Get email delivery of the Cadence blog featured here
Cadence India hosted a first-of-its-kind seminar recently that talked about the need artificial intelligence (AI) processing will have for high speed connectivity. The central theme was that the next era of AI processing mandates hyper-connectivity to deliver an ultra-high-speed and secure interconnection between a range of users, cloud applications, servers, and data sources. This new era will be defined by highly contextualized processing and personalized experiences, delivered as and when you want them, which will impact individuals, society, industries, and the world’s economy. This is the theme that the Cadence sessions covered in technical depth, but the day started with two very interesting keynotes, Dr Avneesh Agrawal, CEO of Netradyne, and Prof Chiranjib Bhattacharyya, professor at the Indian Institute of Science, which is what this blog is about.
In the first keynote, Dr Avneesh Agrawal talked about AI in the context of road safety. Currently, said Dr Agrawal, there are a whopping 1.25 million road fatalities around the world annually. In India alone, there are 150,000 fatalities from road accidents annually. Apart from the massive and tragic human cost, there is also around $800 billion lost annually in financial damages around the world from road accidents.
Dr Agrawal presented work being carried out by Netradyne to develop maps in high definition (HD) dynamically to improve road safety standards with its Driveri solution, which could help contain the high number of road accidents. Driveri is “a vision-based driver recognition safety platform designed to enhance driver safety within the commercial fleet market.” It focuses on identifying, recognizing and rewarding positive driver performance. Driveri is driven by vision-based analytics and provides a deeper view to commercial fleet managers and owners into event context – providing them with better resources to recognize and coach drivers.
While the Driveri technology is being targeted to commercial fleets at this time, one can easily see how such information could help individuals as well – in fact, while reading more about the technology on the Netradyne website I kept thinking how useful it would be to track the driving of my 19-year old son who has just got his driver’s license!
HD maps are expected to replace existing maps or at least be an alternative to the current maps designed for human use with accuracy of only a few metres. The way HD mapping is currently done is by using LIDAR (Light Detection and Ranging), which is very expensive and based on collecting data. “Our philosophy is that if you need to improve something, you must first measure it,” said Dr Agrawal. No wonder then, that Driveri, using edge computing, will analyse every second of driving, with cameras facing the driver and the road ahead included in its 360-degree coverage.
Netradyne’s cloud-based solution is one of the many ways of mapping as there is no standard yet. The solution has mapped 5 million miles and is targeting one billion miles a month by 2021.
The second keynote was by Prof. Chiranjib Bhattacharyya of the Computer Science and Automation department of the Indian Institute of Science, Bangalore, who spoke about “Why Do we Need Depth in Deep Networks”. He that said deep networks – or the future of Machine Learning, as he called it – has arrived.
Prof Bhattacharyya started by saying that “autonomous systems are coming into our lives,” giving some well-known examples - robots reading the news in Japan, self-driving cars and robotic cameras sending pictures of the Fukushima Daiichi nuclear disaster soon after the it occurred.
Shallow or Wide Networks have just one hidden layer between the input and output layers and are easy to learn. But they can be powerful and model large classes of functions. Deep Networks, on the other hand, have two or more hidden layers between the input and output layers and are difficult to train/learn but they can learn far more complicated functions. Shallow Networks though are based on Memorization while Deep ones are based on Generalization, though he added that these are only the initial findings from experiments in Machine Learning.
The two keynotes were very different in nature, but both were fascinating in their own ways and showed how both academic research and practical applications of AI and machine learning are working in parallel, if not collaboratively, towards the same goal – exploring and improving AI and ML innovatively to help improve the human condition.
I expect we will hear more along these lines at the keynotes at CDNLive India, which is coming up on Sep 6 & 7. Watch this space in the next week for more on what to expect at CDNLive India.