CTAM-RL: Continual Threat-Aware Memory for Continual Reinforcement Learning in Autonomous Cyber Defense

Abstract

{ "upload_type": "software", "title": "CTAM-RL: Continual Threat-Aware Memory for Continual Reinforcement Learning in Autonomous Cyber Defense (v1.1.0)", "description": "Code, extracted attack profiles, and per-seed results for the manuscript 'A Continual Reinforcement Learning Framework for Autonomous Cyber Defense Under Evolving Threat Landscapes'. Includes a custom network-defense simulator, a DQN defender, CTAM v1 (Threat Importance) and CTAM v2 (Knowledge Utility) replay memories, uniform and Prioritized Experience Replay baselines, dataset profile-extraction scripts for CICIDS2017, CSE-CIC-IDS2018 and UNSW-NB15, and all results reported in the manuscript. Raw datasets are not redistributed.", "version": "1.1.0", "license": "MIT", "creators": [ { "name": "Memarpour, Kimia" }, { "name": "Memarpour, Bahar" }, { "name": "Shirini, Kimia" }, { "name": "Samadi Gharehveran, Sina" } ], "keywords": [ "continual reinforcement learning", "autonomous cyber defense", "experience replay", "catastrophic forgetting", "forward transfer", "intrusion detection" ], "language": "eng" }