It’s exciting to see growing interest and participation in the AGI community.
This is another brief post to share two examples. In this case, they both build on one approach to AGI, and that is HTM.
FXAI – Explorations Into Computational Intelligence
A blog that is pretty well described by the title. The author, Felix Andrews, has been focussing on HTM, implementing a version of CLA in Clojure that runs in the browser and follows Visualisation Driven Development. The latest post describes a new algorithm for selection of winning columns based on local rather than global inhibition. Global inhibition is one of the compromises that the NUPIC implementation makes in favour of computational performance.
My Attempt at Outperforming DeepMind’s Atari Results
DeepMind’s successes caused a big splash in the AI research community and tech industry in general. This blog by Eric Laukien documents progress of a project that, as the title suggests, has the goal of achieving better performance than DeepMind. His approach is to incorporate Reinforcement Learning with HTM to create an agent that can learn how to act in the world. This is the only other example we’ve seen of an MPF implementation that can take actions.