(논문 요약) Stop Regressing: Training Value Functions via Classification for Scalable Deep RL (Paper)

핵심 내용

• 문제: scaling value-based RL methods that use regression to large networks, such as high-capacity Transformers, has proven challenging.
• 방법: use classification in place of regression for training
  
• GT: value, distribution 을 category 화.
  
  • Categorical Distributional RL 의 경우, 각 가장 가까운 두 지점에 weight 양분.
    
• 결과: value 를 regression 할때 보다 성능 개선.