392247 ISY Project: Deep Learning Architectures for AI (Pj) (SoSe 2019)

Following the success of the project in 2018 [a, b, c] we will continue development of a cognitive agent for challenging tasks in simulated environments.

[a] https://twitter.com/MSFTResearch/status/1099006255857713153
[b] https://youtu.be/aqUzh_jHSpY?t=1159
[c] http://github.com/crowdAI/Marlo

The emerging field of deep reinforcement learning has led to remarkable empirical results in rich and varied domains like robotics, arcade games, and multiagent interaction.
In the project, you will gain practical experience with Deep Learning Architectures for AI in simulated environments, apply neural networks to learn and produce complex purposeful actions in the context of one of the following environments on your choice [1-13]

[1] https://youtu.be/XjsY8-P4WHM (Atari)
[2] https://www.pommerman.com
[3] https://gym.openai.com/envs/MontezumaRevenge-v0
[4] http://vizdoom.cs.put.edu.pl
[5] https://www.crowdai.org/challenges/marlo-2018
[6] https://www.crowdai.org/challenges/neurips-2018-ai-for-prosthetics-challenge
[7] https://youtu.be/Ul0Gilv5wvY (PFNN)
[8] https://youtu.be/vppFvq2quQ0 (DeepMimic)
[9] https://github.com/MarkPKCollier/NeuralTuringMachine
[10] https://github.com/brendenlake/BPL (one-shot learning)
[11] https://github.com/Unity-Technologies/obstacle-tower-env
[12] https://www.youtube.com/watch?v=-L4tCIGXKBE (Differentiable Physics for Tool-Use and Manipulation Planning)
[13] https://www.crowdai.org/challenges/neurips-2018-ai-for-prosthetics-challenge

Literaturangaben

https://rebrand.ly/DRL