Loading
Cookies help us deliver our services. By using our services, you agree to our use of cookies. Learn more

NIPS, Minecraft, and VizDoom

29 May 2018

4-up-images

This summer we have a lot of exciting things planned on crowdAI.

We will start with the VizDoom Challenge, where you are tasked with training an Agent which can finish randomly generated Doom levels using data normally available to humans (frames as RGB images), and without using any auxiliary information (depth maps, or in-game states, etc). Another version of the same challenge will pit all submitted agents against each other in a deathmatch on unknown maps, and the best frag collector , emerges victorious in bloody battle !

deathmatch

Then we will be starting the NIPS 2018 AI for Prosthetics challenge, which is the reincarnation of the very successful NIPS Learning to Run Challenge, co-organised again with our collaborators from Stanford. This time, the simulated model will have a prosthetic leg attached, and the task is to train an agent capable to learning policies necessary to walk in an unsupervised way. The model(s) which covers the maximum distance in the given time horizon wins ! Future applications of this problem can help simulate the gait of a person with different prosthetic legs attached.

Then follows the NIPS 2018 Adversarial Vision Challenge, that we are co-organizing with Bethgelab and Google Brain. The participants can take on the role of an attacker or a defender (or both). As a defender you are trying to build a visual object classifier that is as robust to image perturbations as possible. As an attacker, your task is to find the smallest possible image perturbations that will fool a classifier. The challenge hopes to collaboratively come up with a much better understanding of how do we build models which are much more robust adversarial attacks.

Illustration Adversarial Examples

Finally, we also have the MARLĂ– challenge in collaboration with Microsoft Research, which is a Multi Agent Reinforcement Learning Challenge on Minecraft. Where participants will have to submit agents which can collaborate and compete with agents submitted by other participants to solve a variety of tasks in Minecraft, from competitive Build Battles to Collaborative Pig Chases, we have got it all covered. If you love Minecraft, you will definitely love this challenge !

minecraft

We are excited about all these challenges, and are eagerly looking forward to all the amazing things we will discover together by trying to tackle all these challenges together ! If you are new to Machine Learning we will try our best to help you learn as much as you can; if you are a Machine Learning veteran, we will ensure that you never get bored ! And if you want to sponsor prizes for any of these challenges, please do not hesitate to get in touch with us.

Cheers, Mohanty