I would like to invite you all to a DeepRacer Challenge with exciting prizes and interesting challenges (and a novel AWS DeepRacer training stack!). Round 1 is already live but there’s time until October 31st to participate. Read on for details!
NeurIPS is a Neural Information Processing Systems conference with quite a bit of history:
The conference was founded in 1987 and is now a multi-track interdisciplinary annual meeting that includes invited talks, demonstrations, symposia, and oral and poster presentations of refereed papers. Along with the conference is a professional exposition focusing on machine learning in practice, a series of tutorials, and topical workshops that provide a less formal setting for the exchange of ideas.Source: https://nips.cc/
It includes various sessions like talks, demonstrations, symposia or papers presentations, but also a more hands on opportunity to get involved with:
- Workshops (see the schedule)
- Tutorials (see the schedule)
- Competitions (see the schedule – but bear in mind it’s not exactly up to date 🙂 )
The NeurIPS 2021 AWS DeepRacer Challenge has been organised by AWS and AIcrowd and is already live, read on for a few details and join!
AWS DeepRacer Challenge
This year the challenge is to build a model using project provided by AIcrowd: https://gitlab.aicrowd.com/deepracer/neurips-2021-aws-deepracer-starter-kit/ . It looks like it’s the simulation environment wrapped in a docker container, leaving the training/control setup to the participant. There is a sample where the agent simply takes a random action.
Your part is to prepare an implementation of a DeepRacer Agent class: deepracer_base_agent.py. It has two methods, the first one is register_reset and the other is compute_action, which I believe means you need to tell the car how to react when a new episode starts and how to adjust the behaviour when a new observation is made.
The challenge will require some git skills, you will fork the project into a private repo and submit that as a submission. I believe submission is determined based on a tag placed on the code in the private repository.
There are two rounds of evaluations, the first being in simulation with objects to avoid and second on a physical DeepRacer car. It’s not exactly clear to me on what needs to be submitted as it sounds like one could try skipping the RL part completely in round one and submit some code to get them through the track (if they know how to deal with effectively translating the observations into actions), that said second round is on a DeepRacer car and that only works with a model upload (unless you change the firmware which I’m pretty sure won’t be happening here).
How to join
Head over to the AIcrowd page of NeurIPS 2021 AWS DeepRacer Challenge and click the Participate link.
You will also find more details there. AIcrowd also share a Discord channel for the challenge and a Discourse forum.
There are two types of prizes: AWS Credits and AWS DeepRacer cars.
Top 10 participants in Round 1 will receive $500 worth of AWS Credits each.
Top 3 Winners in Round 2 will receive $1000 worth of AWS Credits and 1 AWS DeepRacer each, and next 4 Participants will receive $500 worth of AWS Credits each.
On top of that, participants with Top solutions will be offered co-authorship for a research paper outlining the solutions, published in NeurIPS 2021.
Hope you will enjoy the event, have fun and good luck!