Tampa-based, Lumina Analytics announced today that it has developed a new form of machine learning called Random Contrast Learning or RCL. In early testing, RCL has outperformed deep learning algorithms in Natural Language Processing across multiple measures. These include recall, training speed, inference speed, and size of resulting model. Though RCL is applicable to a wide range of machine learning use cases, Lumina has limited its development to date to Natural Language Processing. Lumina is scaling its language model incrementally to build a large language model to be known as M5. Lumina’s expects M5 to compete well with industry-leading, large language models. The RCL model is projected to operate at lower cost and with better overall performance.
“RCL appears to represent an important breakthrough in the science of machine learning,” said Lumina’s Chairman and CEO, Allan Martin. “It is important for several reasons. It reduces the cost of training large language models near two orders of magnitude. Its accuracy is superior to deep learning models. Its inference speed is an order of magnitude faster, and its resulting models are two to three orders of magnitude smaller. All this makes machine learning better, cheaper, and more accessible.”
Lumina’s Chief Data Scientist, Dr. Morten Middelfart, added, “RCL runs on CPU hardware, which is more readily available and has advantages over GPU, including cost. In addition to the attributes Allan noted, RCL scales linearly without regard to size. This is in stark contrast to deep learning, where entropy massively increases as models become larger. I am excited to share our work with the world. This is the most exciting development in machine learning I have seen or been involved with in my thirty years in the field. I am very proud of our team and the work we have done to achieve this breakthrough.”
For more such updates and perspectives around Digital Innovation, IoT, Data Infrastructure, AI & Cybersecurity, go to AI-Techpark.com.
