Deep learning has already completely taken over speech recognition, face recognition, and object recognition. Going forward deep learning is the technology that will solve sensing and perception for machines. You're going to want sensing and perception in your phone and every computer you use, but even more than that: every drone, every self-driving car, every kind of robot is going to need multiple deep learning chips. Each of those chips is going to need vastly more FLOPS and more memory bandwidth than the biggest CPUs and GPUs of today.
Beyond sensing and perception, I believe that deep learning will also be the technology that solves planning, natural language, reasoning, and creativity for machines. This is much more speculative, but you can see the beginnings of planning in the DeepMind Atari work, the beginnings of natural language processing and reasoning in machine translation and various question answering systems, and the beginnings of creativity in Deep Dream and other generative models people have done. Of course once all those pieces are solved then AI is solved. The market for a technology that solves AI is practically unlimited.
Intel is constantly searching for the next big application that will require more processing power so people need to buy faster CPUs and they can justify spending $X billion on their next fab. In recent years they've been struggling to find it. Deep learning is it. The appetite for FLOPS and memory bandwidth is unbounded for the foreseeable future. Unfortunately for Intel, CPUs are weak on both compared to GPUs. Maybe Xeon Phi can morph into a deep learning system?
I am similarly very bullish on this field, even for what I once thought were incredible claims that it can eventually understand the semantics behind documents, do question and answering, and even reason.
The main idea is that reasoning is just a sequence of 'thought vectors' that can be encoded within a recurrent neural network.
Sounded almost outlandish to me until I watched Richard Socher's lectures and started to understand that words can be represented as vectors, and that these vectors can be then encoded into new vectors of even higher representation. 'Thought vectors' may not be so outlandish after all.
You will likely be interested in this recent presentation by Hinton where he explains how he believes the brain does deep learning: https://www.youtube.com/watch?v=kxp7eWZa-2M
Beyond sensing and perception, I believe that deep learning will also be the technology that solves planning, natural language, reasoning, and creativity for machines. This is much more speculative, but you can see the beginnings of planning in the DeepMind Atari work, the beginnings of natural language processing and reasoning in machine translation and various question answering systems, and the beginnings of creativity in Deep Dream and other generative models people have done. Of course once all those pieces are solved then AI is solved. The market for a technology that solves AI is practically unlimited.
Intel is constantly searching for the next big application that will require more processing power so people need to buy faster CPUs and they can justify spending $X billion on their next fab. In recent years they've been struggling to find it. Deep learning is it. The appetite for FLOPS and memory bandwidth is unbounded for the foreseeable future. Unfortunately for Intel, CPUs are weak on both compared to GPUs. Maybe Xeon Phi can morph into a deep learning system?