This week was my final week at the Neuromorph Workshop in Telluride, CO. This week wasn’t as lecture heavy as it was work heavy in terms of projects that me and my team needed to finish. One of the projects that I have been primarily working on throughout the course of the workshop is a mountain biking game coded in python. The aim of this mountain biking game is to collect data on sensory error feedback and motor steering using something like a mouse. The aim of the game is to stay on a path for a set amount of time and trying not to deviate from the path as much as possible. The more that the player deviates from the path the more error is accumulated and then recorded. We plan to separate the sensory and motor brain activity, and with their individual components decode them into code to play a replica game of what the player is doing. In a perfect senecio, the replica game can make the move the player is thinking about a second before they actually preform it.
The part of the game that I was told to do was make a demo and add some levels for the players to complete. I did both of those last week, so it was onto decoding the EEG brain data into a replica game. I didn’t work on the decoding as much as I hoped, instead I was more observing the process of the decoding which I will try to describe as simply as possible.
The decoding didn’t use a typical LLM, in fact the method for decoding the EEG data was robust. My team used a method called cross validation, which is essentially changing the training and testing set of a model to see what produces the most optimal results. Instead of using cross validation to test a model, they just found the most optimal training/test set and used that as the decoding model. This optimal set was then used as weights to help decode the EEG data into the players current position. This cross validation method produces a lot of noise because of set weights and a lack of a model. Though the replica was noisy and the biking position barely corresponded to the players actual position in the game, it was progress. Hopefully the game will be continued next year and the decoder will produce better results.
There were three demos this week, one with Telluride Academy (a kids camp), one with people from Telluride, and a final presentation to the rest of the participants in the workshop. In terms of the demo with the Telluride Academy kids, I helped out by explaining to them a simplified version of what my project group has accomplished. After that demo, on Thursday night was the demo for people from all over Telluride to come in and look at the different project groups. I helped explain what we have been doing and also looked around at the other demos. Some of the other demos that caught my eye were a visualization of the brain and its activity in Unity, event cameras, and neuromorphic hardware, namely SpiNNaker 2. Finally on Friday, the final presentations were presented. I got up with my group to explain what we have been doing during the workshop and then watched everybody else’s presentations. There were many spectacular presentations that involved terminology that I didn’t understand, but the advancements made during the 3 week period of the workshop blew me away.
Friday evening there was a final banquet where I got to play poker, eat food, and watch some comedic presentations. The presentations used neuromorphic jargon as puns like for example “one shot” instead of “one-hot encoding” which was a presentation about a persons ability to pour a shot based off price. There were also many awards that were given to people that stood out during the workshop.
Overall, the neuromophs were very kind and insightful, and were never afraid to ask questions. They also always answered any of my questions regardless of how stupid they sounded. I learned so much during my short 3 weeks with the neuromorphs, and would suggest it to anyone who wants to have their mind blown.
There are no comments published yet.