Unlocking the secrets of the Rubik’s Cube with Reinforcement Learning at UmojaHack Africa 2023

Published

Categories

InstaDeep is pleased to continue its support for up and coming AI talent in Africa by once again sponsoring the UmojaHack Africa machine learning hackathon. This year – 2023 –  is a pleasing multiplicity of 3s: 3000+ students, hailing from more than 300 universities and 30 countries across the continent, will compete in real-world machine learning challenges. There is also a prize pool valued in excess of over $10,000 for the top-ranked individuals and teams. UmojaHack will be streamed live via dedicated Zoom links, with all submissions and results hosted on the Zindi data science competition platform.

Nicolas Lopez Carranza, InstaDeep’s Head of BioAI, remarked on the event, “I am delighted that we have been able to support UmojaHack Africa again.  We think the Rubik’s Cube challenge will add a dimension of fun and creativity, while also helping the participants refine and improve their RL abilities in a collaborative but competitive environment. Having an expert in the field and cube enthusiast like Tristan to set it up made it a no brainer decision to go with this challenge for the advanced track. We are looking forward to checking the results and announcing the winner!”

The Challenge

As Gold sponsor, InstaDeep is once again providing the advanced-level challenge for the event, this time bringing an intriguing problem based on the almost 50-year old Rubik’s Cube puzzle. Deceptively simple-looking, the six coloured faces of the Rubik’s Cube are said to have more than 3,000,000,000 combinations but only one correct solution, making it an ideal candidate for the reinforcement learning techniques that participants will use.  The challenge is to solve the puzzle in the fastest time possible.

Participants will be presented with  a large number of randomly scrambled virtual cubes to attempt to solve, at various levels of difficulty. The solution file will be a list of environment states and actions taken to transition between them, hopefully leading to a solved cube. Supporting scripts will be provided to generate the file in the format required for the marking platform, and scores will be the proportion of correctly solved cubes. Because this is a Reinforcement Learning (RL) challenge, participants will generate their own training data using the environment provided.  Participants will also have access to Colab starter notebooks and github repos to help with their attempts.

Nicolas is hosting the challenge along with the problem’s architect (and expert Rubik’s Cube collector and puzzler) Tristan Kalloniatis, Research Engineer at InstaDeep.  The pair will be on hand throughout to offer advice and answer questions, and also prepared this briefing video:

For more information about UmojaHack Africa 2023 (18-19 March), and to sign up for the event, please visit https://umojahack.africa/

The Prizes

The winners of the Rubik’s Cube Advanced Challenge will be awarded cash prizes, (1st place – $600; 2nd place – $500; 3rd place – $400); plus a year-long DataCamp License worth $400 USD, providing access to over 360 courses, certifications, tracks, competitions, and ready-to-use templates and datasets. Alongside individual challenge prizes, there will also be an additional $100 country prize, given to the highest ranking user/team from a specific country that didn’t win any other prize, across all challenges.

About Zindi

Zindi (zindi.africa) is the largest African data science competition platform with the mission of building the data science ecosystem in Africa. Zindi hosts a community of over 60 000 data scientists, engineers, analysts and practitioners across Africa dedicated to solving the world’s most pressing challenges through machine learning and AI. Data practitioners come to Zindi to hone their skills, build their professional profile, connect with job opportunities, and to share ideas with their peers across Africa and beyond. Private companies, start-ups, governments, and NGOs work with Zindi to source the best data science and AI solutions, as well as proven talent for their specific needs.