Aquarium Learning

Aquarium is a comprehensive machine learning (ML) data operations platform that enhances model performance through embedding generation, processing, and querying. It provides ML teams with tools for targeted data collection, error analysis, and model improvement.

Key features:

- Embedding Generation and Processing: Simplifies creation and management of neural network embeddings without complex infrastructure.
- Indexing and Querying: Efficiently indexes and queries embeddings to surface critical model performance issues.
- Automatic Error Surfacing: Identifies the most critical patterns of model failures, helping prioritize issues.
- Visual User Interface: Provides a visual interface for inspecting and collaborating on model errors.
- Targeted Data Collection: Enables sifting through large pools of labeled and unlabeled data to find specific objects or scenarios.
- Few-Shot Learning: Uses technology to bootstrap new classes with minimal examples, accelerating new model development.
- Scalability: Handles datasets with hundreds of millions of data points, suitable for large-scale ML projects.
- Data Security: SOC2 Type 2 certified with Anonymous Mode for protecting sensitive data.
- Hands-on Support: Offers solutions engineering resources, customer success syncs, and user training.

How it works:

1. Users interact with Aquarium through its web application.
2. The visual user interface allows inspection of model errors.
3. Users collaborate with team members to determine solutions for identified issues.
4. The platform's tools are used to find and collect specific data points to improve model performance.
5. Users can upload datasets and use Aquarium to automatically identify edge cases where the model performs poorly.
6. Additional data for edge cases can be collected and used to retrain the model.

Integrations:

Classification, 2D Object Detection, 3D Object Detection, Segmentation

Use of AI:

Aquarium leverages generative AI techniques, particularly in its few-shot learning capabilities, to bootstrap new classes with minimal examples. This feature is built on advanced neural network embeddings.

Target users:

- ML Teams seeking to improve model performance through targeted data collection and error analysis
- Enterprises requiring scalable and secure solutions for managing large datasets
- AI Researchers interested in leveraging advanced embedding technologies and few-shot learning

How to access:

Aquarium is available as a web application, making it accessible to a wide range of users. It is not open source, ensuring a high level of security and support for enterprise customers.