Overview
- Quadrant provides machine learning services to developers for building state-of-the-art deep learning models
- What’s unique about this framework is that it can produce high accuracy results on deep neural networks, without requiring much data
- It uses semi-supervised algorithms and works with noisy and unlabelled data as well!
- Quadrant’s generative machine learning algorithm won first prize at the ‘CATARACTS medical imaging grand challenge’
Introduction
Imagine building a deep learning model, from scratch, that requires minimum data points. It’s almost unthinkable. So far, whenever we think of deep learning, we know we’ll require millions of data points to build a good enough model. I’m sure most of you who are familiar with DL must have done the cats and dogs classification problem. That model needed tons and tons of images to be trained properly.
D-Wave, one of the world’s leading companies in quantum computing software and systems, this week launched a business unit called Quadrant that provides machine learning services to developers for building state-of-the-art deep learning models. What’s unique about this framework is that it can produce high accuracy results on deep neural networks, without requiring much data. This is done by building generative models which combine the flexibility of deep neural nets with probabilistic graphical models.
If you do have this data, then that’s not an issue. But the real practical challenge is getting this amount of data AND labelling it as well, so the algorithm understands what it has to classify (also called supervised learning). These tasks are time consuming and expensive and usually out of scope for most organizations. For instance, in healthcare to detect a rare disease, you will need a lot of data to generate a high accuracy model. But you won’t find many images of this disease. That’s a real problem.
Quadrant aims to solve this conundrum. Below is the comparison they have shown of their algorithms against the others currently in use, or now defunct. The difference is stark.
The team at D-Wave has developed semi-supervised algorithms that make use of images with noisy labels. These include images found on social media, search engines, etc. and consequently are less expensive to obtain. The algorithms also make use of unlabelled data to train the model and produce amazingly accurate results.
Quadrant solutions run on standard GPU-based systems using D-wave’s quantum technology. As mentioned by the team, D-Wave plans to integrate the Quadrant solutions in hybrid quantum/classical platforms for use with its next generation quantum system, which is currently in development with prototypes being tested.
Our take on this
This heralds a breathtakingly refreshing new breakthrough in deep learning. Imagine the uses of such a framework – medical imaging, finance, sports, telecommunications, marketing, advertising, among others. It has the potential to integrate smaller sized organization into the deep learning scene as well.
D-Wave has already announced it has partnered with SIemens Healthineers to win first place at the CATARACTS medical imaging grand challenge. The team used Quadrant’s machine learning algorithms to identify surgical instruments in the given videos, with high accuracy. The possibilities are endless.
Subscribe to AVBytes here to get regular data science, machine learning and AI updates in your inbox!
Senior Editor at Analytics Vidhya.Data visualization practitioner who loves reading and delving deeper into the data science and machine learning arts. Always looking for new ways to improve processes using ML and AI.
