This Week in MLC@Home
Notes for Sep 21 2020
A weekly summary of news and notes for MLC@Home

Summary
The new client (v9.61) is out of testing and deployed on the main MLDS app, and Dataset 3 WUs are out! We've only released 10 WUs per graph at the moment, will up that to at least 100 going forward. Dataset 3 WUs require at least MLDS app v9.61, and all previous version of the app have been deprecated. Also, the MLDStest applications have now been marked as beta, which required users to specify the "run test applications" preferences to participate in testing. Please consider doing so, as testing on mldstest this time around uncovered many issues before wide deployment, but also understand it is unstable and you may not get credit for WUs deployed through the testing app.

Preparations for Dataset 4 (MNIST and TrojAI) are underway. Paper writing continues for a conference deadline at the end of the month.

News:

• Dataset 3 WUs released: 100 automata to model, 100+ models per graph. WUs take longer to run, with appropriate increase in credit granted! Also, memory requirements per WU have now dropped below 350MB!
  
• Focus on the next few weeks will be monitoring Dataset 3 WUs to make sure there are no issues, and paper writing.
  
• We'll do an official release of a preliminary Dataset (1+2) once we have at least 1000 examples of each machine type, and we're getting closer!
  
• New server has not shipped from the vendor yet, but should be "any day now".
  
• We haven't forgotten about badges! We're just focused on the paper and new WU generation at the moment. That said, if volunteers would like to offer potential designs for badges, head on over the the forums and join the discussion.
  

Project status snapshot:

Tasks
Tasks ready to send 13588
Tasks in progress 22210
Users
With credit 767
Registered in past 24 hours 62
Hosts
With recent credit 2044
Registered in past 24 hours 15
Current GigaFLOPS 35869.59

Dataset 1 and 2 progress:

SingleDirectMachine      10002/10004
EightBitMachine           9999/10006
SingleInvertMachine      10001/10003
SimpleXORMachine         10000/10002
ParityMachine              690/10005
ParityModified             148/10005
EightBitModified          5445/10006
SimpleXORModified        10005/10005
SingleDirectModified     10004/10004
SingleInvertModified     10002/10002 

Last week's TWIM Notes: Sep 14 2020

Thanks again to all our volunteers!

-- The MLC@Home Admins

Great, thanks! Can you pls quickly elaborate on the design of the new dataset 3 table on the main project page as seen on https://www.mlcathome.org. What does the design represent or how should we interpret the color-coded table?

Good question! There are 100 cells, one for each random automaton (numbered 0-99). Y axis is tens digit, X axis is ones. The number in each cell is the number of networks successfully trained to mimic that automaton. Right now they're all 0/red... once WUs start to come in, we'll start to see numbers go up, until we have 100 in each cell. Red = 0 comp,ete, Yellow = some complete, Green = complete. So the table will slowly turn from red to green over time.

Just a way to show the data compactly versus datasets 1 and 2's tables. We'll be playing around with it some more to see if we can improve the representation going forward.

Keep up the great work! I am getting the new rand_automata_ workunits now. These are some big boys taking average of 6 hours 11min with estimated computation size of 60 TFLOPS, compared to about 1.5 hours and 23 TFLOPS with the previous types.

It seems like all my WU's have computation error straight away what would cause this?

Hmmm... First, thanks for volunteering and you're doing everything right (processor/ram/OS version are all fine). You're not the only one having this issue as it effects a portion of our windows users.

Our app is failing with "0xc0000138", which I believe is an indication there's some library/dependency issue. Unfortunately, the MLDS application can't be linked statically, so it has to ship with some dependent libraries, and it looks like those libraries are either insufficient or conflicting with existing libraries on your system. Also unfortunately, Windows isn't very helpful about saying what library or why, so its difficult to debug. It's also highly dependent on the host-specific environment. so its difficult to replicate in our testing. See e.g.: https://stackoverflow.com/questions/10349155/debugging-process-load-exception-0xc0000138-ordinal-not-found .

Are you a developer or willing to test a few things for us? If so, please open a thread under the "Windows" topic in the platform-specific problems message board and we'll continue there, we might get more testers there too.

One quick thing to try is to make sure you have the latest copy of the "vcredist runtime" installed from microsoft. We ship our app with our own copy of some DLLs from in there, but maybe if you have an old version and we try to run with a new version .. maybe that causes issues? It shouldn't, but worth a shot.. and if it does help, it'll help us narrow down the problem.

Again, sorry it isn't working for you!

Hi thanks for the reply - no I'm not a developer but I did try reinstalling latest vcredist to no avail - new WU errored again on start..

Also tried resetting project - still errored - removed project & reconnected - still errors..

Thanks for trying anyway. We'll continue working it on our end as best we can.