Stop Regressing: Training Value Functions via Classification for Scalable Deep RL — mentioned by matrix2596

Fri, 22 Mar 2024 11:53:30 GMT

I'm building upon insights from this paper (https://arxiv.org/pdf/2403.03950.pdf) and believe that classification can sometimes outperform regression, even when dealing with continuous output values. This is particularly true in scenarios where the output is noisy and may assume various values (multi modal). By treating the problem as classification over discrete bins, we can obtain an approximate distribution over these bins, rather than settling for a single, averaged value as regression would yield. This approach not only facilitates sampling but may also lead to more favorable loss…

Paper and discussion history. A mention is not an endorsement.

Stop Regressing: Training Value Functions via Classification for Scalable Deep RL · HackerPapers discussions

Stop Regressing: Training Value Functions via Classification for Scalable Deep RL — mentioned by matrix2596