A very strange rendering "engine": Navigating the dreams of my handmade neuroscience-inspired AI! (AOgmaNeo)
CireNeikual
Member since
Nov 10, 2020
Projects
Recent Activity
I was able to get an AOgmaNeo agent to play Atari Pong on a Teensy 4.1 microcontroller. I believe it is the most resource-efficient general-purpose reinforcement learner!
A robotic demo of my project here (AOgmaNeo, from-scratch AI system): https://www.youtube.com/watch?v=JrCU5My2L6I
Project blog post:
Findings about Unsupervised Behavioral Learning (UBL)
Project blog post:
New release 1.5.0 + a possible paradigm shift?
Project blog post:
The Return of Topology + Experiments with Fast Weights
Project blog post:
UBL demo + collab!
Project blog post:
"GAN Theft Auto" but with AOgmaNeo
Project blog post:
Moving Away from Dual-Encoder Setup - New Findings + An Interactive Demo!
Project blog post:
A guide, "imagination", LD48, and lots of behind-the-scenes experiments
Project blog post:
Dual Encoder Setup
Learning a world model of the old MS maze screensaver from pixels using AOgmaNeo
Project blog post:
More Error-Driven Encoders and Decoders + Upcoming Demos
Project blog post:
ESR-light and conservative Q-learning
Video Recall Demo for AOgmaNeo, our hand-crafted (no C++ std lib!) machine learning system that doesn't backprop and learns fully online: https://www.youtube.com/watch?v=64xac-fITd0&feature=youtu.be
New video on my AOgmaNeo software running on a real-world robot. This video here is using the Python bindings to better interface with sensors, but you can also use it directly in C++ with no external libraries (not even the C++ standard library!)
https://www.youtube.com/watch?v=Dw6Upzcom0A&feature=youtu.be
Project blog post:
"Do not use" inputs, multi-step predictions and error-driven encoders
Project blog post:
Demo Work and Topology
While experimenting with new encoders (particularly ones based on ART - adpative resonance theory), I also ran some experiments with AOgmaNeo where an agent must find a red box. Since AOgmaNeo builds a world model during this reinforcement learning process, I can actually see "dreams" of what it has modeled in the world.