Neural networks, also known as artificial neural networks (ANN), are computing systems inspired by the behavior of the human brain. They are a subset of machine learning and are at the core of deep learning algorithms. ANNs enable machines to recognize patterns, with practical use cases that include speech recognition, object detection, and image classification.
How do ANNs Work?
Modeled on the neurons in the biological brain, neural networks are composed of a collection or layers of nodes that are connected to another. With an assigned weight and threshold values, each artificial neuron can receive, process, and pass a signal to other neurons. The weight can increase or decrease the signal at a connection depending on the received input. Moreover, neurons can have a threshold whereby a signal can only be sent when it crosses a given threshold. Each connection, called edge, is akin to the synapses that occur in the human brain.
Neurons are usually aggregated into layers. Signals travel from the input layer, passing through one or more hidden layers, and finally to the output layer. With training, the weight values are adjusted so that the final output of the network gives the expected answer to what the ANN is designed and trained for.
For an in-depth discussion on ANNs, check out IBM Cloud Learn Hub.
Types of Neural Networks
Since the oldest ANN was created by Frank Rosenblatt in 1958, it has then evolved into a broad family of techniques that are used for different purposes. The most common types are:
- Feedforward neural networks, to put simply, move data in one direction and never backward. They are foundation for computer vision, natural language processing, and other neural networks.
- Convolutional neural networks (CNN) are similar to feedforward neural networks but are commonly used for computer vision-related tasks such as image recognition and pattern recognition. To illustrate, a CNN takes an image input, assigns weights and thresholds to various objects within the image, and differentiates one from the other, or identifies the image as the output.
- Recurrent neural networks (RNNs) are identified by their feedback loops, allowing previous outputs to be used as inputs. They are mostly used in natural language processing and speech recognition.
Applications of ANNs
Because ANNs can find patterns in data or model complex relationships between inputs and outputs, they can be used across disciplines and industry use cases.
In automotive, some examples of applications include vehicle control, driver assistance systems, and vehicle identification. In healthcare, neural networks are applied in medical imaging and diagnosis. In finance, applications are data mining, visualization, and automated trading systems. In retail, they are utilized in visual search, behavioral analytics in-store, and personalization platforms. In media and entertainment, examples include general gaming, social network filtering, and recommender systems.