Research ideas that may or may not have been acted on
A collection of project ideas brainstormed and developed through exploratory research in Winter quarter.
|GAN Swarm or Actor Critic, RL Swarm||Many agents that can compete adversarially and maybe also communicate and cooperate.||Would this implement swarm intelligence methods of distributed agent level computation or is it a swarm just due to the large number of agents? What sort of environment? What sort of goals?|
|Synthetic Language||Give models a protocol to communicate between each other as inputs and outputs of each||Has it been done? What kind of models would benefit most from this?|
|Fungal Computing||Grow mushrooms (maybe we can talk with the Indoor Farming RSO) and try to compute with them.||How? Could we instead use priciples of how fungi communicate and grow in silico instead of literally growing them?|
|Memory ML network||Use something like a Hopfield network to encode memory in NNs.||Is this similar to any parts of the brain? How is memory represented in the brain?|
|Hebbian Learning||Update the weights of NNs based on the firing rate of previous and next layers similar to how the brain stregthens and weakens synapses.||What work has been done on this already? What type of NNs do we want to apply it to?|
|Predictive Coding||Predict some feature of the environment as well as the next input from the environment ot the model, RL based agents. We could measure the difference in performance between non-pred. coding and pred. coding networks.||Has it been done? At what level of granularity do we want to predict the next input? The exact inputs or larger features of it?|