Boltzbit has raised a £1.6 million seed round for its ‘generative AI’ deep learning platform.
The London-based startup says its neural learning platform allows data scientists to create thousands of AI models without writing a single line of code.
The new funding round was led by Speedinvest, with participation from IQ Capital. Boltzbit will use the funds to expand with the opening of a Berlin office, growing its engineering team and research resources.
Generative AI has broader potential than the narrow AI often used in software today. It can automate challenging tasks like synthesising high-resolution images, engaging in natural conversations with customers or even writing complex code at the level of human developers.
While tech giants such as Google, Meta, Amazon and Microsoft are racing to develop practical generative AI models, the technology is too costly and complex to be part of most businesses’ everyday machine learning toolboxes.
“Current generative neural networks are designed for data synthesis that is often highly limited in practice. This approach is also inefficient and counter-productive for training generative AI,” said Boltzbit co-founder and CEO Dr Yichuan Zhang.
“We’ve changed this with a breakthrough in our unique dynamic generative neural networks that learn many generative tasks shared by multiple real-world use cases all together, making our generative AI vastly more efficient and effective.”
The firm says its platform allows data scientists to prototype different neural network architectures within minutes rather than days, without code or manually tuning training parameters, and to benchmark AI solutions on the tasks they really care about.
With Boltzbit’s solution, businesses can leverage a large amount of unlabelled data, including images, natural language data, time‐series data and other unstructured data to solve business challenges including the generation of promising drug and vaccine candidates without slow lab experiments; automating data discovery and integration for carbon accounting; learning about images and text to improve neural search and recommendation solutions; and answering questions about products or engaging in conversations with customers.
Boltzbit was co-founded by Zhang along with Dr Jinli Hu (chief research scientist) and Dr Hanchen Xiong (CTO), a team with previous experience at companies like Google, Twitter and Microsoft.
Zhang has been working on fast and scalable inference algorithms for unsupervised AI for more than 10 years. His research at the University of Cambridge into deep generative AI applications for automated data cleaning and outlier detection of unstructured data is the foundation of Boltzbit’s platform.
Xiong is an award-winning machine learning researcher and engineer who was previously a senior machine learning engineer and data scientist at Twitter, while Hu is an expert in probabilistic machine learning, who has had papers published at high-profile AI conferences. Prior to co-founding Boltzbit, he led development of Gambit’s first deep learning based betting systems.
The startup has collaborated with researchers from the University of Cambridge to work on a new generative AI training algorithm for more complex neural network architectures. This algorithm can significantly boost the performance of generative AI on multiple unsupervised tasks simultaneously.
“We’re seeing a shift around the world from task-centric AI to data-centric AI that supports multi-tasking and innovative capability,” said Zhang. “Boltzbit aims to be the leader of this exciting trend to create the next generation of AI to help all businesses to solve challenging problems using the power of data.”