Industry Outlook: Powering New Applied AI Solutions #emtechdigital

by

Naveen Rao is the Corporate Vice President and General Manager, Artificial Intelligence Products Group, Intel.

Naveen views Moore’s law as an economic law.

Naveen is talking about biological comparisons in order to use biology as a guide. There is a great diversity in the ecosystem. There are 300 pounds of insects for every pound of humanity. Genetic processes are a learning process. Naveen believes there is diversity across the use cases for artificial intelligence. There will be winners across the segments.

Naveen was at Nervana from 2014 to 2016. It was changing the way we think of computation and AI. Their machine intelligence platform allowed the application of state-of-the-art neural network learning to nearly any kind of data problem. On August 9, 2016, Nervana was acquired by Intel, for an estimated $408 million.

Naveen and Bill Dally are both sharing the stage in a small panel. They are discussing the opportunity for more speed from analog systems. Bill discussed how it is more than just speed and supporting 13 frameworks to make it easy for developers to save their time.

They are discussed that AI is not just Deep learning. Their is Bayesian inference and neuromorphic AI and much more. They will support all of the approaches.

However, Bill Dally says neuromorphic has never been competitive. Spiking is just too inefficient. Naveen Rao says that the human brain has an existence proof that there is interesting research possibilities. Bill thinks that there the calcium channels of the brain are driving the success of the spiking solution in the brain.

There was a question about the AI race between China and the USA. The moderator felt there is not a zero sum game in the AI competition.

Hardware innovations will track software innovations. There is a co-evolution. There emerged a need to handle sparse matrices efficiently. This came from software and algorithms and hardware adapted to meet this need.

Nvidia has not found a case where analog is better yet. They are looking for it.

Nvidia has competing teams. One team will try to do the best job with analog and another will try to do the best job with digital. They will then pick the best option.

Naveen says Quantum computing still needs to show that there is an advantage as a substrate.

Bill feels that qubits to get to 99.99% accurate instead of 99%. He thinks quantum computers are decades away from being commercially relevant.

Bill dismisses the NISQ at the 100 qubit level. Nextbigfuture notes that the IonQ trapped ion could achieve the level of accuracy in the near term.

Nervana is now part of Intel.

Naveen gave another on AI and hardware in the middle of 2018.

SOURCES – Live Reporting by Brian Wang of Nextbigfuture at EmTech Digital 2019, Intel, Youtube