My First Paper and the birth of TReX

Introducing TReX

My desk in Microsoft Research :)

The article describes a bit about how i got into research and how i started building TReX to help myself and all researchers out there.

July 2019, i got my offer letter from Microsoft Research to join them as a Research Intern. Couldn’t be more happier. Didn’t really know that i was going to work on a paper. Had little idea about how research works. I read research papers back then too. A lot of them. Didn’t understand many things but i kept reading them anyway. Point is, i had zero idea about the fact that i can also write papers.

To me it was a normal internship when i got the offer, till i spent some time there.

My love for research became more definitive. And i started wishing to be a researcher one day. Had a clear route ahead of me.

Met people and made friends, wish i could chat with everyday.

Ah. It gets very nostalgic, as im writing this in December 2020 in the covid pandemic, whilst having to spend the last 6 months of my term in MSR as Work-From-Home. I so wish everyday, that i could spent the last 6 months going to office and be among these great minds.

Well, coming back to what this article is about. How TReX was born. And how as i worked towards my first paper, i experienced the points that i believe if automated would make the life of every researcher simpler. And not just simpler, i believe its a necessity.

My goal with TReX is to have researchers focus on the actual research side of things, and not spend time on the logistics, like on which server an experiment is running, what is the current value of metric x for the experiment y. Is experiment x performing better than y. Having to put the outputs of experiments side-by-side for comparison, for putting in the paper. Keeping track of logs to know when to kill an experiment. Collaborating in an efficient manner about the things that needs to be done. Vizualizing the inputs outputs of the experiment. Keeping a track of where all the files are for different experiments. And many many more.

I faced these things to some extent as i was working on my first research paper. I used to write down tables on my table (pun intended). Keeping a track of what the status of experiments are and on which servers which experiment is running. And note, all these things while very important to research takes a lot of time, which at the end of the day can be automated, and which is exactly what my aim is with TReX.

To have everything organized in one place.

Went on happy about life in MSR. Working on the paper, solving bugs, observing what can be done better in the problem we were trying to solve, running experiments, facing the challenges i mentioned above and handling them as well. Hoping to do a PhD next.

From March 2020, i slowly started realizing that i wanted to do a startup. (Why you ask? Story for another blog.)

Generally, i started asking myself, what product do i want to create.

Almost immediately, i knew i wanted to work on a Augmented Reality Glass. That was the most challenging and coolest and the most useful thing that i personally need as i go about my everyday life. And i believe would create a tremendous impact on the society if done the right way.

I listed out all other things i think has potential and im interested in. I listed out pros and cons. Long story short, i reasoned out why i wanted to work on glasses.

I doodled this randomly on May.

You may ask. Wait. What happened to TReX.

I was working on my paper full-time. I didnt get much time. I used to keep a check of who the competitors are for the AR glass and what they are upto. But thats it. I didnt start working towards it.

Fast forward to July, i realized i didn’t have much experience in startups. And it would be best to start with something completely software only compared to a consumer electronics product, considering i’m new to hardware it would take me some time to build and get to the MVP stage. And there’s nothing in the whole wide world that i needed a tool more for than tracking my research experiments. Even when i would do the vision research for the glass, i would need a tool. Thus, decided on starting off with it.

My term ended on August in MSR. And then, i started writing the code for TReX, i think from Mid September, 15th to be exact, if im not wrong.

Working on the System Design for TReX.

I realize i have a tendency of naming things, and then not getting the time to work on it (a judgement based on my personal projects i did back in my undergraduate years). So, I didn’t name it earlier. And i started thinking of names only after i started working on it and started telling people about it.

After some considerable amount of pondering, the name TReX hit me and I knew that was it. (Wish i would know exactly how in my unconsciousness i came up with that name.)

If you dont already, T-racking Re-search e-X-periments.

Yes, and that’s how TReX came into life. Not sure about the Jurassic one though.

Here’s a video introducing TReX.

A short video introducing TReX.

Before ending, heres a lil bit about the paper i was working on, it was Training Conditional GANs with fewer paired samples. We worked on a Training algorithm for Conditional GANs to generate real looking faces using just 5 paired examples. You can find more details here: https://arxiv.org/abs/2010.12622.

Hoping all the researchers in both academia and industry would benefit a lot out of TReX. And not just researchers, we are aiming Data Scientists too. Feel free to get in touch with me over social.

PS. Im working towards the Glass too :)

https://twitter.com/amArunava

Machine Learning | Python

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store