It's Been Iv Years Since Nosotros Posted Deep Learning Is Vc Worthy (And What's Coming)

Let's run across how it turned out.
August 21, 2014
From recode:

Nervana Raises Second Round This Year, every bit Silicon Valley Bets Big on Deep Learning

Nervana Systems, to a greater extent than or less other instrumentalist inward the all of a abrupt hot “deep learning” space, has closed its instant circular of working capital alphabetic quality inward the final iv months.

The San Diego startup said it raised $3.3 1000000 inward Series H5N1 funding led past times DFJ, which comes on tiptop of a $600,000 seed round in April.

DFJ’s Steve Jurvetson will take a spot on the company’s board as role of the latest investment. Allen & Co., AME Cloud Ventures in addition to Fuel Capital likewise participated.

Deep learning is a cast of artificial tidings that researchers conduct maintain credited alongside recent leaps in areas similar spoken language recognition in addition to icon search. That has sparked growing involvement in Silicon Valley, with Google, Facebook in addition to Twitter making notable acquisitions or hires inward recent months in addition to diverse prominent players betting their ain money on the space.
As Re/code explained inward an before piece:

Deep learning is a cast of machine learning inward which researchers endeavor to educate calculator algorithms to spot meaningful patterns past times showing them lots of data, rather than trying to computer programme inward every dominion nigh the world. Taking inspiration from the agency neurons operate inward the human brain, deep learning uses layers of algorithms that successively recognize increasingly complex features — going from, say, edges to circles to an pump inward an image.
Notably, these techniques conduct maintain allowed researchers to educate algorithms using unstructured data, where features haven’t been laboriously labeled past times human beings ahead of time. It’s non a novel concept, but recent refinements conduct maintain resulted inward important advances over traditional AI approaches.

Nervana is aiming to distinguish itself inward the nascent land past times focusing on edifice hardware optimized for deep learning software — in addition to vice versa....MORE

 On August 9, 2016 Intel purchased Nervana for a rumored $408 million.

Here's Semiconductor Engineering, September 11, 2018 alongside the storey that spurred this stroll downwards Memory Lane:

Intel’s Next Move

Gadi Singer, vice president in addition to full general managing director of Intel’s Artificial Intelligence Products Group, sat downwards alongside Semiconductor Engineering to verbalize nigh Intel’s vision for deep learning in addition to why the society is looking good beyond the x86 architecture in addition to one-chip solutions.

SE: What’s changing on the processor side?
Singer: The biggest alter is the improver of deep learning in addition to neural networks. Over the past times several years, the changes conduct maintain been therefore fast in addition to profound that we’re trying to assess the potential in addition to what nosotros practise alongside it. But at the same time, you lot likewise ask to measuring dorsum in addition to recall nigh how that fits inward alongside other complementary capabilities. That’s role of the overall transition.

SE: What actually got this going was a recognition that you lot could develop algorithms alongside machines rather than past times hand, right?
Singer: The master approach was from the 1960s, in addition to it went dormant until [computer scientist Geoffrey] Hinton in addition to others constitute a ameliorate agency to bargain alongside multiple layers effectively inward the early on 2000s. The large breakthrough, when deep learning was recognized every bit a major computational force, occurred a twain of years ago. That was when ImageNet showed you lot tin attain near-human accuracy alongside icon recognition. We started to run across bully results on spoken language recognition. Around 2015 in addition to into 2016, results began to await promising plenty to live on a major alter factor. At that time, the footing was basically flat, at to the lowest degree inward price of images. It was relatively elementary images in addition to simple, direct speech. Most of the drive was proving things were possible alongside deep learning therefore you lot could attain to a greater extent than or less grade of accuracy or to a greater extent than or less laid upward of results. In price of the agency to create in addition to evidence models, the principal architectures were CPUs in addition to GPUs. The agency to practise the employment before that was C++, similar to a greater extent than or less of the predecessors to Caffe, in addition to alongside proprietary environments such every bit CUDA. It required a lot of expertise in addition to drive inward edifice the compute architecture, every bit good every bit inward the deployment. In price of who was involved, if you lot await at the engineering inward the land today, those were the early on adopters.

SE: What’s changed since then?
Singer: Over the final few years, we’ve seen the coming of historic catamenia of deep learning. The information itself has croak much to a greater extent than complex. We’ve moved from 2D to 3D images. We’re working alongside Novartis, which is looking at 3D microscopic images of cells, trying to position potentially malignant cells. The images themselves are 25 times to a greater extent than complex inward price of data, but what you’re identifying is a to a greater extent than refined model.

SE: Where does Intel jibe inward alongside these architectures. One of the large problems alongside AI in addition to deep learning is they’re changing quickly, therefore you need a real flexible architecture. What does Intel computer programme here?
Singer: In the past, the employment contention was clear. You knew what you lot needed for a graphics fleck or a CPU fleck ii or 3 years out, in addition to companies competed on having the best solution for a known problem. Deep learning is a infinite where companies compete based on who best understands the employment every bit it evolves. You ask an architecture that is able to empathize in addition to foresee trends, in addition to live on create for what is coming when it’s out at that spot inward the marketplace inward total production in addition to deployment—not when it’s beingness designed in addition to tested....

 ...MUCH MORE