[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-35945":3,"doc-seo-35945":29},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},35945,4810365810221,"Aurora","https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d",6,"Technology","Automated Penetration Testing Using Deep Reinforcement Learning","Penetration testing is commonly performed manually and depends on pentesters’ tacit expertise, making the process time-consuming and error-prone. The paper introduces an automated penetration testing framework that uses deep reinforcement learning to recommend attack strategies for training and to recreate attacks for defense practice. The method first builds a realistic topology using Shodan and MulVAL-generated attack trees, then uses search and a Deep Q-Learning Network to select the most exploitable attack path, achieving 0.86 accuracy.","","cbCaiduKRCADi4zI","https://ap.wps.com/l/cbCaiduKRCADi4zI","pdf",310469,1,9,"English","en",105,"# Introduction\n## Automated penetration testing and motivation\n## Related AI and planning approaches\n## Attack trees and reinforcement learning background\n## Paper contributions","[{\"question\":\"What is the main goal of the proposed automated penetration testing framework?\",\"answer\":\"To automatically find the best attack path for a given network topology, in a way that resembles human attacker decision-making. It is designed for guided attack training and can also support defense training by recreating attacks in a test environment.\"},{\"question\":\"How does the framework build the attack representation before reinforcement learning?\",\"answer\":\"It uses Shodan to collect relevant server data and construct a realistic network topology, then applies MulVAL to generate an attack tree. Search algorithms are used to enumerate possible attack paths and form the matrix representation used by deep reinforcement learning.\"},{\"question\":\"How does the Deep Q-Learning Network (DQN) contribute to selecting an attack path?\",\"answer\":\"After generating candidate attack paths from the attack matrix, DQN is used to discover the most easy-to-exploit path among the candidates. Evaluation on thousands of input scenarios shows the optimal path is found with 0.86 accuracy.\"}]",1782766897,23,null]