Aquarium Raises $2.6M Seed Funding for Machine Learning Data Refinement

Aquarium Secures Seed Funding to Accelerate Machine Learning Model Deployment

A new startup, Aquarium, founded by former Cruise personnel, is focused on streamlining the process of refining machine learning model data and accelerating its transition to production environments. The company recently announced the successful completion of a $2.6 million seed funding round.

Seed Round Details

This seed round was led by Sequoia, with additional participation from Y Combinator and a group of angel investors. Notably, Cruise co-founders Kyle Vogt and Dan Kan were among those contributing to the funding.

Addressing a Key Bottleneck in Machine Learning

CEO Peter Gao and head of engineering Quinn Johnson, the co-founders of Aquarium, identified a significant challenge during their time at Cruise. They observed that identifying weaknesses within model data frequently hindered the deployment of models into real-world applications.

Aquarium is specifically designed to tackle this problem. The platform aims to provide a solution for improving the quality and relevance of data used to train machine learning models.

Improving Model Performance Through Data Management

“Aquarium functions as a machine learning data management system,” Gao explained. “It empowers users to enhance model performance by focusing on the data used for training – often the most critical factor in achieving successful production deployment.”

The founders have noted a surge in the development of machine learning models across diverse industries. However, many teams encounter difficulties due to the complexities of data iteration and the ongoing need to identify and acquire pertinent data.

The Iterative Cycle of Model Improvement

“A substantial portion of model improvement, and the effort required for production readiness, centers around determining what data needs to be collected next, what requires labeling, and what necessitates retraining and error analysis,” Gao clarified. “This iterative process is central to our approach.”

The ultimate goal is to develop models that surpass human performance. Sterblue, a customer utilizing Aquarium’s services, provides a compelling example of this potential.

Sterblue Case Study: Enhanced Accuracy and Cost Reduction

Sterblue offers drone inspection services for wind turbines. Previously, inspections relied on human technicians. However, by leveraging drone-captured data and training a machine learning model, they were able to automate the process.

Through the use of Aquarium, Sterblue refined its model, achieving a 13% improvement in accuracy while simultaneously reducing the cost of human reviews by 50%, according to Gao.

Commitment to Diversity and Bias Mitigation

Aquarium currently employs seven individuals, including the founders, with three being women. Gao emphasizes that diversity is a deliberate priority for the company.

Recognizing the potential for bias in machine learning model creation, Aquarium believes that assembling a diverse team is a crucial step in mitigating these inherent biases.

Company History and Product Availability

The company was launched in February of last year and participated in the Y Combinator Summer 2020 program. Throughout 2020, the team focused on product refinement.

Aquarium recently transitioned from a beta program to general availability, making its platform accessible to a wider range of users.