Guest Post (Part II): Deep Reinforcement Learning with Neon says:

The deep learning ecosystem has evolved a lot since then. Supposedly a new deep learning toolkit was released once every 22 days in 2015. Amongst the popular ones are both the old-timers like Theano, Torch7 and Caffe, as well as the newcomers like Neon, Keras and TensorFlow. New algorithms are getting implemented within days of publishing.

Was a new deep learning toolkit released once every 22 days in 2015?

  • 1
    The original claim, which that article is just quoting and linking to, is here. That kind of phrase is often used to mean "17 deep learning toolkits were released in 2015"; that is to say, "a new deep learning toolkit was released for every 22 days in 2015".
    – Dan Getz
    Feb 11 '16 at 1:28
  • 1
    I don't think there is a notable claim here, it has become part of research culture to make research code available (e.g. jmlr.org/mloss). I suspect something similar could have been said about support vector machines back in the late 90s. I would hope the algorithms are implemented before publishing though! ;o) Feb 11 '16 at 7:57

Kyle McDonald wrote a Python script that dumps basic statistics information from GitHub for all the Deep Learning Toolkit projects that are included in this list.

This Reddit comment, from November 10, 2015, describes the script, and summarises the results:

From 2010-2015 a new toolkit was released every 47 days. This year, it's every 22 days.

The Reddit comment, and the toolkit list is maintained by a user called kcimc, which is presumably the same Kyle McDonald as the script writer.

Some clarifications:

  • This is a count of the releases, not a count of the total number of toolkits. So, a new release of a toolkit occurred, not necessarily the release of a new (novel) toolkit.

  • This is a standard English idiom that means that on average there was a release for every 22 days. It does not imply there was a release exactly on day 22, 44, 66, etc.

While this wasn't published in a peer-reviewed source, the claim is fairly prosaic, and it isn't hard to reproduce the results by hand, so it seems reasonable to tentatively accept the claim, at least for the first 10 months of 2015.


There are 37 such toolboxes listed here which suggests the claim is entirely plausible in the sense of "17 deep learning toolkits were released in 2015" (see comment by @Dan Getz).

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .