What Is Deep Learning?

The idea is that by using multiple layers of nonlinear processing units, each successive layer that uses the output from the previous layer can be supervised, or, more importantly, unsupervised. This will allow the machines to analyze patterns (unsupervised) and classify (supervised) problems of all kind. 

Maybe you’ve heard of it, but probably not. For now, NVIDIA is completely dominating the field, and there is pretty much no press about it all.

Deep learning, also known as deep machine learning, is a type of machine learning that is based on a set of algorithms that creates high level abstractions in data using different model structures. Basically, it’s the future of highly complex problems - deep learning will be the solution.

The idea is that by using multiple layers of nonlinear processing units, each successive layer that uses the output from the previous layer can be supervised, or, more importantly, unsupervised. This will allow the machines to analyze patterns (unsupervised) and classify (supervised) problems of all kind.

So, basically, it means that the machines themselves can learn. This is pretty wild.

By teaching computers the ability to detect patterns and concepts in data, there will be large ramifications for speech recognition, natural language processing, video classiciation, computer vision, etc. 

By teaching computers the ability to detect patterns and concepts in data, there will be large ramifications for speech recognition, natural language processing, video classiciation, computer vision, etc. 

The most common kind of deep learning involves artificial neural networks that have been inspired by actual neural networks in our brain. This because there are two types of cells in the primary visual cortex: simple and complex cells. Artificial neural networks often depict cascading models of both kinds of cells.

As computers have given us the ability to create larger algorithms, they have also given us much larger amounts of data. However, sometimes the computers produce more data than they can actually understand. The main motivation between deep learning is to foster computational capacity in order to understand all data.

As I have mentioned, NVIDIA has been pretty much all alone in terms of major manufacturers. GPUs have had a massive impact on deep learning. NVIDIA has created the fastest deskside deep learning machine - DIGITS DevBox, which has its own Dev Program, and the possibility to build your own DevBox.

This little Linux-powered device is a supercomputer. What would have taken months can now take days.

With four Titan X GPUs with 7 TFlops of single precision, alongside 336.5 GB/s of memory bandwidth and 12GB of memory per board, it’s essentially a lightning fast powerhouse. NVIDIA DIGITS utilizes the design, training and visualization of deep neural networks for image classification. It also has a pre-installed standard Ubuntu 14.044 with Caffe, Torch, Theano, BIDMach, cuDNN v2, and CUDA 7.0. This little Linux-powered device is a supercomputer.

What does it all mean? That your results for any experiments come back a thousand times faster. What would have taken months can now take days. You can also explore multiple network architectures and you have an insanely accelerated dataset manipulation. And it’s all on a tiny computer that fits on your desk.