should you work on this idea ?

So, through some wild rituals, you divined a hypothesis — a short, interesting, and unproven statement. Is it worth your time to pursue it further and whole heartedly dedicate your next few months on it? In this chapter, we’ll cover how to vet your research idea before working on it. Unlike the chaotic process of coming up with an original research idea, one can easily describe the process of vetting an idea (or other professional endeavors), even though implementing this process requires significant discipline and skill.

minimize backtracks

The worst thing that can happen to a researcher is realizing, only after extensive…

voodoo rituals to procure research ideas

For the sake of everyone involved in research, I hope there isn’t a tractable algorithm we can run to get creative research ideas: That’ll poop the party for everyone! However, there are some unreliable maneuvers you can take to facilitate interesting ideas, like voodoo magics, and this chapter will focus on that. However, take whatever I write here with a giant slab of Pakistani pink salt, as it is an inherently a chaotic process that resists characterizations.

finding the mojo

This could be wrong, but I think –or at least, find it helpful to think– that creativity is mostly functional, in a sense…

a look behind the scenes

I have been planning to write this series for a while now, and the beginning of 2021 is an arbitrarily auspicious time to start. This series serves two purposes: First, as an account/guide on the process of developing original research for publication, written by someone who has done it a few times. Second, to share some funny stories of my time as a PhD student / postdoc at MIT, and keep it very personal. I remember Philip Guo’s the PhD grind being very informative when I started grad school, and I wanted to make something similar for other PhD students…

learned representation in the style of w2v

Given 2 objects, A and B, what makes them similar? There can be many answers, but one of the most elegant answer is based on substitution :

A and B are similar, if in most use cases of A, you can substitute it with B, and you wouldn’t notice much of a difference.

The idea of substitution has popped up multiple times, the ones I’m aware of are the liskov-substitution-principle, and of course, distributional-semantics. But instead of research literature, we’re just gonna do it with Clash Royal cards instead. It’s going to be a lot more relaxing and accessible.

Clash Royal


a dataset of crosses that serves no ulterior motives

pytorch or tensorflow ?

At ICML 2019 I asked “Pytorch or Tensorflow ?” at arbitrary passer-bys. Depending on their response, I drew a “x” on either side of the notebook. After 1024 queries I have obtained the above picture.


This was the question I got the most while talking with people. There were some conceptions that I was doing it for a research, or if I worked at facebook or google. I joked that I was collecting a dataset of crosses, or for fun, or to simply watch people’s reactions. …

Subverting traps in Human vs AI games

As of last month, [AlphaStar], an AI player, beat 2 top Starcraft players with a score of 10–1. Did AI conquer Starcraft the same way it conquered Go? Is AlphaStar actually SkyNet? Or are we just falling into some avoidable traps?

In this (my second :D) blog, we will explain which games have unbeatable AIs, and which games do humans still have a shot of winning. We will explore what it takes for a game to be “solved” and the gaps still lacking in AI approaches. No click baits, no mindless hypes, no needless jargons. Enjoy.

Note: All external links…

In this blog ( my first blog on medium o_o ), I will explain the challenges in making a Dota bot, appreciate how OpenAI addressed these challenges, and lay out its fundamental flaws. I hope that by explaining how it works in layman terms, we can watch its upcoming matches with the right mindset.

The OpenAI bot is trained like a pigeon: It is conditioned to associate short-term objectives such as last hits and surviving with positive reinforcements. By accomplishing a sufficiently large number of these short-term goals, the bot wins the game by coincidence, without planning its victory from the start.

The OpenAI Five Problem

Every project needs the right problem statement. I believe the OpenAI Five’s problem statement as follows: Beat a team of human in Dota in any way possible with a program. This view is both powerful and liberating.

It is powerful as it creates a tangible spectacle…

Fundamentally speaking, the OpenAI bot is trained like a pigeon. It is to our advantage to respect the bots’ feats of acrobatics while simultaniously exploit their lack of plannings and judgements.

I have been contemplating on starting a blog, with the recent OpenAI Five benchmark, it seems like a good time to start. The bots defeated the humans convincingly 2–0, but I think the bots can be beaten with the right approach.

Understanding Precedes Victory

In the moment when I truly understand my enemy, understand him well enough to defeat him, then in that very moment I also love him. — Ender’s Game

There can be no convincing victory without understanding. Similarly, I would like to divide this topic into two 2 parts:

Understanding OpenAI Five

In this more academic part, I will focus on explaining the…

some entries here would be nice

# does title work like markdown ?

Apparently not.

I am not sure what happened here

Evan Pu

Research Scientist (Autodesk). PhD (MIT 2019). I work on Program Synthesis in the context of Human-Machine Communications

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