Deep neural networks, sometimes referred to as neural nets, are a type of artificial neural network that are biologically inspired to mimic the learning model of the human brain and are used to run calculations on large sets of data. Deep neural networks can “learn” from the data they process, and they rely on multiple layers to perform complex calculations and solve problems by providing real world solutions. Deep learning networks will take in data at the input layer, process it through multiple hidden layers (with an activation function at each layer to give a specified input), and eventually provide answers as part of the output. Deep neural networks are considered a black box in the sense that while the input and output is easily understood, a close look at the structure of the analysis itself (the “black box”) will not increase understanding of the functions being performed.
Learning models of machine learning, such as deep neural networks, are able to learn more as they ingest more data. This means that a large data set is more likely to provide truer insights than a smaller data set is. Neural networks can include optimization algorithms, such as a gradient descent, to help train models. The more training a model has, the more accurate their results are likely to be once they are fully put into practice. Labeled data is also helpful in learning models, as it helps the networks learn to recognize correlations between features and their optimal results. Neural network models learn through input and output at each layer of their analysis and can sometimes include a “certainty” score at the output between 0 and 1 that correlates to the likelihood of it being right. For example, if a deep neural network is attempting to perform speech recognition, it might say that a word had a 0.98 chance of being the word “Hello.” This allows data scientists to have firm numbers when reporting on the certainty of outcomes as part of testing with neural networks.
Neural networks have recently been garnering public attention, though they have been around for decades. The New York Times has released multiple articles about their usage, and they have even been mentioned on TV shows such as Person of Interest and The 100. The increased interest in both deep neural networks and convolutional neural networks has brought attention to many ways that they can be utilized, such as:
- Natural language processing (NLP): NLP is utilized by many consumer devices, such as Alexa and Google home devices. Deep neural networks can perform language modelling and assist in translating spoken word into data that can be processed by the device. This makes it more likely that the device will correctly understand spoken instructions.
- Image processing:Deep neural networks are often used in computer vision, which processes images and videos. Images and videos can be analyzed in a variety of ways, including assisting in diagnosing health conditions and for allowing autonomous vehicles to properly “see” in order to drive safely.
- Marketing: Deep networks can assess the ways that consumers interact with retail locations, web shopping platforms, and even mobile applications to determine what consumers are looking for, and how best to market to them. This allows for businesses to optimize on the data they collect as part of a sale in order to boost sales in the future.