Alpha 2048 - Reinforcement Learning in the Browser
summary
A JavaScript-based reinforcement learning agent runs entirely in the browser using an in-browser JS RL implementation (open-source) and a full JS stack. It trains live against the 2048 game with a visible policy network and a simple UI that shows states, rewards and a drag-and-drop data view, demonstrating measurable score improvement over time without any server backend.
AI-written summary · about
chapters (11)
- 0:00Reinforcement learning in browser
- 6:02RL system implementation
- 10:20Action space & game mechanics
- 14:33Decision processes & training
- 16:34Exploration strategies
- 23:34Exploration parameters
- 27:12RL challenges
- 29:07Custom policy & correlation
- 31:08Live demo: game engine
- 35:36User interaction & data viz
- 40:35Q&A
related talks
29mReinforcement Learning in the Browser
20mAlpha2048
26mGames Are Smarter Than Us
33mAlpha2048 - how I taught my computer to play 1024
56mCan I do AI?
31mDesigning game AI with reinforcement learning
39mTF-Agents: A Flexible Reinforcement Learning Library for TensorFlow
41mReinforcement Learning and Minecraft