Hoursec

Hoursec Self-Learning Machines for Inference and Training On-A-Chip solves the problem of Energy Efficiency for the Edge Computing Market.

We already have good willingness to pay feedback from potential customers. We are in the process of co-developing our solution together with early adopters in a B2B pilot project case. However, we are working in a plug-and-play solution that will accelerate sales. We have got traction for collaborations with International Research Institutions like CERN and LIGO, together with financial institutions, Helbling group and Quantum computing startups. Hoursec has been selected as top five startup of the year at Edge Computing World 2021 and top three Swiss Technology Awards.

Machine Learning (ML) models can only be trained in the cloud given their expensive computational and lengthy time requirements in combination with large data sets. Massive processor muscle is applied to deal with both complexity and the required time. It is no surprise that all these training activities, from voice to image to biometric data to autonomous driving is projecting an almost exponential growth in power consumption by datacenters. Clearly, in the light of today’s climate and environmental footprint, this is not acceptable. What is also becoming unacceptable are the risks associated

Hoursec innovative learning model (HW/SW) architecture is based on a proprietary paradigm shift for continuous training and inference. Our architecture combines matrix vector multiplication and content addressable memory into a hyper dimensional computational kernel which dramatically reduces the time for training and inference. The ability to not only run but also train ML models “on the job” results in four fundamental shifts in the application of ML: (1) privacy: personal data will remain local, (2) power consumption will be drastically reduced for both training and operation, (3) deploymen

Our Training and Inference on Chip has a Software and Hardware co-development component that we are going to License in exchange of royalties per device or royalty and a yearly subscription for recurrent. This business model is a better proposition than the current approach where customers need to re-train expensive ML models on a monthly basis besides having to pay for expensive talent to parameterise their architectures.

Founder Institute Switzerland