User of the Day

User profile
Profile Phil1966
Member of L'Alliance Francophone

News RSS


[wildlife] "monster" workunit questions

So I had a couple ideas about the monster workunits. For those of you who are having them, are you running other workunits at the same time? Could you let me know the memory usage of those workunits? Does their performance improve if you only run one EXACT workunit at a time?

What I'm thinking may be happening is that the larger workunits require more memory, and if there are enough other workunits (especially other large EXACT workunits) running at the same time things may end up paging to disk because of memory requirements.

Travis Desell on Wednesday, April 26th
leave a comment


[wildlife] updated workunit deadlines

So I'm now dynamically generating workunit deadlines based on how much work they'll require (so deadlines are based on credit). Right now I'm setting the deadline to 1 day per 500 credit awarded -- let me know if this seems correct of if I should extend it further. I've also stopped the one search which should have been generating the really large workunits, so I imagine newly generated workunits should not have the excessively long multi-day runtimes you've been seeing.

Travis Desell on Monday, April 24th
leave a comment


[wildlife] status update

Just a status update for everyone. Due to the "monster" workunits people have been seeing I've made a couple tweaks server side and started to run some new searches (although the results I've been getting in have been extremely cool).

These new searches will start back with the smaller workunits and get progressively larger in size, so if you start seeing results with names like 'exact_genome_<timestamp>_6_<more numbers>' or 'exact_genome_<timestamp>_7_<more numbers>' those are from the new searches. If these start to see runtimes wildly out of line with credit please let me know.

Also, I've updated the work generation to have dynamically generated deadlines as opposed to a fixed 7 days. This should make the deadlines get longer as the workunits get more complex, so people won't see WUs that will take longer to compute than their deadline anymore. If the deadlines seem a bit off please let me know.

On top of that, I've been working on a new version of the application which will also be able to run on GPUs. I think this may address some of the other weird issues we've been seeing with the "monster" workunits as well. Will keep you posted as this develops but hopefully in a few weeks I'll be able to start testing an OpenCL version of the application that can run on NVidia or AMD GPUs.

Travis Desell on Tuesday, April 18th
leave a comment


[wildlife] grizzly chases geese

Susan has made a really cool video stitching together some of the imagery you've been looking at where a grizzly bear chases some geese around. Please take a look on youtube:

https://www.youtube.com/watch?v=UBEjeqBZSW8&feature=youtu.be

Feel free to talk to Susan if you have any questions in her thread in the video/image watching forum:

http://csgrid.org/csg/forum_thread.php?id=2388

Travis Desell on Wednesday, April 12th
leave a comment


[wildlife] credit inconsistencies

I've updated the algorithm which calculates credit in order to deal with credit inconsistencies we've been seeing recently. If you're still seeing wildly varying credit for workunits that take the same amount of time on your machines please let me know. Note that there are some 200-300 workunits in the queue that were still generated with the old credit algorithm so it may take a little time for the new workunits to start flowing out. Hope this fixes things!

Travis Desell on Friday, April 14th
leave a comment


[wildlife] "monster" workunits status update

Some users have been reporting that some workunits are taking significantly longer to run than what wingmen are reporting -- which is a pretty weird bug. See this forum thread.

For a status update, what I believe is happening is that certain parameters to the backpropagation is causing this issue on some operating systems/architectures. I've whipped up a page to track these "monster" workunits, which can be found here. This page shows all workunits with results that had runtimes more than 4x apart from each other. Quite a few of these had initial learning rates set at 1e-08 which is extremely low. I'm thinking this might be part of the problem so I've updated things server side to prevent the initial learning rate from being any lower that 1e-05.

I'm hoping that with this change, newly generated workunits should not have the problem anymore. If you're still seeing it please let me know. If you see it happen with a workunit that doesn't have a very low initial learning rate, please let me know as well as I'll need to do some more digging.

Travis Desell on Friday, April 21st
leave a comment


[wildlife] a push for UAV imagery analysis

Susan is making a push to try and analyze recent UAV imagery taken from the Hudson Bay! She'd like you to help us find birds, bears and caribou in our newly collected images from an unmanned aircraft (aka. Drone) in Northern Manitoba. The wildlife found in the images will be used to create training datasets for us to automate post-processing of large datasets from aerial imagery.

You can help us by reviewing images for the UAS Estimating Snow Geese project. We'll be using your results to expand on previous publications to test our software which will take your wildlife observations and automatically count snow geese. The more training information we get from you the more accurate our results.

Travis Desell on Wednesday, March 29th
leave a comment


[wildlife] EXACT MNIST and EXACT CIFAR

Given the issues some people have been having with the workunits that are working on the CIFAR-10 dataset eating up too much ram (~1.5GB) as opposed to the workunits working on the MNIST dataset, I've updated the applications and split things up. There is now an EXACT MNIST application and a separate EXACT CIFAR application. Now you can go into your preferences:

http://csgrid.org/csg/prefs.php?subset=project

Select "Edit preferences" and then you choose to run either or both of these applications. I'm keeping things running for the old applications, so if you're running EXACT or EXACT2 the validator/assimilator will still accept things and give you credit for work in progress as we transition over to searches running on these two new applications. I hope this helps!

Currently I'm just running EXACT MNIST workunits as I get things going. Once I start sending out EXACT CIFAR workunits I'll give those a credit bonus over EXACT MNIST due to the additional memory requirements.

Travis Desell on Sunday, March 26th
leave a comment


[all] cooling failure - potential downtime

I've been informed that the cooling system has gone down in the university datacenter that houses the virtual machine that the Citizen Science Grid runs on. As of now, they're not going to be shutting down the virtual machine, but it's still a potential. I just wanted to get a warning message out to everyone in case the site goes down for a period of time. Our system administrators are working on it and hopefully we don't experience any downtime, but if we do they'll be doing their best to get things back up and running.

Travis Desell on Saturday, March 25th
leave a comment


[wildlife] EXACT2 v0.19

I noticed an issue where the datafiles for v0.18 were missing a parameter, which was leading to some WUs not validating and some errors on the server side. I've updated the application to v0.19 which should fix this issue -- although may cause some of the WUs generated for v0.18 to error out. Sorry about that!

Travis Desell on Wednesday, March 22nd
leave a comment

Active Projects

Feel free to scroll through our image gallery above to get a better idea of what is going on with the project and see the field biologists in action, talk to us on the message boards, or watch videos about the project on our Google+ YouTube Channel. The following projects are actively being worked on by these Wildlife@Home team members:

Contact Information

Please email any questions or suggestions to Travis Desell, or feel free to post a message in our forums.

The Wildlife@Home team includes:

  • Travis Desell, Assistant Professor of Computer Science, University of North Dakota
  • Susan Ellis-Felege, Assistant Professor of Biology, University of North Dakota
  • Chris Felege, Educational Outreach Coordinator, Instructor of Biology, University of North Dakota
  • Alicia Andes, Graduate Research Assistant in Biology, University of North Dakota
  • Connor Bowley, Graduate Research Assistant in Computer Science, University of North Dakota
  • Nickolas Conrad, Undergraduate Research Assistant in Biology, University of North Dakota
  • Kyle Goehner, Graduate Research Assistant in Computer Science, University of North Dakota
  • Marshall Mattingly III, Graduate Research Assistant in Computer Science, University of North Dakota
  • John Palarski, Undergraduate Research Assistant in Biology, University of North Dakota
  • Lindsey Wingate, Undergraduate Research Assistant in Computer Science, University of North Dakota
With agency support from:
  • Kaylan Carrlson, Manager of Conservation Planning, Ducks Unlimited.
  • Tanner Gue, Conservation Specialist, Ducks Unlimited.
  • Aaron Robinson, North Dakota Game and Fish.
  • Terry Shaffer, U.S.G.S. Northern Prairie Wildlife Research Center.
  • Mark Sherfy, U.S.G.S. Northern Prairie Wildlife Research Center.
With previous support from:
  • Robert Bergman, Graduate Research Assistant in Computer Science, University of North Dakota
  • Paul Burr, Graduate Research Assistant in Biology, University of North Dakota
  • Benjamin Carpenter, Undergraduate Research Assistant in Compute Science, University of North Dakota
  • Becca Eckroad, Undergraduate Research Assistant in Biology, University of North Dakota
  • Jaeden Lovin, Undergraduate Research Assistant in Compute Science, University of North Dakota
  • Adam Pachl, Undergraduate Research Assistant in Biology, University of North Dakota
  • Julia Johnson, Undergraduate Research Assistant in Biology, University of North Dakota
  • Nitin Karodiya, Graduate Research Assistant in Computer Science, University of North Dakota
  • Eric Kjeldergaard, Graduate Research Assistant in Computer Science, University of North Dakota
  • Leila Mohsenian, Undergraduate Research Assistant in Biology, University of North Dakota
  • Kelly Sagen, Undergraduate Research Assistant in Biology, University of North Dakota
  • Rebecca VanderClute, Undergraduate Research Assistant in Computer Science, University of North Dakota

Publications

All publications, talks and posters about DNA@Home can be viewed on the Citizen Science Grid's publications page.

Support

Funded through a grant from the National Science Foundation (NSF).

Wildlife@Home is currently being supported by NSF award no. 1319700 through the Division of Intelligent Information Systems's Information Integration and Informatics (III) program.

Wildlife@Home is collaborating with Ducks Unlimited on a project invovling blue winged teal.

Wildlife@Home has been generously supported by a collaborative research award and new faculty SEED grant from UND's Office of Research Development and Compliance. The project's video streaming server is hosted by UND's Computational Research Center and the volunteer computing server is hosted by UND's Scientific Computing Center.


North Dakota Game and Fish has provided financial support for field logistics to collect sharp-tailed grouse videos.

USGS has provided financial support for camera equipment, video storage, and field assistance to collect data for the piping plover and interior least tern.

Powered by BOINC

Wildlife@Home is in part powered by the Berkeley Open Infrastructure for Network Computing (BOINC).