CADENT: Gated Hybrid Distillation for Sample-Efficient Transfer in Reinforcement Learning
Mahyar Alinejad, University of Central Florida, [email protected], Yue Wang, University of Central Florida, [email protected], George Atia, University of Central Florida, [email protected] · 2026
Read CADENT: Gated Hybrid Distillation for Sample-Efficient Transfer in Reinforcement Learning by Mahyar Alinejad, University of Central Florida, [email protected], Yue Wang, University of Central Florida, [email protected], George Atia, University of Central Florida, [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-02532.txt.
Description
Mahyar Alinejad and 2 co-authors combine two kinds of teacher knowledge so reinforcement learning agents can transfer skills with fewer samples, in a 2026 arXiv paper. Policy distillation gives fine-grained action advice but misses long-term strategy. Automaton-based methods capture task structure but cannot guide individual actions. Domain shift between source and target environments defeats either alone.
Their framework, CADENT, merges strategic automaton knowledge with tactical policy knowledge into one guidance signal. An experience-gated trust mechanism weighs teacher guidance against the student's own experience at each state-action pair. Guidance therefore fades gracefully where the target domain differs. Across sparse-reward grid worlds and continuous control tasks, CADENT improved sample efficiency by forty to sixty percent over baselines. Final performance was also superior, making the transfer both faster and better.
Catalog record
| CREATOR | Mahyar Alinejad, University of Central Florida, [email protected], Yue Wang, University of Central Florida, [email protected], George Atia, University of Central Florida, [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-02532