An overview of deep learning

 

An overview of deep learning

Artificial intelligence (AI) used to be mainly focused on rule-based systems that made predictions based on preset sets of rules supplied by a subject matter expert. These systems, on the other hand, were flimsy and reliant on these "expert views," leading to their demise. These approaches were superseded by a more data-driven approach, machine learning, as the scale and volume of data expanded.

 Deep learning vs. machine learning

Machine learning is a set of methods and tools that enable machines to recognise patterns in data and utilise this underlying structure to reason about a problem. Machines attempt to comprehend these fundamental patterns in a variety of ways. But what is the relationship between machine learning and deep learning? We give an outline of how deep learning fits into this field and examine some of its uses and problems in this post. Deep learning is being misunderstood as a competitive technique in the machine learning area.

Deep Learning in humans:

The fundamental architecture for deep learning was inspired by the structure of a human brain in an effort to develop systems that learn similarly to humans. As a result, a number of key terms in deep learning may be traced back to neurology. Deep learning architecture includes a computing unit called a perceptron that permits modelling of nonlinear functions, similar to how neurons constitute the essential building blocks of the brain.

The simple perceptron is where deep learning's magic begins. The perceptron tries to comprehend data representation by stacking many layers, each of which is responsible for comprehending a different aspect of the input. A layer is a group of computing units that learn to recognise a set of values that occur repeatedly.

 Each layer of perceptrons is in charge of deciphering a certain pattern in the data. The design is termed neural networks because a network of these perceptrons resembles how neurons in the brain create a network (or artificial neural networks).

Neural network with a shallow depth:

 You observed the computations that take place within each perceptron in the previous section. You'll see how these perceptrons integrate into the network now, as well as how the flow is finished.

 A neural network has three layers in its most basic form: input layer, hidden layer, and output layer. A shallow neural network is one that has just one hidden layer, as depicted in the diagram below.

The calculations mentioned in the preceding sections take place for all neurons in a neural network, including the output layer, and forward propagation is one of these passes. The output layer must compare its findings to the actual ground truth labels after one forward pass is finished, and change the weights depending on the discrepancies between the ground truth and projected values. Backpropagation is a method that involves a backward transit across the neural network. While the mathematics underpinning back propagation is beyond the scope of this article, the following are the fundamentals of the process:

 The network aims to reduce an objective function, such as the error incurred over all points in a data sample, to the smallest possible value.

 Applications:

Health-care use cases have been a natural fit for implementing deep learning, thanks to easier access to accelerated GPUs and the availability of large volumes of data. Cancer detection using MRI imaging and x-rays has surpassed human standards of accuracy thanks to picture recognition. Other prominent health-care-related applications include drug development, clinical trial matching, and genomics.

Autonomous vehicles: Though automating self-driving cars is dangerous, it has just come a step closer to being a reality. Deep learning-based models are trained and tested in simulated scenarios to evaluate progress on anything from recognising a stop sign to spotting a pedestrian on the road.

 In the future:

With all of the knowledge in this article at your disposal, you're ready to take the next step in your deep learning adventure. There are a number of variants and enhancements to the artificial neural network that aid in attaining previously unheard-of levels of accuracy for various purposes. There will be more pieces in this series, so stay tuned.