Intro to Aquarium
Aquarium is an ML data operations platform that helps teams find issues, validate fixes, and add the right data to improve their machine learning datasets.
ML models are defined by a combination of code and the data. While there's a lot of great tools for debugging and understanding code, there's not a lot of tooling for debugging and understanding the actual data. Our interactive views and collaborative platform allows your teams to work more efficiently on data-centric workflows with the goal of:
- Speeding up your ML workflows
- Saving engineering time
- Reducing operational risk
You should use Aquarium when you're trying to build or improve an ML model.
Most gains to model performance come from improving datasets rather than model code. And as a result, it's hard to make significant gains in the code without large time investments.
Thats where we come in! Aquarium can improve on your team's ML tasks like:
Assessing Data Quality
- Find labeling errors and subsets of your data that have interesting/problematic patterns
- Easy to use interface encourages other members in the process to get involved which leads to freeing up your ML engineers time
Comparing Model Performance
- Diagnose the causes of critical model errors across multiple versions of your model
- Compare multiple models with regression tests to ensure each iteration of your model is truly an improvement
Collecting Relevant Data
- Quickly identify the highest value data to collect that improves the model performance
- Reduce manual time spent trawling through unlabeled data to figure out what to label next
Teams have seen up to a 25% increase in model performance in a single cycle of dataset iteration with up to 8x less time spent than in their previous workflow!
Aquarium supports the following ML tasks:
- 2D Object Detection
- 3D Object Detection
- Semantic Segmentation
We provide a variety of quickstart guides to help you get familiar with our data upload process, and we also provide documentation around how to complete certain ML tasks and workflows.
In addition we have detailed documentation on how to upload different kinds of data listed here: