ML Contests: June 2020 Update
Hi there, I hope this newsletter finds you well! I feel like I went for a quick nap when lockdown started here in the UK in March, and woke up in June. As always, lots of exciting new competitions to get stuck into.
For those of you using PyTorch, you might be interested in an article I wrote recently on an easy way to speed up tabular data training in PyTorch by more than 20x.
AI for Healthcare
From neuroimaging to prostate cancer grade assessment, melanoma classification, mental health statement classification, and stalled blood vessel analysis for Alzheimer's research, there are some really interesting competitions happening in the healthcare/medicine space across Kaggle, DrivenData, and Zindi.
I mentioned in the February update that we're expecting more reinforcement learning competitions this year, and that seems to be happening now. CodaLab has a competition on learning to run a power network, BienData has one on dispatching and repositioning vehicles for ride-hailing platforms, and Kaggle is hosting the fourth iteration of Two Sigma's Halite competition.
This last one is particularly interesting. It's sponsored by the systematic investment firm Two Sigma, and involves moving ships around a game world to collect limited resources. It hasn't fully launched yet, so there's plenty of time to get up to speed on it before it starts. It's likely to be a big one - over 15,000 people have taken part in previous iterations of this competition!
NeurIPS, widely regarded as the top global machine learning conference, always has a series of interesting competitions. The first of this year's competitions have been announced, and are based around understanding 3D models of clothing. In addition to the $4.5k prize money, the kudos and potential trip to the conference makes this one worth looking at! There are three tracks to this competition, all hosted individually on CodaLab, with the main competition description on ChaLearn.
Dozens of other competitions
One last competition worth highlighting is Facebook's multi-modal hateful meme hateful meme identification competition, with a $100k prize pool. This requires classifying memes as hateful based on a combination of image processing and NLP. There are many more new competitions though, so check out mlcontests.com for the full list.
As always, please hit reply if you have any feedback, and do let your friends know about this mailing list!
PS: If you're looking for a cloud GPU provider for your next project, check out Genesis Cloud. They're much cheaper than AWS and many others, and run their data centre on 100% renewable energy in Iceland. Use this link for $100 free credits to get you started!