At the ReworkDL conference in Boston last month I listened to a fantastic presentation by Ryan Adams of Whetlab on how they’d created a business to add some science to the art of tuning deep learning engines. I signed up to participate to their closed beta and came back to the UK very excited to use their system once I’d got my architecture in place. Yesterday they announced that they had signed a deal with Twitter and the beta would be closed. I was delighted for the team – the business side of me is always happy when a start-up is successful enough to get attention of a big corporate, although I was personally gutted as it means I won’t be able to make use of their software to improve my own project.
This, for me, is a tragedy. The key purpose of the Whetlab software was to put deep learning back into the hands of everyone – taking it away from the select group of “Hinton Post-Grads” that are experts in this area. While there are off the shelf solutions (Metamind, H2O, Skymind etc) these still have limitations – there comes a point where you just can’t get the results you need without opening the black box of deep learning and tinkering to get the accuracy.
Without products like Whetlab, the ability to do this tinkering will be either slow product of trail and error or owned by the few with experience – who are being bought out by the big players as soon as they’ve shown success. The exciting new technologies that should be advancing all sorts of solutions are gradually being assimilated by a few search engines and social media platforms1, staff who regularly shared their research embargoed from speaking in anything other than abstract terms2 and all this just to ensure that the most relevant advert3 clogs our browsers.
I hope some aspect of the Whetlab solution remains accessible to the community, or that a similar solution springs up to replace it. Deep learning has so many potential applications and problems it can solve and I believe it is important it stays accessible.
Update: If you are an academic or non-commercial institution then you can make use of their academic code at https://github.com/Hips/spearmint – sadly, this is of no use to me but I’m leaving this here so others who may be covered by the license can at least make use of this.