Image for post
Image for post

The Current Times

In the current status of the world, the grocery retail industry has been thriving, with the average increase of sales being 26.9% across the board. Although this is great news for the grocery retail industry, there is a popular saying that goes, “more money, more problems”. So as sales may increase for businesses across this sector, so will their shrink rate, extended queues at tills, staffing needs, etc.

In the grocery retail industry, shrinkage, or shrink, is the term used to describe a reduction in inventory due to shoplifting, employee theft, administrative/human errors such as unintentional/intentional record keeping, pricing, etc. A common misperception is that retailers simply absorb shrinkage as part of the cost of doing business. …


En route to eradicating the need for cloud in any AI training and retraining processes. Part 2 of 3.

Image for post
Image for post
Live view of our cluster of edge devices, running a live distributed and collaborative training.

If you haven’t read the first post in this series, we suggest that you read it in order to get the most benefit from this article. It described our process for building a cluster of 100 edge devices, in order to train deep-learning and machine-learning models in a distributed manner using a cluster of edge devices, without requiring the use of the cloud while achieving near-perfect accuracy. …


Fresh Produce Recognition at Point of Sale

Image for post
Image for post

Trend Spotting

Coming back from NRF 2020 we’ve spotted the next big trend in grocery retail:

Image Recognition of Barcode-less Items

Self checkouts will soon be using computer vision technology to recognise barcode-less items such as fruits and vegetables. No more frustrated customers browsing through endless screens to locate the Banana’s SKU or delaying the queue trying to distinguish between Pink Lady and Red Delicious apples.

At first, we were cautious about calling this a trend, but one after the other, we noticed major industry leaders investing in initiatives aimed at this vision. NCR, Toshiba, Zebra, Shekel, DataLogic, Fujitsu, HP, NEC, StrongPoint and Diebold Nixdorf to name a few. …


Edgify’s Collaborated Method for Distributed Learning, to be Fully Released at NeurIPS this Year!

Image for post
Image for post

Overview

There is a growing interest today in training deep learning models on the edge. Algorithms such as Federated Averaging [1] (FedAvg) allow training on devices with high network latency by performing many local gradient steps before communicating their weights. However, the very nature of this setting is such that there is no control over the way the data is distributed on the devices.

consider, for instance, a smart checkout scale at a supermarket that has a camera mounted on it and some processing power. You want each scale to collect images of fruits and vegetables being scaled and to collectively train a Neural Network on the scales to recognize these fruits and vegetables. Such an unconstrained environment would almost always mean that not all edge devices (in this case, scales) will have data from all the classes (in this case, fruits and vegetables). …


Check Out the Check-in Process.

Image for post
Image for post

“You don’t build a business — you build people —and then people build the business”. — Zig Ziglar

A little over a year ago, the time came to put our money where our mouth is when we said that we are a “people-first,” startup company. This forced us to start asking ourselves and our employees more difficult questions on how we’re doing at attracting and retaining the best people, and how we can improve in doing so as a company.

On one hand, we decided it was time for our company to embrace some type of (formal) review and feedback system. On the other hand, I wanted to adopt a system that would be simple and easy to implement, preferably automated, and one that would exert use and value throughout the year. It was important that this would be built to be dynamic and updated as we go, and not only as a once a year, painful and bureaucratic HR requirement. The timing we chose could be considered as “too early” given the stage of the company, the size of the team and the shutter of horror I got from the senior managers after first raising the idea of a Check-in process. …


En route to replacing the cloud for all AI training. A three part series on setting up your own cluster of edge devices (1/3).

Image for post
Image for post

Intro

This is the first post, of a three-part series, describing how we built a cluster of 100 edge devices, in order to train deep learning and machine learning models without ever using the cloud, whilst achieving close to perfect accuracies.

These 100 edge devices, are intended to replicate real world edge devices, such as self checkout POS, cameras, connected cars, medical devices, etc..

This first post focuses on the various aspects that we considered while building our edge-device clusters, including hardware, network limitations, power supply requirements, the actual construction of the cluster and more. The next post of this series will address cluster deployment and ongoing management; and finally, post 3 will present the challenges and our overcoming, as we encountered them, while training a deep-learning model on top of various types of hardware (edge). …


(3/3) an Edgify Research Team Publication

Image for post
Image for post

In the first post of this series, we presented two basic approaches to distributed training on edge devices. In the second, we explored compression methods for those approaches. Diving deeper into the real-world challenges of these training methodologies, we now introduce a problem with common architectures that arises from certain common data distributions — and the solution we have found for it.

Non IID Data Distribution

Training at the edge, where the data is generated, means that no data has to be uploaded to the cloud, thereby maintaining privacy. This also allows for continuous, ongoing training.

That being said, keeping things distributed, by its very nature, means that not all edge devices see or are privy to the same data sets. Take Amazon Go, for example. Their ceiling cameras capture different parts of the store, and hence different types of products. This means that each camera will have to train on a different dataset (and will therefore produce models at varying levels of quality). …


(2/3) An Edgify.ai Research Team Publication

Image for post
Image for post

In the first post of this series we presented two basic approaches to distributed training on edge devices (if you havent read our first post in this series, you should start here). These approaches provide benefits such as AI data privacy and the utilization of the compute power of edge devices. Large-scale distributed training systems, however, may fail to fully utilize that power (as we’ve come to see in our own experiments), as they consume large amounts of communication resources. As a result, limited bandwidth can impose a major bottleneck on the training process.

The two issues described below, pose additional challenges to the general limitation of bandwidth. The first has to do with how synchronized distributed training algorithms work: the slowest edge device (in terms of network bandwidth) “sets the tone” for the training speed, and such devices need to be attended to, at least heuristically. Secondly, the internet connection on edge devices is usually asymmetric, typically twice as slow for upload then for download (table 1). This can severely impede the speed of training when huge models/gradients are sent via the edge devices for synchronization. …


How many times have we heard sentences like: “this is how we have always done it” or “everyone else does it this way” ?

Image for post
Image for post

Working in teams has a lot of advantages. We share tasks, ideas, goals and most importantly a sense of belonging. To sustain and maintain a working group structure, the team must form a set of unwritten rules for each member to follow. Any new member joining the group must follow these rules or the group might treat the new member as an outsider. In other words, to be an efficient member of the group — that person has to conform to the group’s rules.

Conformism is a very efficient tool in helping the group reach its common goals. However, with all its advantages, conformism entails one major flaw. These unwritten rules, make it hard for the group members to challenge fundamental concepts and behaviours of their team. …


(1/3) An Edgify Research Team Publication

Image for post
Image for post

This is the first, introductory post, in our three part algorithmic series on real-world distributed training on the edge. In it, we present and compare the two fundamental methods for this type of machine learning (and Deep Learning in particular). The post following this one will then address communication compression, and the final one will attend to the challenge that non-IID distributions pose, specifically for Batch Normalization.

Distributed Edge Training

With the increased penetration and proliferation of IoT devices, and the continuous increase in connected, everyday devices (from smartphones to cars to self check out stands), the amount of data collected from the world is increasing exponentially. …

About

Edgify.ai

A foundational shift in the world of AI training. Deep Learning and Machine Learning training directly on edge devices.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store