Artificial intelligence startup Galileo Technologies Inc. today launched from stealth mode after raising a $5.1 million seed funding round led by The Factory.

The Factory was joined by several angel investors, including Anthony Goldbloom, chief executive of Google LLC’s Kaggle unit. Kaggle operates a popular machine learning website that AI experts use to participate in research initiatives.

One of the most important steps involved in developing a neural network is the so-called AI training process. As part of this process, developers hone their neural network’s accuracy and speed by having it ingest a large amount of training data. Before a software team can begin the task, it has to ensure that the training data being used for an AI project doesn’t contain errors.

Galileo has developed a software platform that promises to reduce the amount of effort required to find and fix errors in AI training datasets. According to the startup, its platform is capable of speeding up the process by a factor of 10.

Currently, the task of finding errors in AI training datasets is largely done manually using spreadsheets and custom scripts. Performing the task manually can require a significant amount of time, particularly when the dataset being scanned for issues contains a large amount of information.

Developers have to ensure that the data being used to train an AI is similar to the information the AI will process once it’s in production. For example, if a company is building a neural network to organize business documents, developers might train the neural network using a collection of documents from one of the company’s systems of record.

Software teams also have to scan their datasets for more fine-grained errors. Often, the individual records in an AI training dataset are stored alongside a piece of metadata that helps the AI being trained interpret the information. Developers must ensure that both the records and the associated metadata are accurate.

The risk of data errors going unnoticed is another challenge, according to Galileo. It can be difficult for developers to detect every single accuracy issue when scanning an AI training dataset manually. In some cases, unresolved accuracy issues can reduce the reliability of the AI that the dataset is being used to train. 

Galileo says that its platform saves time for developers by automating many of the task’s most time-consuming aspects. According to the startup, developers can deploy its platform by adding a few lines of code to their AI projects. From there, Galileo’s algorithms automatically detect issues in AI training datasets and suggest ways to fix them.

Developers find ways to improve their neural networks through trial and error. A software team might, for example, create three versions of the same AI training dataset and train a neural network on all three versions to determine which produces the best results. As part of its platform, Galileo provides features that help software teams manage this testing process more efficiently.

“We are building Galileo with the goal of being the intelligent data bench for data scientists to systematically and quickly inspect, fix and track their ML data in one place,” stated Galileo co-founder and Chief Executive Officer Vikram Chatterji.

Chatterji launched Galileo after previously working as a project management leader at Google AI. Galileo co-founders Atindriyo Sanyal and Yash Sheth earlier held senior engineering roles at Google and Uber Technologies Inc.’s AI group, respectively.

“Finding and fixing data errors is one of the biggest impediments for effective ML across the enterprise,” said The Factory investor Andy Jacques. “The founders of Galileo felt this pain themselves while leading ML products at Apple, Google and Uber. Galileo has built an incredible team, made product innovations across the stack and created a first of its kind ML data intelligence platform.”

Galileo says its platform is currently being tested by multiple Fortune 500 companies and startups. Using the new funding, the startup plans to grow its headcount and accelerate engineering initiatives. 

Image: Galileo