AI Spectrum

In the world of AI, a key concept is how to train a neural network to perform a particular task efficiently and accurately, then a hardware solution is created that uses the results from that training - and this is called inferencing.The difference between these two concepts - training and inferencing - often creates confusion among people, and that's why, in today's episode, we are diving deep into explaining these two terms and how exactly they differ. We are also painting a clear picture of the industries that use artificial intelligence and machine learning and what they're working on, so tune in, and find out more!  In this episode, you will learn:The difference between training versus inferencing a neural network. (00:46)Examples of frameworks that help with the training process of a neural network. (01:24)The stage AI & ML is at, currently, in terms of safety-critical applications. (04:42)The industries that are currently using AI & ML, and the types of applications they’re focusing on. (06:52)Connect with Mike Fingeroff:LinkedInConnect with Ellie Burns:LinkedInResources:Catapult High-Level SynthesisSiemens EDA

The field of artificial intelligence and machine learning - just like any other industry where innovation happens -  faces lots of challenges, and specialists are relentlessly looking for ways to overcome them.In this episode, Mike and Ellie tackle some of these challenges and discuss the different compute platforms, their limitations, and the surge of new platform development, as well as the many challenges that hardware designers face as they try to move AI to IoT edge devices.Tune in, and learn some of the challenges of implementing the latest cutting-edge neural network algorithms on today's compute platforms. In this episode, you will learn:The amount of energy neural networks use. (00:54)Why analog starts to be in the spotlight again. (04:30)How applications moving to the Edge impacts training and inferencing. (05:39)Data movement requires most of the energy consumption. (07:50)Connect with Mike Fingeroff:LinkedInConnect with Ellie Burns:LinkedInResources:Catapult High-Level SynthesisSiemens EDA

The gap between what the best AI applications can perform today versus the human brain is vast. Among many other differences, power efficiency and learning speed are two of the most challenging factors the AI & ML industry is dealing with when trying to design brain-like neural networks. Today, in the final episode of the series, Mike and Ellie discuss that gap and the challenges that hardware designers have in their design flow. They also touch on the clashing requirements of coming up with a generic AI application that can perform many tasks versus applications that perform one task really well. Tune in, to find out what the AI industry is doing to narrow the gap between the brain and artificial intelligence. In this episode, you will learn:The gaps between AI applications and the human brain. (00:45)The Holy Grail of AI: one-shot learning. (01:48)The energy consumption of the human brain versus deep neural networks. (02:50)The industry’s struggle of creating specific networks versus generic ones. (03:56)The resources required by one of the most complex neural networks. (06:08)The industry’s challenge of keeping up with the rapid changes in AI architectures. (06:57)Connect with Mike Fingeroff:LinkedInConnect with Ellie Burns:LinkedInResources:Catapult High-Level SynthesisSiemens EDA

Artificial intelligence is becoming increasingly more common in the workplace. To really understand how it works and the benefits that it can bring about, talking to people with first-hand experience is key. To learn more about how AI technology is being used, we turn to our very own experts here at Siemens.  In today’s episode, I’m talking to Roberto D'Ippolito, Senior Technical Product Manager of the HEEDS team at Siemens Digital Industries Software based in Belgium. We’ll discuss the range of possibilities within AI, where all that data comes from, and how to create value from it. AI has the potential to offer big advantages over the competition, and machine learning puts all of the information into focus. You’ll also learn where HEEDS fits into the simulation equation, the key benefits of using the technology, and the process of designing automated vehicles so that unpredictable situations are accounted for. We’ll wrap up by touching on a few misconceptions about AI, and where it might lead us in the future.  In this episode, you will learn:How we can utilize AI industrially and in general (1:48)The role of HEEDS (2:57)The key benefit of AI and machine learning technology (6:51)How the adaptive sampling strategy is being used (9:06)How machine learning meets the challenge of designing autonomous vehicles (11:02)The AV design process (14:13)Where all of the data is coming from (18:16)Challenging beliefs and misconceptions about AI (23:21)The future of AI in engineering (25:00)Connect with Roberto D'Ippolito:LinkedInConnect with Thomas Dewey:LinkedIn

Computer vision is one of the fastest-growing AI fields. This has been fuelled by the progress made in data models training and its widespread adoption. Automation that results from this increases the quality of products and lowers the cost of production.In today’s episode, I’m talking to Shahar Zuler, a data scientist and machine learning engineer at Siemens. We'll discuss object recognition in factories and the unique challenges being faced in its deployment. Tune in and learn more about computer vision in machine learning as well as the use of synthetic data in model training.Some Questions I Ask:How do you see AI impacting the industrial industry? (3:06)What are the unique challenges of employing AI/ML in the industrial environment? (10:59)What are you doing at Siemens to help solve the industrial environment’s AI/ML challenges? (19:33)What do you do to validate the correctness of synthetic data? (23:15)Can you predict what you think will happen with machine learning in the next 10 years? (26:57)In this episode, you will learn:Different tasks of computer vision machine learning (11:30)How to train an object detection model (16:34)How synthetic images are used in ML model training (20:56)How to validate synthetic data (23:38)The benefits of partnerships between Siemens and their customers (25:08)Connect with Shahar Zuler: LinkedInConnect with Thomas Dewey: LinkedIn

Artificial intelligence has come a long way in the last few years and it is making a significant impact in many industries. However, there is still notable reluctance to hand over more operations to AI-based systems because they are still not seen as being robust enough to be fully relied upon.In today’s episode, I am talking to Michael Berger, the head of Munich Reinsurance’s AI Insurance Unit, and Boris Scharinger, a senior innovation manager at Siemens Digital Industries. We’ll discuss AI performance risk insurance and the progress of industrial-grade AI.Tune in and learn more about what’s happening in the field of AI, the challenges it’s facing, and what the future holds for it.In this episode, you will learn:How AI performance risk contributes to the adoption of technology (2:32)What industrial-grade AI concept entails  (7:09)Ingredients of industrial-grade AI (8:03)Challenges facing industrial-grade AI development (10:20)Importance of AI models’ robustness (11:06)How AI risk is assessed by an insurer (13:32)Qualities of a good AI solution (14:36)Experts' thoughts on where AI will be in 3-4 years (17:35)Connect with Michael Berger: LinkedInMunich ReinsuranceConnect with Boris Scharinger: LinkedInConnect with Thomas Dewey: LinkedIn