So I've been trying to come up with a method to reimburse users for the "monster" workunits that some of you have been experiencing. Unfortunately, by the time these come in many of them have been removed from the database so there is not much I can do in terms of reimbursing individual users for that work lost.

That being said, what I'm going to do is on this coming monday, May 8th, I'll provide a reimbursement to each user with credit from the "EXACT MNIST Convolutional Neural Network Trainer" application. I'll provide a one time award to everyone for 15% of the credit they've generated by that application. This was the best I could come up with as the users who experienced the most "monster" workunits also would have crunched the most workunits as a whole.

That being said, if you're experiencing some workunits from the "EXACT MNIST Convolutional Neural Network Trainer" application that are seeing pathologically long runtimes (multiple days), please feel free to abort them, or any workunits that won't be completed before Monday, May 8th for that matter. After the 8th I will still leave the validator up and going for this application to collect any stragglers, but this award will be one time only based on credit for that application as it stands on Monday.

Sorry for all the trouble these have caused and I hope this is a reasonable solution!

More...