Log in

No account? Create an account

Previous Entry | Next Entry

Conference talk notes

Talk quality is all over the board at this meeting.

The best one so far was this morning, on accident. I meant to go to room B113, but went to room B117 instead.

The speaker talked about testing ideas about the evolution of network information-processing structures, using a set of robots. The network type he used was pretty straightforward: two sensory inputs, 6 intermediary processing nodes, then two motor output units. The robots were floating units with fins and motors, swimming in a 10-meter pool of water. His experiments involved running a set of trials on robots configured with different network connectivity structures, then measuring how much light they received from a light source centered in the middle of the pool. At the end of a set of trials, he would choose the robots with the most successful network structures (i.e. highest light harvesting), mutate them in some fashion, and run the trials again.

Within 7 generations, the robots got noticeably better at light-harvesting. Contrary to expectations, the most successful robots didn't have a higher degree of modularity (basically, compartmentalization of processing within the network) compared to less successful robots. Instead, success appeared to be tied to sparsity and pruning.

This seemed to me like one of the more informative cases where robots have been used to test biological principles.

The quality of talks at this meeting can be all over the board, probably because it's something of a catchall for Zoologists.

My brother and I both like it, but for the purpose of interacting with our different individual social circles.

Latest Month

September 2018


Powered by LiveJournal.com
Designed by Naoto Kishi