In recent years GPUs have become the most important component for heavy calculations in the worlds of artificial intelligence, but now, the Israeli startup Deci claims they can upgrade them with software, to train and run AI models better.

AI for AI

Deci uses ML to build and optimize neural networks automatically, making them more efficient and suitable for large-scale instant activation. According to the company, its deep learning platform allows ML developers to get performance improvements of up to ten times from the models they built, whether it is computer vision, processing, or understanding natural language or audio. In practice, Deci's platform uses the same models, to redesign neural networks and to maximize their runtime performance - all while maintaining accuracy. This way you can achieve better performance on any hardware, including standard processors, GPU, and FPGA accelerators, as well as dedicated artificial intelligence chips for end devices and cell phones.

The user can enter various parameters to suit the model such as task type, accuracy requirements, calculation speed, target hardware (smartphone, computer, cloud, etc.) and let Deci do the rest; the engine builds the model optimally. According to the company, this should produce a significant improvement in performance results and running and operating costs.

Recently, the company launched the 2.0 version of its development platform, which allows data scientists to develop accurate and fast models tailored to the application requirements and the end hardware on which the models are designed to run. Earlier this year, it released a family of computer vision models that it claims are leading to the best performance in the world in terms of CPU accuracy and computational speed, which reduces the computational gap between CPU and GPU by more than 50%.

Another $25 million to the box office

Yesterday (Wednesday) Deci announced the completion of a Series B round totalling $25 million. The round was again led by Insight Partners, and with participation from Jibe, Emerge, Fort Ross, Square and ICON. This round comes less than a year after the company's previous round of funding, which saw $21 million raised. Deci was co-founded by Yonatan Geifman, Ph.D., an expert in deep learning (CEO), together with tech entrepreneur Jonathan (Joe) Elial (COO), and Prof. Ran El-Yaniv (Chief Scientist), a computational learning researcher and professor at the Henry and Marilyn Taub Faculty of Computer Science at the Technion. The company currently employs about 50 people in its offices in Israel and the United States.