Thommen George Karimpanal

Lecturer in AI, Deakin University

[HARL_workshop @HarlWorkshop]

Its been a while, but we finally have the videos from our workshop @ICRA2024 ready: https://t.co/bvOkpK2LEY

[HARL_workshop @HarlWorkshop]

Thanks to all the invited speakers, panel members, paper authors and participants for a great 2nd edition of the workshop!

[HARL_workshop @HarlWorkshop]

Do send in your questions here: https://t.co/AMEPoImxv0

[Thommen George Karimpanal @thommengk]

Please send in/post questions you would like to see being discussed during the panel discussion session at our workshop on 17th May at ICRA 2024 !

[HARL_workshop @HarlWorkshop]

We are collating questions for the panel discussion session to be held during our Human-aligned RL workshop (https://t.co/2Rzk1vI8Wq) on 17th May at @ieee_ras_icra . What topics/questions relating to human-aligned RL for Robotics would you like to see being discussed?

[Thommen George Karimpanal @thommengk]

We look forward to receiving your best works!

[HARL_workshop @HarlWorkshop]

If you are still considering submitting your work on Human-aligned RL, we have good news- we have decided to extend our submission deadline by an additional week! The new deadline is on 11th March, 23:59 AoE. We do not anticipate any further extensions.

[HARL_workshop @HarlWorkshop]

The workshop on Human-Aligned RL for autonomous agents and robots will be held on 17th May 2024 at ICRA 2024 @ieee_ras_icra in Yokohama, Japan. Please find all details here: https://t.co/QOFJuocDnI

[Mayukh Das, PhD @mayukhdastweet]

Consider submitting your article to the new spcl issue/research collection on Knowledge-guided Learning & Decision-Making in Frontiers in AI, that I am editing along w/ @devendratweetin @thommengk & @janadoppa
#human_in_the_loop #ArtificialIntelligence

https://t.co/mZzJskmzJk

[Thommen George Karimpanal @thommengk]

Happy to be recognised as a top reviewer in NeurIPS 2022! (https://t.co/1MrvpAmgQO). I hope more conferences and journals make steps towards acknowledging reviewers' efforts.

[Thommen George Karimpanal @thommengk]

If you are interested in pursuing post-PhD research in machine/deep/reinforcement learning, please consider applying: https://t.co/oXUFC9DToA

[HARL_workshop @HarlWorkshop]

Thanks to all the participants for contributing to a successful HARL@ICDL 2021 workshop! We have made all the videos available on our YouTube channel here: https://t.co/EH8uRpWOzX
@ICDL_EpiRob

[Thommen George Karimpanal @thommengk]

Prof. Mahadevan's interesting take on the state of machine learning research today. https://t.co/3slmcDtnrk

[Thommen George Karimpanal @thommengk]

Sharing our recently published work on automatically incorporating domain priors into the reinforcement learning framework:
https://t.co/VXEwCs0F4p
Arxiv link: https://t.co/lvpXGf9gGe

[Thommen George Karimpanal @thommengk]

Is machine learning all hype or solid science? by Sridhar Mahadevan https://t.co/M4wZMSQW4x

[Santu Rana @santura]

Half-way through 2020, @DeakinA2I2 has 11 papers accepted at top-tier international conferences in AI, Machine Learning and Computer Vision: one in CVPR, two in ICML, two in ICLR, two in AAAI, two in AISTATS, and two in IJCAI. https://t.co/kC7hIt9DMi

[hardmaru @hardmaru]

There’s also a disincentive for inventors of an algorithm to thoroughly compare its performance with others—only to find that their breakthrough is not what they thought it was.

“There’s a risk to comparing 𝘵𝘰𝘰 carefully.”

https://t.co/OOAq2qMJie

[Thommen George Karimpanal @thommengk]

Plenty of interesting projects for PhD aspirants in the areas of reinforcement learning and deep learning.

[hardmaru @hardmaru]

Learning Human Objectives by Evaluating Hypothetical Behavior

“One benefit of our method is that, since it learns from synthetic trajectories instead of real trajectories, it only has to imagine visiting unsafe states, instead of actually visiting them.”

https://t.co/xqjuC9YiyY

[hardmaru @hardmaru]

The agent first learns a world model driven only by exploration, without task-specific rewards.

It then receives rewards to adapt to multiple tasks, such as standing, walking, running, flipping using either zero or only a few tasks-specific interactions.

https://t.co/5ccxqdYxUB

[Thommen George Karimpanal @thommengk]

"We have to learn the bitter lesson that building in how we think we think does not work in the long run." https://t.co/AnCwr4bNIE