Genetic algorithms and rosetta@home

Message boards : Rosetta@home Science : Genetic algorithms and rosetta@home

To post messages, you must log in.

AuthorMessage
Stephen

Send message
Joined: 5 Jun 06
Posts: 23
Credit: 2,570,438
RAC: 0
Message 19479 - Posted: 29 Jun 2006, 17:57:16 UTC

I've taken a graduate course on GAs and learned how effective they can be for problems with large search spaces, especially those with lots of local minima or maxima. I even participated in a distributed GA search for optimizing electronic hardware design while it lasted.

Has any work been done to look at how GAs can help Rosetta@home's goals?

Stephen
ID: 19479 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
James Thompson

Send message
Joined: 13 Oct 05
Posts: 46
Credit: 186,109
RAC: 0
Message 19492 - Posted: 29 Jun 2006, 19:48:00 UTC - in response to Message 19479.  
Last modified: 29 Jun 2006, 19:48:38 UTC

I've taken a graduate course on GAs and learned how effective they can be for problems with large search spaces, especially those with lots of local minima or maxima. I even participated in a distributed GA search for optimizing electronic hardware design while it lasted.

Has any work been done to look at how GAs can help Rosetta@home's goals?

Stephen


Genetic algorithms are definitely a very powerful idea. We currently have a visiting graduate student who is using a Genetic Algorithm to combine different parts of predicted protein structures. There's short thread about this here.

One of the problems with using something like a GA in Rosetta@Home is communication. While it's plausible to have each computing node with an independently evolving population of structures, that process is limited by the number of structures that can be constructed on that single node in a reasonable amount of time. The most straightforward approach to adding this information would be prohibitively expensive in terms of communication, especially when we consider the number of users that we have on intermittently active or slow connections.

Hope that this answers your question.
ID: 19492 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Stephen

Send message
Joined: 5 Jun 06
Posts: 23
Credit: 2,570,438
RAC: 0
Message 19495 - Posted: 29 Jun 2006, 20:09:08 UTC - in response to Message 19492.  

Those are certainly questions to be looked at.

The distributed hardware evolution project (DHEP) I was part of took the approach of having each client be an island on a grid. Each island would evolve its own population of solutions. There was also in place a mechanism that would promote migration among neighboring islands. Over time, the average fitness of the islands' populations would go up. The project generated neat graphics showing the fitness distribution color coded on a map of the "geography".

The whole project, client and server (http://dhep.dbestern.com/), was implemented in java. I wonder if the code is still available (http://freshmeat.net/projects/dhep/) for any of the developers who might want to look at it for ideas. Maybe Miguel Garvie could be reached at his yahoo email.

Stephen
ID: 19495 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Dimitris Hatzopoulos

Send message
Joined: 5 Jan 06
Posts: 336
Credit: 80,939
RAC: 0
Message 19497 - Posted: 29 Jun 2006, 20:15:11 UTC - in response to Message 19492.  

One of the problems with using something like a GA in Rosetta@Home is communication. While it's plausible to have each computing node with an independently evolving population of structures, that process is limited by the number of structures that can be constructed on that single node in a reasonable amount of time. The most straightforward approach to adding this information would be prohibitively expensive in terms of communication, especially when we consider the number of users that we have on intermittently active or slow connections.


I hope that one day soon, the trickle-up, trickle-down facility of BOINC could be used to achieve what you described.

As far as frequency of communication, I'm sure that many people with permanent Internet could easily contact every e.g. 2hr (0.1day)
Best UFO Resources
Wikipedia R@h
How-To: Join Distributed Computing projects that benefit humanity
ID: 19497 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote

Message boards : Rosetta@home Science : Genetic algorithms and rosetta@home



©2024 University of Washington
https://www.bakerlab.org