18911 Hardy Oak Blvd

San Antonio, TX 78258

  • White LinkedIn Icon
  • White Twitter Icon
Send Us a Message

What is Deep Learning and Why you should care about GPUs?

It has been almost 60 years since the term artificial intelligence was first coined by John McCarthy, one of the founders of the discipline. A machine is considered to have human-level intelligence if it passes the Turing test, which is used to determine whether a computer can mimic human cognition. An intelligent machine can perceive its environment and take appropriate actions to achieve a specific goal.

Deep learning is based on class of algorithms called “Artificial Neural Networks” inspired by the structure and function of the brain to perform a machine learning task. Deep learning is a subset of machine learning, which itself is a subset of Artificial Intelligence. Deep learning algorithms find associations between a set of inputs and outputs to extract patterns in the data. The learning process is achieved by tuning and adjusting weights in the network. The goal of the deep learning algorithm is to minimize the error between predicted and actual values.

Deep learning is a subset of machine learning, which itself is a subset of Artificial Intelligence.

The “Deep” in Deep Learning refers to the multiple numbers of stacked layers in these neural networks. These structures are therefore distinguished from the simpler single-layer networks by their depth. Early neural network models were shallow, consisting of one input and one output layer with one hidden layer. An architecture is considered “Deep” if it has more than three layers. Stacking multiple layers is what allows these networks to achieve their state-of-the-art accuracy that sometimes surpasses human-level performance.

There are two major architectures of deep networks: Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). CNNs learn higher order features in the data through convolutions. They are the most popular networks used for image classification and recognition tasks. RNNs, on the other hand, have shown great promises in Natural Language Processing (NLP) tasks. What makes RNNs ideal for text and handwriting is their ability to recognize patterns in sequences of data. Recurrent networks possess a memory capable of storing and recalling previous associations

Why Deep Learning Now?

The reason we are seeing a rapid adoption of deep learning because we finally have two main key ingredients to make deep learning practical:

1. Availability of large amounts of high-quality labeled data (Big Data) required to train deep learning algorithms. To improve the model’s accuracy, it is crucial to have enough training examples to teach our neural networks.

2. Availability of high-performance graphics processing units (GPUs). Deep learning tasks require serious amounts of multi-core computing power. GPUs are specialized computer processors originally designed to process 3D game rendering, but their capabilities are now being harnessed to accelerate compute-intensive tasks. With the recent breakthrough of Nvidia’s parallel computing platform CUDA, researchers and developers are able to dramatically speed up their computing applications. With the availability of thousands of cores inside a GPU, training time for a deep learning model is significantly reduced from months to minutes.