Simple Regret Optimization in Online Planning for Markov Decision Processes · HackerPapers discussionshttps://hackerpaper.kaipfister.de/p/1206.3382/Collected HN comments and submissions linking this paper.enMon, 02 Nov 2026 11:16:22 GMTSimple Regret Optimization in Online Planning for Markov Decision Processes — mentioned by eutectichttps://news.ycombinator.com/item?id=9915896urn:hackerpapers:mention:1206.3382:9915896Mon, 20 Jul 2015 13:34:12 GMT<p>The idea that top-level programs use UCT is just folklore at this point. I know that at least MoGo uses a greedy version of Monte-Carlo Tree Search enhanced with RAVE (Rapid Action Value Estimation), a way to share information between branches of the search tree, a pattern database which runs an evaluation function on small patches of the board, and a set of simple heuristics. The consensus is that UCT is bad because it tries to minimize cumulative regret (the average performance over all simulations) at each node, rather than the strength of the chosen move, and that the various heuristic…</p><p><a href="https://hackerpaper.kaipfister.de/p/1206.3382/">Paper and discussion history</a>. A mention is not an endorsement.</p>cs.AI