Games Are Smarter Than Us
summary
Builds a tiny 2D game in the browser using canvas and exports from bitmellow then runs reinforcement learning in Javascript with reinforce.js (and mentions tensorflow js) to train an agent to collect mushrooms; flappy‑bird style learning takes about 25 minutes to converge, models save/load, and a single trained network transfers to new sprites and rules so it can learn chasing or fleeing by swapping the reward sign. The demo works live and improves over minutes.
AI-written summary · about
chapters (5)
- 0:00Start
- 1:20Village of Boredom
- 12:37Mountain of Wisdom
- 14:33Reinforcement Learning!
- 22:04River of Opportunities
related talks
29mReinforcement Learning in the Browser
43mAlpha 2048 - Reinforcement Learning in the Browser
56mCan I do AI?
31mMake coding fun: developing a game in the browser
20mAlpha2048
44mArtificial Intelligence and Machine Learning in the browser
25mJavascript as Play A Primer on Getting Started with Video Games
32mDeep Learning in JS