‘Experiment Tracker’ Funded for Deep Learning
New software tools aimed at the builders of deep learning platforms and used to analyze and accelerate model training continue to advance as investors zero in on startups coming up with unique ways to track and replicate experiments while visualizing the performance of those models.
Among the tool vendors is Weights & Biases, a San Francisco-based startup that this week announced a $15 million funding round led by early stage investor Coatue Management. Weights & Biases has so far raised $20 million.
The startup released its first tool last year, an “experiment tracker” used to monitor nearly 1 million deep learning projects thus far. Users are estimated to be in the thousands, the company said Thursday (May 30). Developers say the tool can be used to scale experiments from a single researcher to a team, and from a single model to thousands.
CEO Lukas Biewald noted in a blog post announcing the funding round that the startup was in the inviable position of not needing the funding, allowing the startup to “pick a partner completely aligned with our mission.”
Biewald previously co-founded Crowdflower, later rebranded as Figure Eight, that collects training data for machine learning models. The goal is to accelerate the annotating of training data. Figure Eight was acquired in March by Australian machine learning specialist Appen Ltd. for $175 million.
Figure Eight released a report earlier this month that found AI and machine learning practitioners still spend an inordinate amount of time on data management tasks like cleansing and labeling.
Biewald said the new funding would be used to expand Weights & Biases’ engineering and product development teams as it launches marketing and sales activities.
Coatue, lead investor for the Series B round, also has contributed venture funding for Databricks and Domino Data Labs.
The funding reflects growing requirements among developers to manage and track machine learning and deep learning experiments. While data management approaches are emerging, developers say they still need tools like those offered by Weights & Biases to help streamline deep learning experiments.