Computer Systems Intelligence Lab.
CSI-Agent Group @ SKKU
Reinforcement Learning
CSI-Agent group works on incorporation of data-driven multi-task, multi-modal, continuous learning model architectures in the context of reinforcement learning to address the problems (sample-inefficiency, low generalization) of real-world adoption of reinforcement learning. CSI-Agent group develops pretrained behavior-driven models that can be adopted for downstream agent task learning in real-world settings.
On-going projects : Policy generalization with multi-modal transfer (NRF, 2023~2025), Open domain multi-modal self-directive intelligence (IITP, 2022~2026)
Embodied Agent
CSI-Agent group conducts the research on various embodied agent scenarios including drone, robot, and autonomous driving where the embodied agent learns to take optimal actions for achieving its goals in the environment, and can adapt its behaviors to changes in the environment or its goals. To do so, CSI-Agent group employs several machine learning techniques such as imitation learning, reinforcement learning, and multi-modal learning.
On-going projects : Personified agent learning (IITP, 2022~2026)
Intelligent System
CSI-Agent group focuses on system intelligence areas including network-storage-learning system optimization, autonomous system control, federated learning with NPUs, and networked real-time CPS (Cyber-Physical System), aiming at the application of data-driven deep learning technologies on large-scale real-world system automation that requires optimal sequential decisions.
On-going projects : Federated learning with NPUs (IITP, 2022~2024)