Israeli startup DeepCube, which develops technology that optimizes Machine Learning deployment, announced a $7 million Series A funding round. The investment was led by Canadian Awz Ventures, with participation from Koch Disruptive Technologies and Nima Capital.

Reducing the ML model size without compromising on accuracy

“The main issue with Deep-Learning models these days is their enormous memory size, which usually means that day-to-day operations become pretty costly. As a result, end-user installation almost doesn’t exist, and costs are extremely high for Datacenters,” explains Dr. Eli David, DeepCube CEO.

DeepCube has developed a collection of Deep-Learning methods, which are backed by no less than 8 patents that allow automated optimization for each Deep-Learning model, reducing their size by more than 90%, without compromising on accuracy.

In addition, the company developed an inference framework for Deep-Learning that runs on different hardware devices - at Datacenters and on end-user devices - this makes it easy to run models that have been optimized and improve speed and memory usage. According to David, running the models on DeepCube’s superior system makes it possible to improve the model’s speed and memory usage over the current standard non-DeepCube infrastructure.

In a conversation with Geektime, David tells that “while various devices offer small improvements of between 10%-20%, DeepCube’s software-base solution provides significant upgrades to speed and capacity, and it doesn’t matter which hardware device is used. The hardware’s specs or price have no significance, as the user will receive notably improved performance at less cost, all by utilizing DeepCube.

DeepCube was founded in 2018 by Dr. Eli David and Yaron Eitan. The company has 20 employees split between Tel Aviv and NYC based offices. The company notes that the funds will be put towards expansion of its R&D efforts, as well as invested in its product’s go-to-market. To date the company has raised $12 million, with an unreported $5 million Seed round.