The Latently Deep Learning Certificate offers self-motivated participants with the opportunity to build state of the art deep learning models and gain industry experience. We provide all the raw materials for success including the environment, access to a community (Slack), mentorship (Brian, Angie and the community), and a curated library of the hottest research publications on deep learning (access to our Paperpile). Over the course of 90 days, participants independently choose and build a deep learning model. A model implementation can be completed at any time during the course of the program. Successful participants are those who commit to the program and effectively manage their own time and effort, demonstrating their engineering maturity. Hot and historically important models are the highest priority, but all models are eligible. You may earn multiple certificates, but we reserve the right to decline to accept trivial implementations or trivial modifications of existing implementations.
In exchange for your implementation you'll earn 5% * 1/N phantom stock in Latently, where N is the number of certificates issued, payable on change of control of the company, acquisition or IPO. No phantom stock will be issued until the program is considered a success, which is defined as 1000 certificates issued. So invite your friends :) The program will continue indefinitely once 1000 certificates are issued.
As part of the certificate program you'll gain valuable experience and access to our 16 GPU cluster provided by IBM. If you successfully implement the paper or patent and we approve the implementation we'll also issue you a formal certificate that recognizes the engineering maturity required to successfully replicate a paper or patent and you'll also have priority access to interview for salaried positions when they open up in the future.
Code must be developed in public on our GitHub and released using The Unlicense (all code is in the public domain). Latently reserves the right to change these terms to help ensure the success of the program and company. Latently reserves the right to issue the Latently Deep Learning Certificate for other purposes, not limited to novel implementations of algorithms, collaborations, multiple certificates for one difficult paper or patent, etc.
Ideally, a successful implementation of a paper or patent implies writing code that replicates the findings in that paper or patent. However, due to hardware constraints this may not be possible and so the nearest reasonable approximation may be acceptable if you get permission from us. If a certificate is issued for an implementation that was not a replication, that paper or patent may be eligible for further development that results in a certificate being issued so that replication can be achieved.
- Code must be written in Keras, Sonnet or Tensorflow. In the case of Keras, Tensorflow must be the backend.
- Code should be factored out into a resuable library that makes it simpler to implement new papers.
- If existing implementations of a paper exist, you must not look at them. All code must be implemented with a cleanroom mentality.
- Code should either be written on your personal laptop or computer or on Latently's GPU cluster. Do not write code for this project at work or at school.
- Code is subject to code review. Commits should be small and frequent.
- Repositories for specific papers should be named as follows: Author1Author2Author3EtAl1996_1. In other words, the last name of the first three authors followed by EtAl if there are more than three and followed by the year. If more than one paper has that code then add an underscore and increment a counter that starts at 1.
- Dependencies are discouraged.
- You may not depend on libraries with viral licenses.
- Why are you doing this?
This project increases the talent pool for AI/ML which benefits both engineers and companies. Additionally, by abstracting implementations into libraries that implement the publically available literature we can more easily see what is patented and what is not in addition to discovering prior art that can be used to invalidate patents. It also helps new inventors know what inventions are OK to incorporate into new inventions and what inventions they will need to get permission to use.
- Why The Unlicense? What is The Unlicense?
We chose The Unlicense because we wish to place no additional restrictions on how the inventions are used and The Unlicense puts the code into the Public Domain. The primary purposes of the code is to document what's been done and to give engineers an opportunity to grow their skills.
- Can I use the code?
The code is in the Public Domain, however, there may be additional restrictions on its use as documented in patents and otherwise. If you use this code in your business you do so at your own risk - Latently assumes no responsibility.
- How do I join?
Contact brian@latent.ly with your resume.