Aquarium

Improve ML models by improving datasets they’re trained on

Machine learning models are only as good as the datasets they’re trained on, yet it’s extremely difficult to improve dataset quality. Aquarium uses deep learning to find problems in your model performance and edit your dataset to fix these problems.
Discussion
Would you recommend this product?
2 Reviews5.0/5
Hi all, @quinn_johnson and I are the co-founders of Aquarium! We previously led the perception and ML engineering teams at Cruise, where we built the deep learning systems for a self driving car. Our goal is to give ML teams the same world class tooling that companies like Waymo, Tesla, and Cruise already have. We want to scale the improvement in your ML models proportional to the amount of data you have instead of the amount of engineering time you put in. Our tools find common problems in datasets like data anomalies, label quality issues, and patterns of model failures. We then integrate into your labeling provider to help you edit your dataset or sample the best data to improve your model performance the next time you retrain. Some of our customers have already improved their model performance 20% using our tools! If you’re interested in learning more or trying our tools on your own dataset, email me at pgao@aquarium-learn.com and we can get you set up! We’ll also be here answering questions.
As a founder of a company that does ML, I encountered problems with improving the performance of models first-hand. I love how Aquarium handles that for you, so that you can focus on your core business, instead of spending time implementing your own infrastructure. I highly recommend this product to everyone. Congrats @quinn_johnson and @pgao with your launch and thank you for solving this problem!
Awesome work and congrats on the launch @pgao @quinn_johnson !
Congrats on the launch Peter! I can definitely see Tribe using this for deep learning projects. Excited to explore this further