Numericcal

Numericcal is a California based artificial intelligence company providing mobile application and IoT developers with the ability to leverage Deep Neural Networks (DNN) on edge devices. The company’s platform allows developers to deploy, optimize and manage DNN models with analytics. The company operates with a B2B business model with license fees to operate and update neural networks.

Numericcal is developing a software platform that cuts down the neural network available on the cloud and makes the code and data small enough to be operational on a smartphone or an IoT device. This neural network transfer is performed over the internet; however once implemented, user data is not sent to the internet. Utilizing deep learning on edge devices like smartphones, TVs, and embedded hardware cuts down on network latency, outages, costs and improves privacy.

The platform architecture contains two sections; an edge SDK on an operating device, and a Cloud SDK on a cloud infrastructure containing the DNN repository. The edge SDK sends device hardware and metadata information to the cloud SDK, where DNN profiling tables can be configured to send a specific DNN model relating to the edge hardware. This configuration allows developers to hold multiple DNN models and data for different devices. The device specific models reduce the amount of code and data an edge device has to compute. The platform uses asynchronous API based on ReactiveX to handle large DNN models. The company initially released its platform for Android devices with RxJava API on Dec 22, 2017.

Numericcal also provides machine learning model optimization tools and support by providing benchmarks on model change, hardware platform specific analytic, and bottleneck identification.

Ranko Sredojevic, Shaoyi Cheng, and Zach Thigpen founded the company on May 1st, 2016. The company is headquartered in California, United States.