Deep Learning refers to set of techniques for learning in Neural Nets. So does the statement "reimplementation of Deep Learning algorithms replicating performance from papers"
mean implementing these algorithms using which Neural Networks learn their weights and measuring do they perform as well as said in the research paper? or implementing a model that some research paper suggests and replicating its results?

Context: OpenAI Machine Learning Fellow position demands this in an applicants profile state exactly in the following words:

"We look for candidates with one or more of the following credentials:
Open-source reimplementations of deep learning algorithms which replicate performance from the papers"

Therefore, What is it that they are technically looking for?


Accompanying ML/AI (and computer science, in general) papers with the relevant code is much desirable, both for reproducibility of the paper results as well as for faster integration of these results into applications.

Some researchers do so by themselves; but this has lately become also a community job, where a third, independent individual or team, not affiliated with the paper author(s), offers a code implementation starting from the paper.

A good, very recent case is the reproduction of the World Models paper (blog post, Github repo). Many of the implementations available at GitXiv are indeed provided by the community.

A great write-up of such an effort, containing many lessons learned, made big raves recently: Lessons Learned Reproducing a Deep Reinforcement Learning Paper.

Understandably, individuals that have such efforts in their CV are a good recruiting option...

  • $\begingroup$ In addition to this, do you find the stated "statement (Open-source reimplementations of deep learning algorithms)" demanding for reimplementation of some Machine Learning Model's based on some paper or the reimplementation of the very learning techniques themselves (e.g. back-propagation)? $\endgroup$ – Shoaib Anwar Apr 18 '18 at 19:13
  • $\begingroup$ @ShoaibAnwar probably the former; a re-implementation of BP is not of any particular value, apart from a purely educational perspective, and such efforts abound $\endgroup$ – desertnaut Apr 18 '18 at 19:17

I believe that this:

  • "reimplementations of deep learning algorithms" He is asking that they are looking for people who have made and completed an AI that perform similarly or exactly as those given in papers (IDK which papers)
  • "Open-source" The Fellow wants to be able to see the source code of the project.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.