Deep Learning AI

Decentralized Deep Reinforcement Learning

What Is Deep Learning AI?

 

Deep Learning is a type of AI machine learning algorithm. These particular algorithms were actually inspired by the human brain and its artificial neural networks.

Andrew Ng who formally founded Google Brain, which subsequently resulted in many of Google's services incorporating deep learning technologies.

 

Andrew has referred to deep learning as within the same context as traditional artificial neural networks. There was a talk in 2013 titled 'Deep Learning, Self-Taught Learning and Unsupervised Feature Learning' he commented on the idea of deep learning:

"Using brain simulations, we hope to:

– Make learning algorithms much better and easier to use.

– Make revolutionary advances in machine learning and AI.

I believe this is our best shot at progress towards real AI"

Andrew also made comments on the scalable nature of deep learning and how it will outperform and outlast older learning algorithms as time goes on:

"for most flavors of the old generations of learning algorithms … performance will plateau. … deep learning … is the first class of algorithms … that is scalable. … performance just keeps getting better as you feed them more data".

 

 

 

 

 

 

Why Deep Learning?
Slide by Andrew Ng, all rights reserved.

 

We can now see the pioneering nature of deep learning from some of some the biggest names in tech. That's why we've used deep learning algorithms within our cctv video analysis software. Not only do we incorporate the innovation of deep learning, but we go one step further and implement deep reinforcement learning within our vision analysis software.

What Is Deep Reinforcement Learning?

Deep reinforcement learning takes both reinforcement learning and deep learning and combines them. This allows for the ability to solve a wide range of complex problems and decisions that previously would require a human to solve or make. The nature of deep reinforcement learnings' abilities means many domains such as robotics and smart cities are seeing the birth of new applications from deep reinforcement learning. We've decided to utilize these powerful algorithms to enhance the power of CCTV and surveillance monitoring. 

Why We’re Here

   
We believe in augmenting human intelligence, not replacing it. We advocate AI as a tool to improve human productivity, not as a `lights-out’ black-box to replace human judgement. In 2005, Professor Sean Gong anticipated the growth of video data and initiated a multi-year research programme to apply machine learning to transform the capability of video analysis. The goal was to take world leading research and create practical products that were scalable and reliable for the demands of the commercial world. That’s why we founded Vision Semantics. That’s what’s driven our creation of the next generation AI platform.

 

Changing the Rules of the Road


Today, AI is based on big centralized data “learn and forget” algorithm models with limited data privacy, supercomputing processing power, and big power consumption.   This has resulted in AI being applied in narrow applications and these applications do not transfer well across domains.  As a result, AI applications are dependent on extensive data training and retraining when applied in other domains.  Therefore, the platforms are not scalable.

 

To date, big technology AI has been able to capture user data without many legal limitations or user restrictions. This is now changing and going forward user privacy and data ownership will be tightly managed. Centralized training become more difficult with EU ruling on EU-US Privacy Shield. The old machine learning approach dependent on easy access to data is not fit for the future, which is why VSL is providing a radically different AI platform. The edge is the next stage of the evolution of AI technology because of the physical constraints, the cost constraints, and the practical constraints of running all AI applications in the cloud. It simply won’t make sense to send all the data for applications like video and audio streaming to the cloud and back down for every situation, every endpoint.

Fundamentally Different Approach – Decentralised AI Platform


VSL’s algorithms are fundamentally different to competitors’ approaches to machine learning. The company has developed unique patent-protected Deep Learning and Reinforcement Active Learning algorithms for optimising domain transfer Zero Shot learning in computer vision Re-Identification.  The algorithms are not dependent on large data training sets (N Shot learning) and transfer well across domains. This enables it to deliver the next generation AI platform, more advanced than Google Federated Learning, based on small data with data privacy, small processing power and low energy consumption enabling the rollout of a Decentralised Deep Learning AI Platform.


This combined with a ‘Human-in-the-Loop’ Search & Learn results in a continuous improving system without the need for large scale labelled training data sets at all target domains. We call it Dynamic Search & Learn, a lifelong model learning – the more you use it the better it gets.

 

 

 

 


 

Decentralised AI Platform Enables Multiple Sector Opportunities


The advances in silicon have enabled the distributed devices (Edge Devices) to run AI algorithms to have the potential to run locally. VSL’s Distributed AI platform architecture makes it happen. The rollout of 5G connectivity and Internet of Things (IOT) forest of sensors provide both connectivity and data collection & processing at the Edge.  VSL enable the capability of AI to be deployed on a massive scale bringing AI-IOT to life.


Today, VSL provides RE-ID for computer vision in a network of CCTV cameras. Tomorrow, everybody-worn camera or mobile phone can be enabled by the Distributed AI Platform, vastly accelerating the benefits for Public Safety and Smart City Applications at the edge of the network.


The Decentralised AI Platform will transform the outcomes for Retail, Commercial, Medical and Industrial sectors. The lower architecture and energy cost and the flexibility of AI models and domain adaptive edge-localised model training will speed up new applications. The key applications within real-world products such as controlling home devices or providing driver assistance in a car, are running on the edge and require real-time responses. Working through a cloud architecture will no longer make sense. 

 

Decentralised AI Platform Enables the Consumer


Every consumer with a smart phone or internet connected consumer device has the ability to become an AI Edge Device and reap the benefit of AI without loss of data privacy.  Alongside smart phones, camera, consumer devices, there are about 100 Billion edge CPUs which can process a VSL Edge Algorithm, bringing the benefits local to the consumer and without the need to send the data to a remote data centre.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Contact Us For a Free Trial


Get in touch
today for a free trial of our industry leading, deep learning AI solutions. Implement our AI enhancing platform into any video software for CCTV and surveillance.

Leadership diagram.JPG

VSL believes that the future is distributed user ownership of data (individuals or organisations), not centralized data owned by technology companies.

VSL executes machine transfer learning (domain adaptation) on distributed “small data”, Information & knowledge correlation over networks without data sharing.

VSL implements AI for Edge domain adaptation (zero-shot transfer learning), more being used, better it gets for each user so it transfers well across domains.

VSL implements Human-In-The-Loop: AI augments human decision making, not replacement of humans, not “black out” solutions.

VSL implements 

Heterogenous Semantic Correlation of multi-source for joint decision making (Re-ID, geo-loci, text). Users can be comfortable with AI that helps them with their local information without being intrusive.

Vision Slide.JPG