Two former Google Deepmind researchers who worked on the company's Nobel Prize-winning Alphafold Protein Structural Prediction AI and its Alphaevolve Code Generation system have launched a new company with a mission to democratize access to advanced algorithms.
The company, called Hiverge, came out from stealth today with a $5 million seed fund led by Flying Fish Ventures, with participation from Ahren Innovation Capital and Alpha Intelligence Capital. The legendary coder and Google's chief scientist Jeff Dean is also a startup investor.
The company uses AI to build a platform called “Hive” to generate and test new algorithms to run critical business processes, from product recommendations to delivery routing. Large companies that can afford to employ their own data science and machine learning teams may develop bespoke algorithms, but this functionality is out of reach of most small businesses. Small and medium-sized businesses often had to rely on off-the-shelf software that came with pre-built algorithms that could not be ideal for that particular business and its data.
The Hive system also promises the possibility that human data scientists will discover anomalous algorithms that can produce excellent results that may not be able to develop through intuition or trial and error, Alhussein Fawzi, co-founder and CEO of the company. luck. “The idea behind Hiverge is to enhance these companies with the best best in class algorithms,” he said.
“You can apply [the Hive] Machine Learning Algorithms allows you to apply that to planning algorithms,” explained Fawzi.
At Google Deepmind, Fawzi led the team that developed the Alphatensor AI in 2022. This has discovered a new way to do matrix proliferation, a basic mathematical process for training and running neural networks and many other computer applications. The following year, Fawzi and the team developed Funsearch. Funsearch used large-scale language models to generate new coding approaches and used automated evaluators to eliminate erroneous solutions.
He also worked on the early stages of what became Google Deepmind's Alphaevolve system. This involves working together several LLMs as agents to create an entirely new codebase to solve complex problems. Google believes that Alphaevolve has found a way to optimize LLMS. For example, I found a way to improve the way Gemini does matrix growth and provides a 23% speedup. We also optimized another important step in the type of AI architecture underlying LLMS, how to improve the work and speed.
Hiberge co-founded with him was his brother Hamza Fauzi, a professor of applied mathematics at Cambridge University, and he serves as the company's technical advisor. and Bernardino Romera-Paredes, a member of the Google Deepmind team that created Alphafold and is now Chief Technology Officer at Hiverge.
Hiverge has already demonstrated the usefulness of the Hive system to beat the Airbus Beluga Challenge. This asks competitors to find the most optimal loading and storage methods for aircraft components carried by Airbus Beluga XL aircraft. The solution developed by Hiverge provided 10,000 times faster speeds over existing aircraft-mounted algorithms. The company also showed that it could adopt a machine learning training algorithm that has already been optimized and speeded up three more times. We also found new ways to improve computer vision algorithms.
Alhussein Fawzi said that although Hiverge, based in Cambridge, UK, currently has six employees, will use the money raised in the latest funding round to expand its team. “We're also moving from research to building products,” he said.
The company plans to make technology accessible through cloud marketplaces such as AWS and Google Cloud. Here, customers can use the system directly with their own code. The platform analyzes which code represents a bottleneck, generates improved algorithms, and provides engineers with recommendations.
