Each selected student will undertake the “Zerotha Foundation Program.” The first project will include: guiding you in reproducing an already existing paper without code. Later, we shall see how things go for you and if both have time. Zerotha is a pre-stage of any formal research, and involvement in Zerotha is entirely informal. Here, the idea is to train you in the preliminaries of research methodology- that is it. Usually, our students spend at least 8-12 months on the initial project. Please note, Zerotha students do not work with mentors in any collaboration (e.g., research articles) for an open-source project until they finish the two-phase foundation program. All mentors are also working full time in different organizations and can only allocate a few additional (voluntary) hour(s) on Saturday for mentorship (that too, not regularly). However, we take full responsibility of your effort and time, too. Hence, students are expected to be self-disciplined and you will get full support from us. Unfortunately, if you are looking for quick success or fast research pace, Zerotha is not for you. We still believe in the old-style research learned from our professors, which is slow and has all the time to discuss ideas, with no pressure to be a SotA.
Following are the key points of the foundation program:
- First Phase:
- The student will be part of a dummy research setup on a mutually agreed topic.
- In the first two months, a student will be assigned to open Github issues from ongoing research projects by different research groups.
- Students will then read papers in two-three related areas. You will then work to reproduce one-two papers and note the findings (little things/issues)
- After reading the literature and reproducing codes, the students will write a 5-page summary of their findings (ACM or ACL format).
- 90-95 percent of the student drop out in the first phase.
- The first phase normally takes 3-4 months.
- Second Phase:
- Mentors will select one of their already published papers in the domain of student’s interest. We maintain a 1:2 (student:mentor) ratio.
- The Students will be taught about the fundamentals of research: mentors will help the student understand why they did this work, giving their reference list to let the student read related papers. Mentors will then teach how they wrote different sections in the paper, their research hypothesis, and how they formalized it. Help the student in reproducing the work, why to chose a particular research setting, why only Adam optimization? Why a particular dataset, etc.- all the little things which are implicit. The student will work on re-implementing the work after understanding the idea. We focus a lot on problem identification rather letting you solve a pre-defined problem. Hence, the primary goal of the Zerotha program is to make you a “problem identifiers.” We believe that once you start identifying the right research problems, you will be able to solve them with your technical skills.
- The students will then write a dummy paper in their language on which mentors were using as a dummy setup. The student needs to submit an 8-page paper (ACM or ACL format) at the end of phase 2. We expect the student not to look at the mentors’ original published work, which was used as a dummy setup. Three reviewers will review the paper and provide the overall acceptance or rejection. In case of rejection, the student is required to submit a second version considering reviewers’ comments.
- 2nd phase normally lasts around 4-6 months.
- Special note for students residing in India/Africa/South-East Asia
- If communication in English is an issue for you, we speak Swahili, Gujarati, Hindi, Rajisthani, Marwadi, and Marathi (if this helps).
Before you approach us, you need to do a pre-requisite:
- Try making following code up and running: DCA.
- Ideally, reproduce the results (it’s reproducible). If not, List down the steps which you tried and why it did not work. List down your errors and steps you took to resolve those errors. Send us a Github link listing your effort. Also, write a two page summary of the paper and attach it in the email.
- Send us your Github link having previous open-source implementations. Strong coding skill is a mandatory requirement. We point you to the following video which we believe you must see as an aspiring researcher: Link
- Your email without the above two steps will not be entertained.
- Anery Patel (2019), next position: research intern in TIB, Hannover, Germany (she also got Masters admit from Northwestern University and NYU in computer science)
- Manoj Prabhakar (2019-2020), next position: PhD student, Hasso Plattner Institute, Berlin
- Anson Bastos (2019-2020), next position: PhD student in IIT, Hyderabad
- Abhishek Nadgeri (2019-Present), next position: Masters admit in RWTH, Aachen joining October 2020.
- Unknown (2020-Present)
- Unknown (2020- Present)
Two of our current students wanted to be anonymous, hence we do not put their name on the website respecting their privacy.
Besides Zerotha, there are several initiatives from Individuals and organisations to mentor students. We provide a list of resources that may help you.
- From Danish Pruthi, CMU: https://twitter.com/danish037/status/1302078153213513728
- From Cardiff NLP group: https://twitter.com/cardiff_nlp/status/1302875140208893952
- From Dr. Nishant Sinha- a virtual Lab for young researchers: https://github.com/ofnote
- From Victor Zhong- getting started with NLP: https://www.victorzhong.com/blog/getting-started-in-NLP-ML-research.html
- From Eric Jang: https://twitter.com/ericjang11/status/1274102047017582594
- From Dr. Shomir Wilson: https://shomir.net/advice_for_students.html
- The Dr. Isabelle Augenstein: The Big Directory – primarily people from underrepresented groups in NLP, as well as supporters who are interested in actively increasing diversity in the NLP community. You can reach out to many researchers from this directory who may be ready to mentor you.
- NLP with friends: a free space for talks by students in NLP: https://twitter.com/NLPwithFriends
- Some other resources listed in the Gihub page: https://github.com/OrthoDex/ml-research-starter-kit