Khimya Profile picture
Jul 2, 2020 4 tweets 3 min read Read on X
Humans have a remarkable understanding of which states afford which behaviors. We provide a framework that enables RL agents to represent and reason about their environment through the lens of affordances bit.ly/31qjlSv

#ICML2020 paper from my internship @DeepMind 1/4
In this work, we develop a theory of affordances for agents who learn and plan in Markov Decision Processes. Affordances play a dual role. On one hand, they allow faster planning. On the other hand, they facilitate more efficient learning of transition models from data. 2/4
We establish these properties through theoretical results as well as illustrative examples. We also propose an approach to learn affordances from data and use it to estimate partial models that are simpler and generalize better. 3/4
Joint work with an amazing set of collaborators @zafarali , Gheorghe Comanici, @dabelcs , and Doina Precup. Hope to "see" you at @icmlconf! cc @MILAMontreal @rllabmcgill 4/4

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Khimya

Khimya Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @khimya

Sep 26, 2020
“X papers accepted at Y”
~ What people don’t tell about paper acceptances! ~ A thread #AcademicChatter #phdlife
1/N Many papers that you see accepted now would have been very likely a second or a third submission! That doesn’t mean they are not worthy instead they actually took a lot more work including feedback from multiple reviewers, revisions, skilled rebuttals & perseverance!
2/N Yes, many would have made it in first attempt! They are likely the result of dedicated collaborations with some including structured teams! Again not a bad thing, but don’t feel bad because you are comparing this to you toiling the night oil all by yourself+advisor!
Read 17 tweets
Sep 22, 2020
Applying for grad school? Sharing here a few pointers on grad school applications! Pl feel free to share w/ folks who might benefit! Disclaimer: I am not an expert! I came across some questions from different people and I see a pattern, so here we go: #AcademicChatter 0/N
Where to apply? How to do this? Start with a list of schools 3 dream schools, 4 would love to go, 3 safe( will get in most likely)! Google sheets, to do lists, color codes, then marie kondo the hell out of this process! You got this! 1/N
Webpage! I cannot emphasize how nice it is to see a candidates webpage with all information consolidated at one place! I have some students tell me that they don’t have enough content to create a webpage! You need it even more! 2/N
Read 22 tweets
Aug 8, 2020
Why is there zero *explicit* training on almost everything that matters a lot in academia except research? Be it reviewing, writing, communication, presentation skills, or even networking. They are left for each to hone & champion majorly by themselves! 1/N
The advisor and peers ofcourse help one develop such skills and get better at these over time! But it is often after many attempts people in later career stages say “I wish someone told me this earlier..” I don’t even believe the entire burden should be on the advisor alone! 2/N
For instance, we complain about reviews, but how much training is given to anyone for reviewing? Poor quality review comments like “sota chasing” “not novel enough” “not deep enough” and even the scale on the reviews is not very helpful for anyone! 3/N
Read 6 tweets
Jul 2, 2020
Humans have a remarkable understanding of which states afford which behaviors. We provide a framework that enables RL agents to represent and reason about their environment through the lens of affordances bit.ly/31qjlSv

#ICML2020 paper from my internship @DeepMind 1/4
In this work, we develop a theory of affordances for agents who learn and plan in Markov Decision Processes. Affordances play a dual role. On one hand, they allow faster planning. On the other hand, they facilitate more efficient learning of transition models from data. 2/4
We establish these properties through theoretical results as well as illustrative examples. We also propose an approach to learn affordances from data and use it to estimate partial models that are simpler and generalize better. 3/4
Read 4 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(