Interview with Ishan Mishra
Where are you based?
I’m currently based in Waterloo, Canada. 🇨🇦
What do you do (i.e. studying, working, etc.)?
I currently work at OpenMined full time, where I have the privilege of leading the Differential Privacy team for PySyft.
I also graduated not too long ago from the University of Waterloo!
I spend most of my time and energy on differential privacy and building things with AI/ML, and would consider both of those to be my specialty.
I have a background in academic research from Harvard and Harvard Medical School in the past, as well as entrepreneurship (I worked at a VC firm in the Bay Area for a while, and my final year capstone project at University wound up turning into a startup that was featured on BBC, the Discovery Channel, and was an International Runner-Up for the James Dyson Award).
How and when did you originally come across OpenMined?
It’s actually quite a cool story!
One of my best friends and I came up with a startup idea that involved training machine learning models on data that was across different semiconductor fabrication facilities. I looked for resources about how to do this, and discovered this field was called Federated Learning, and found a post about OpenMined on HackerNews.
Not too long after, they advertised a job that entailed building out their end-to-end secure pipeline for remote data science. I applied, and a few minutes before the interview, I found out that the person interviewing me was the author of my favourite ML book (Grokking Deep Learning)!.
I jumped on the opportunity, and I’ve been working at OpenMined ever since!
What was the first thing you started working on within OpenMined?
I was building out our AutoDP library, and adding new functions to it, with the goal of making it as close to the NumPy API as possible. This way, people who used PySyft for remote data science could treat the Syft Tensors exactly the same way they would treat their NumPy arrays.
I remember being nervous when I started- I had never contributed to Open Source before. I think my first pull request was quite literally fixing a single linting error (I think I removed a semicolon or something)- but it was enough to break the psychological barrier and I tried to help out as much as I could.
Looking back, it’s remarkable how a supportive, kind and empathic community like OpenMined can help you grow so fast!
And what are you working on now?
I have the privilege of leading our collaboration with Twitter!
The aim of this partnership is to enable third-party access to non-public Twitter data using privacy preserving technologies. This would allow third parties to investigate things such as the amplification of political content by Twitter’s recommender systems- without exposing any user’s private data, or any of Twitter’s models or IP.
I believe this initiative will set a new standard for algorithmic accountability, fairness, and transparency, and will truly result in ML becoming a lot more ethical. We’re tremendously grateful for Twitter’s ML Ethics, Transparency and Accountability (META) team for driving this initiative, and I’m personally also very grateful to my entire team for their hard work and dedication in helping bring this vision to life.
(Please feel free to reach out over Slack at @Ishan if you’d like to get involved!).
What would you say to someone who wants to start contributing?
One of my best friends, and fellow team member at OpenMined, Ruchi Bhatia, is a 2x Kaggle Grand Master, and I’ve seen her fly through Jupyter Notebooks faster than Homer Simpson goes through a plate of donuts. (Only a slight exaggeration!).
She told me that as part of her New Year Resolution last year, she worked on Kaggle every single day for the whole year, even if it was just to leave comments on a discussion.
That’s the approach I’d recommend to people who want to start contributing- figure out a commitment level that lets you have a healthy balance, and that you feel comfortable with. And in that allotted time, try to grow, learn and help out as much as you can! And if you have to step away for a while to take on exams, or paper deadlines, or something else- that’s totally fine. I guarantee that you’ll have a lot of fun along the way, and will be joining an incredible community with a lot of exciting projects going around.
“The best time to get started was yesterday. The next best time is now.”
Please recommend one interesting book, podcast or resource to the
For learning about Differential Privacy, I’d recommend the book Programming Differential Privacy, this blog series by Damien Desfontaines, and Professor Gautum Kamath’s course on DP at my alma mater, the University of Waterloo.
All of these resources are available for free!
Please do feel free to reach out! I love meeting new people. You can reach me on our Slack workspace at @Ishan :)