Following in the line of the high caliber of speakers at the AI Summit in New York last week, was Intel’s Gayle Sheppard who delivered an exciting insight into how Intel has implemented AI, looking at anything from their AI-legacy, to where they are today.
Sheppard started the presentation by quoting Intel’s Co-Founder, Dr. Robert Noyce: “Until now we have been going the other way, that is in order to understand the brain, we have used the computer as a model for it. Perhaps it is time to reverse this reasoning: to understand where we should go with the computer, we should look to the brain for some clues”.
“Today, Intel is delivering intelligence in the AI-age”, Sheppard said while showing an infographic that listed Intel’s capabilities and their experiences, linking how they correlate with each other.
Sheppard spoke about Intel’s capabilities such as their reasoning systems (Saffron and Intel Xeon), machine/deep learning (Nervana and Intel Xeon), programmable solutions (Arria), depth sensing (Intel Realsense), computer vision (Movidius) and tools and standards (TAP).
“There are two different forms of machine intelligence”, Sheppard said, looking at ‘methods’ and ‘implementations’. Under methods, the subcategories are machine learning and reasoning systems. Under implementations follows deep learning, classical machine learning, memory based and logic based implementations.
Sheppard gave the audience an insight into Saffron’s Natural Intelligence Platform: “A memory and reasoning platform for development and out of the box deployment”.
Sheppard took the audience through Saffron’s whole work-process, from beginning to end, starting with compiling structured and unstructured data sources, also referred to as ‘Ingest’.
Then, Saffron takes the data and analyses it by using source connectors/extractors, which enables the machine to “learn”. Following this, Saffron takes it to the next level by using associative memory representations in order to remember what it has learnt.
Then it moves to the universal query engine for memory-based reasoning, where the Saffron “thinks”, before the final step, which is reaching a conclusion.
Sheppard emphasise how Intel works with delivering value to the enterprise to transform decision making with three steps: speed to insight, transparency and greater accuracy.
Finishing the presentation, Sheppard showed use-cases where Saffron has assisted various companies, starting with USAA.
USAA drives value at customer personalisation, as the company prior to implementing Saffron, had a 20-35% accuracy in knowing its customer’s needs. The results with Saffron shows an increase to 93% accuracy, providing the customers with relevant, individualised product recommendations.
Mount Sinai uses Saffron in healthcare personalisation, as there have been problems with accurately diagnosing two heart conditions with highly similar signatures, where a misdiagnosis may be fatal. Here, the accuracy on average have been 54%, but with Saffron, there have been 96%accuracy in diagnosis after using it on 10.000 attributes.
Boeing has used Saffron to personalise asset maintenance, where issues have been aircraft downtime, time and money lost due to poorly timed maintenance costs. Prior to implementing Saffron there were a 66% accuracy and 16% false alarms, whereas after implementing it, the accuracy has increased to 94% and only 1% false alarms. It has also given faster aircraft turnarounds.
Accenture uses Saffron in software testing and defecting resolution, as 30.000 FTE’s Globally and 1000+ clients scaling for growth it is highly competitive innovation demands. After using Saffron, 10-15% of the productivity have improved, and 50% test script DeDup.