https://mattdown.ai/tags/machine-learning/Recent content in Machine Learning on Matt DownHugoen-usMatt DownMon, 28 Sep 2026 00:00:00 +0000https://mattdown.ai/posts/tenten/Mon, 28 Sep 2026 00:00:00 +0000https://mattdown.ai/posts/tenten/<h2 id="intro">Intro</h2> <p>For a while now I&rsquo;ve been interested in <strong>AI accelerated design space exploration</strong>. It&rsquo;s a compelling approach that allows the rapid iteration of game rules and balance. As someone who&rsquo;s spent a lot of their career working on early stage prototypes, I&rsquo;ve seen firsthand how important it is to nail the core mechanics of a game.</p> <p>So how does that work? You first need to train an agent that is capable of playing the game in a way that it can react to balance changes and even rule changes. Then you define some metrics based on game dynamics that I&rsquo;ll cover in the next section, before changing the game balance and rules to hit those metrics.</p>https://mattdown.ai/posts/mcts/Wed, 17 Jun 2026 00:00:00 +0000https://mattdown.ai/posts/mcts/<h2 id="intro">Intro</h2> <p>Monte Carlo Tree Search (MCTS) is the famous tree-search planning algorithm used in Google DeepMind&rsquo;s AlphaGo as well as many other applications. The algorithm itself is relatively simple and should be seen maybe more as a planning algorithm than a learning one.</p> <p>At a high level, how does MCTS work? Instead of exhaustively searching all possibilities, MCTS searches the state tree in a principled way that prioritises exploration of the most promising paths within that tree.</p>