Intel holds its annual Developer Convention in San Francisco every year bringing together industry experts from the domains of data science, AI hardware, machine and deep learning. This year, Intel AIDC 2018 held a similar event in Bengaluru, India for which we were graciously invited. Intel brought down some of the top AI experts from India and overseas for the sessions. The highlight of AIDC 2018 was Gadi Singer’s keynote which touched upon all of Intel’s AI software and hardware offerings, and elaborated upon some of the key milestones that Intel has achieved thus far.
Gadi Singer (Vice President, Architecture General Manager , Artificial Intelligence Products Group) was joined by Amir Khosrowshahi (VP and CTO of AI Products), Dr. Mohit Kumar (Sr. Principal Data Scientist, Flipkart), Dr. Santanu Bhattacharya (Chief Data Scientist, Airtel), Dr. C. V. Jawahar (Amazon Chair Professor, CVIT, IIIT-Hyderabad), Tapati Bandopadhyay (General Manager and Practice Head, Wipro HOLMES AI and Automation Ecosystem, Wipro) and other luminaries from the AI domain. Here’s everything that went down at Intel AI DevCon 2018, Bengaluru, India.
Gadi Singer started off by explaining that Intel choose Bangalore for AIDC 2018 India owing to the vast number of AI startups and academics that are concentrated in the Silicon Valley of India. Gadi then touched upon the how the Industry has now accepted AI as a mainstream technology. The current trend of AI acceptance began in 2014-2015 when the hardware and technology started focusing on AI as a key focus area. We are now at a place in the AI revolution where AI hardware and software ecosystems have developed to the extent that businesses and academia can finally afford to experiment with new AI methodologies.
“It took a lot of effort and it was all about experimentation. We’re talking really, of the coming of age of Deep Learning”, said Gadi Singer. “And it should come of age. This is about real-world use-cases. So instead of just looking at an image and identifying a violin and a cat. This is now about going through complex 3D medical imaging and identifying potentially malignant cells which is 25-30 times the complexity.”
Gadi then highlighted the fact that many open-source deep learning frameworks have become available over the past two-three years. Another shift in the hardware space is how the technology has now shifted focus from training to inference workloads in order to cater to the changing pace of the AI revolution. One such announcement was the launch of the Spring Crest based NNP L-1000, the first Nervana Neural Network Processor which is slated to be made available in 2019.
Gadi Singer was followed by Amir Khosrowshahi who began with a showcase on Intel’s new open source math library which is aptly named Intel Math Kernel Library for Deep Neural Networks (INTEL MKL-DNN) which has functions for handling Matrix Multiplication, Pooling, Convolution and Normalization among others. Amir then moved onto the optimisations that Intel has done to allow such AI libraries to work on Intel hardware. So much that performance numbers for certain workloads have supposedly increased by a factor of 200.
Amir also put the focus on OpenVino, a compiler that allows AI models trained on powerful frameworks such as Tensorflow on a wide variety of hardware without having the programmers worry about the underlying hardware architecture. He then had a couple of Indian AI startups to share their experiences with Deep Learning frameworks and libraries. Novartis was one such partner who have been using AI along with Intel hardware to analyse cellular images and they’ve improved their training time by 6x.
Next up was Prakash Mallya, MD, Sales and Marketing, Intel India who elaborated upon Intel’s initiatives to build the AI community in India. Intel had announced a couple of measures in April 2017 which involved enabling students and industry professional to gain better access to AI software and hardware. This involved a couple of workshop and outreach programmes at the University level.
“In April 2017, we committed to train 15,000 developers, students, and professors in AI. We have long surpassed our goal; today, over 99,000 people across more than 100 organizations have been engaged with, through relevant trainings and workshops,” said Prakash Mallya. “We have been able to accomplish this through our strong collaboration with the academia, government, and the private sector. This streak will continue as we engage with local platforms such as Analytics Vidhya*, to train data scientists and developers.”
Aside from the sessions, Intel also had a couple of demo zones for some of the aforementioned hardware and software libraries. Here are some of the ones that caught our eye.
https://twitter.com/digitindia/status/1027132322032885761?ref_src=twsrc%5Etfw