Backburner
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.
| Name | Description | Questions |
|---|---|---|
| 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? |