What is MLC@Home?

MLC@Home is a distributed computing project dedicated to understanding and interpreting complex machine learning models, with an emphasis on neural networks. It uses the BOINC distributed computing platform. You can find more information on our main website here: https://www.mlcathome.org.

Neural Networks have fuelled a machine learning revolution over the past decade that has led to machines accomplishing amazingly complex tasks. However, these models are largly black boxes: we know they work, but they are so complex (up to hundreds of millions of parameters!) that we struggle to understand the limits of such systems. Yet understanding networks becomes extremely important as networks are deployed in safety critical fields, like medicine and autonomous vehicles.

MLC@Home provides an open, collaborative platform for researchers studying machine learning comprehension. It allows us to train thousands of networks in parallel, with tightly controlled inputs, hyperparameters, and network structures. We use this to gain insights into these complex models.

We ask for volunteers to donate some of their background computing time to help us continue our research. We use the time-tested BOINC distributed computing infrastructure — the same infrastructure that powers SETI@home's search for alien life, and Rosetta@home's search for effective medications. BOINC is fun — you get credit for each bit of compute that you do, with leaderboards and milestones. All while helping further open research. Please follow the link below to join, and happy crunching!

Join MLC@Home

Already joined? Log in.


[TWIM Notes] Aug 8 2020
This Week in MLC@Home
Notes for Aug 8 2020
A weekly summary of news and notes for MLC@Home


  • This week we focused on behind-the-scenes work
  • A new server for the project has been ordered buy may take a while to ship
  • ARM32 and ARM64 clients have been created in-house but more testing is needed. Performance on the Pi4 especially is quite reasonable.
  • We spent a lot of time this week trying to get both amd64 and ARM clients to compile statically to simplify deployment. However, the resulting binaries still have issues, and that path might not be an viable option.
  • We intend to release a preliminary version Datasets 1 and 2 to the public for your own data science analysis as soon as we have at least 1000 samples from each machine type. This will just be a CSV file of the weights for each trained network and the label of which machine type. When the full 10k for each is done, we're release a larger dataset that includes that plus the trained models and some more metatdata about each WU's training progress.
  • We're working on 2-3 potential different academic papers based on the work so far. One paper based on the analysis of datasets 1 and 2; another one (possibly combined with the previous) on using out weight-space analysis to detect trojan networks hidden in otherwise benign networks; and a possible third paper/talk on the engineering behind building a distributed computing project focused on ML. One conference we're targeting has a submission deadline of late September, so lots to do between now and then.
  • In relation to the above, we're working on an new version of the client that can support CNNs as well as RNNs. This will allow us to build WUs based on more widely available synthetic network data.
  • This week was a relatively quiet week in the forums.

Project status snapshot:
Tasks ready to send 39875
Tasks in progress 19815
With credit 448
Registered in past 24 hours 20
With recent credit 1428
Registered in past 24 hours 19
Current GigaFLOPS 23319.4

Dataset 1 and 2 progress:

SingleDirectMachine       9679/10004
EightBitMachine           9626/10006
SingleInvertMachine       9684/10003
SimpleXORMachine          9647/10002
ParityMachine              305/10005
ParityModified              29/10005
EightBitModified           795/10006
SimpleXORModified         6077/10005
SingleDirectModified      6039/10004
SingleInvertModified      6130/10002 

Thanks again to all our volunteers!

-- The MLC@Home Admins
8 Aug 2020, 14:48:10 UTC · Discuss

[TWIM Notes] Aug 1 2020
This Week in MLC@Home
Notes for Aug 1 2020
A weekly summary of news and notes for MLC@Home.


  • Happy 1 month anniversary! MLC@Home launched June 30th, so yesterday marked the 1 month anniversary. Thanks to all contribute!
  • We took a preliminary look at Datasets 1 and 2,and the results were encouraging. We are able to differentiate the nearly identical machines in Dataset 1 and Dataset 2 based solely on the learned weights, and we were able to plot the weights in 2D weight space, and see some separation/clustering. This leads to come interesting follow-on questions and suggests maybe loss isn't the only way to evaluate a network (thread here)
  • Linux armhf/arm64 support is in progress. 32-bit client it up and running in testing. Follow along with progress here: Linux/armhf and Linux/arm64 support status thread.
  • Dataset 3 continues in internal testing. Should be ready in a few more weeks.
  • The deadlines for new and follow-on WUs has been moved to 4 days instead of 2. This should help the boinc client make some better decisions about scheduling work. However, many already queued work units still have the 2 day deadline, so this will not go away completely.
  • New server will likely take at least several weeks.

Project status snapshot:
Tasks ready to send 62154
Tasks in progress 21940
With credit 382
Registered in past 24 hours 13
With recent credit 1305
Registered in past 24 hours 21
Current GigaFLOPS 23473.01

Dataset 1 and 2 progress:
SingleDirectMachine 9670/10004
EightBitMachine 9618/10006
SingleInvertMachine 9675/10003
SimpleXORMachine 9633/10002
ParityMachine 269/10005
ParityModified 15/10005
EightBitModified 506/10006
SimpleXORModified 3106/10005
SingleDirectModified 3078/10004
SingleInvertModified 3156/10002

Thanks again to all our volunteers!

-- The MLC@Home Admins
1 Aug 2020, 18:07:29 UTC · Discuss

[TWIM Notes] Jul 25, 2020
This Week in MLC@Home notes for Jul 25, 2020

A weekly summary of news and notes for MLC@Home.

  • We've more than doubled the number of host participating in the project, up to 1100 active hosts!
  • Relatedly, MLC@Home Is now listed on the official BOINC project page
  • We've started processing some of the data, for Dataset 1, and hope to have some preliminary results to share later this week
  • Datasets 1 and 2 are continue crunching away, see the live status on https://www.mlcathome.org/
  • Dataset 3 is in internal testing, and will be even larger. Expect a more complete writeup of what these results are soon
  • The project has secured a Raspberry Pi 4, and will be working on an ARM client. No timeline available yet.
  • You can now follow MLC@Home on Twitter: @MLCHome2
  • Lots of discussion on the forum related to validation errors for a small number of results (less than 1%), but are being addressed as they are uncovered, see the forum threads under "Issue Discussion"
  • A new server is being ordered!

Current work status at time of posting:
Tasks ready to send: 81035
Tasks in progress: 15139

As always, you can see the current project status on the project main page at https://www.mlcathome.org/.

Thanks for contributing!
-- The MLC@Home Admins
25 Jul 2020, 19:29:39 UTC · Discuss

MLC@Home and CORAL
Exciting news!

MLC@Home is now a project of the The Cognition, Robotics, and Learning (CORAL) lab at the University of Maryland, Baltimore County (UMBC).

This brings many benefits, not the least of which is lab space to run equipment, new ideas for research to conduct, and a place to collaborate with other researchers who may not be members of the BOINC community. Most importantly, being backed by a university research lab provides some assurance to you, our volunteers, that this project has both academic merit and a purpose beyond a single researcher and their goals. Over the next few days I'll be updating our website to reflect the new affiliation.

As always, please see our main website, https://www.mlcathome.org for details on the current research, or join us by attaching your BOINC client to this URL: https://www.mlcathome.org/mlcathome/
18 Jul 2020, 2:53:29 UTC · Discuss

New testing WUs released
As the first round of 50,000 networks finishes training (what I'm referring to as "dataset 1"), I've released a few new test WUs for the next round ("dataset 2"). If these go well, I'll release the next 50,000 networks to train, that should keep keep the WUs flowing a bit more consistently.
12 Jul 2020, 6:17:41 UTC · Discuss

... more

News is available as an RSS feed   RSS

©2020 MLC@Home Team
A project of the Cognition, Robotics, and Learning (CORAL) Lab at the University of Maryland, Baltimore County (UMBC)