Edgify CEO Ofri Ben Porat: Shifting AI Training

Edgify specialises in using ‘edge devices’ rather than the Cloud to train deep learning models for a range of industries. It does this by using an edge device – e.g. a self-checkout till or connected car – for the analysis and training of information locally before sharing it across a network of connected machines. This is within the distant family of federated learning, but solves a lot of the pain points that typical federated learning cannot deal with in the real world.

It reduces the risks, costs and time associated with transferring sensitive-data to or from an external server. This allows businesses to train on the entirety of their data, and reach accuracy levels never achieved before. I started the business with my co-founder Nadav Tal-Israel in 2015 and we have offices in the UK and Israel.

Edgify’s technology is already being used in supermarkets. Self-checkout machines are able to distinguish between barcodeless groceries before sharing the acquired knowledge or model across a distributed yet collaborative framework of point of sale machines, leading to faster, more accurate and low-touch checkouts.
 
 
A fundamental shift in AI Training - Edgify
 

How did you come up with the idea for the company?

 
Five years ago my co-founder Nadav had an idea for computer vision-based research that would analyse personal photo galleries and instead of extracting information about the photos themselves, we would extract characteristics about the owner of the gallery. This sounded like a great idea for a personalisation engine and so we set off and began building it.

The idea was great but presented one major challenge – nobody was going to let us access their photo galleries on their mobile devices in order to extract them for analysis, and so we launched a deep tech research into the ability to teach the phones how to analyse their own photo galleries.

Fast-forward a couple of years and we had the ability to run very complex computer vision and deep learning processes on mobile phones using a platform that we built for our own use.

A little over two years ago, our researchers came to us and said that our ability to run on the mobile phones was extremely innovative and can be adapted to run AI training on any hardware. After researching this field of distributed yet collaborative edge training, we realised that the tech platform we built for our product was more interesting and had a much higher value potential than the personalisation engine it was developed for.

And that is how Edgify was born, it’s the technology that was developed for our earlier product, only now it’s wrapped and delivered in a way that anyone can use it for their training needs on the edge.
 

 

What advice would you give to other aspiring entrepreneurs?

 
1. Always focus on real market pain points, don’t make them up in your head. The difference between “nice to have” and “must” from the client’s perspective needs to be the focus in the early days.

 

2. Over-communicate everything, both to your team, but most importantly to your investors. All investors are in your corner when things are going well, but in order to keep them there when things are not going according to plan (which they will eventually do) you must always communicate the situation to your investors.

 

3. Don’t chase a high valuation – that will come with sales, R&D and growth. Going too high on valuation will definitely hurt you if you decide to pivot or regroup. At the end of the day, businesses are built on capital not shares.
 

What can we hope to see from Edgify

 
Two weeks ago we announced our $6.5m funding round with technology backed by Octopus Ventures, Mangrove Capital Partners and a semiconductor giant, which is incredibly exciting.

We want to bring federated learning to the real world. As I mentioned before we are already in supermarkets. For now we will be building out our reach in the retail sector. However the application is agnostic, meaning Edgify’s framework can train AI on a network of edge devices ranging from MRI machines, connected cars, checkout lanes, mobile devices, etc!