledge reuse to develop and validate a framework that categorises forms of packaged knowledge in an electronic repository. Design/methodology/approach Drawing on knowledge reuse and service modularity research, a model is proposed. The model is empirically tested using a case study research design. Findings This research highlighted the value of including both context and process as key dimensions when packaging service knowledge for reuse. This study identifies knowledge types present in modular solutions and how they were configured and reconfigured in the knowledge repository. This research identified five ways modularised services were leveraged. In addition to the traditional scale and stretch approaches, already present, but conflated, in the service literature, three other configurations were identified; shrink, separate and segment. Research limitations/implications The findings are based on a single empirical case study which may limit the generalisability of the findings. There is a need for additional research to further validate the model in additional contexts. Practical implications This study provides managers with empirical examples of how a modular repository was used in practice and outlines five ways of recombining contextual and processual elements to enable service codification and reuse. It has implications for how knowledge is decomposed and recombined in repositories, suggesting an explicit separation of context and process knowledge while developing modular elements within both. Originality/value To the best of the author’s knowledge, this is the first study that explicitly uses context and process as dimensions and draws on service modularity to understand types of knowledge reuse in electronic repositories. In doing so, it adds value by developing and validating a model that identifies five types of reuse.