About
This site facilitates a study of neural networks and their ability to predict
simple, deterministic cellular automata systems.
🕸️ vs 👾
This is a distributed study. All experiments are run by members of the public community,
in their browsers, using their computer's resources.
🍿
The main goal of the study is to map the six-dimensional performance metric landscape (Binary Cross-Entropy, Accuracy, Specificity, Precision, Recall, F1)
over the thirteen-dimensional training parameter input space.
There are 393,216,000,000,000 experiments that can be run at the moment, and each one helps map out a point on this surface.
In many situations the nets don't perform well, while in other cases they become highly accurate at predicting the cellular automata.
🧮
Test metric results from completed experiments are aggregated and shown on the Analysis page.
Not all information from the training & testing process is saved, so internalize the training loss curves and test set images while they exist in memory.
🕹️
Future Development
A meta analysis will begin once we hit our first goal of 30,000 points (completed training-test runs)!
Those results will be used in the meta-level, supervised learning study where we explore predicting the test metrics. Can we accurately predict how effective learning will be given the training parameters? 🤔
The inputs for that study will be the training parameters (instead of the cellular automata start state @ t=0) and the outputs will be the test metrics (instead of the cellular automata end state @ t > 0).
So get over to Training and click that button!
🤯
Additional Information