Artificial Intelligence | Lesson 1.6

Deep Learning

What is deep learning? Simply put, deep learning is a type of machine learning that tries to imitate how the brain works. Deep learning models can recognize complex patterns in pictures, text, sounds, and other data to produce accurate insights and predictions. Deep learning models utilize Artificial Neural Networks (ANN) to map the inputs to the outputs. ANN consists of a set of nodes called neurons connected in a specific way (see Figure 1.3). The first column of neurons is where inputs will be given to the model and the last column of neurons is where the output is calculated. All the neurons in the middle will learn new concepts from the neurons connected to them. Today, the terms NN and DL are used almost interchangeably, meaning essentially the same thing. What then do NN or ANN have to do with the brain? It turns out almost nothing! The brain initially inspired neural networks, but the details of how they work are almost entirely unrelated to how biological brains work.

Figure 1.4 | A simplified Neural Network for demand prediction.

Let us illustrate what neural networks do with a simple example. Assume you run a website that sells t-shirts. You want to know your estimated demand based on the price. You might rightfully think the price alone is not enough and create a dataset with more inputs like the shipping costs that the customers have to pay, the amount of money you spent on marketing that t-shirt in a given week, and the material you use for that t-shirt. These are some of the factors that you think will affect the demand for your t-shirts. Let us see what a NN might look like. You know that your consumers are looking for affordability but you do not have direct information about what people think about the affordability of your t-shirt.

• You have one neuron (indicated by blue in Figure 3) whose job is to estimate the affordability of the t-shirts. Affordability is mainly a function of the price of the shirts and the shipping cost.
• A second thing you want to consider is awareness. How much are consumers aware that you are selling this t-shirt? The main element that affects awareness is your marketing. Let us devote a second artificial neuron that inputs your marketing budget and outputs how aware consumers are of your t-shirt.
• Finally, the product’s perceived quality will also affect demand, and perceived quality would be affected by marketing. The marketing tries to convince people this is a high-quality t-shirt, and also the price of something also affects perceived quality. We devote the third neuron that inputs price, marketing, and material and tries to estimate the perceived quality of your t-shirts. Now that we have figured out affordability, consumer awareness, and perceived quality, you can have one more neuron that uses these new concepts and outputs the estimated demand. This is a neural network; its job is to learn to map from supply to demand by extracting new knowledge. This is a fairly small neural network with just four artificial neurons. In practice, neural networks used today are much larger, with easily thousands, tens of thousands, or even much larger than the number of neurons.

To clarify, the way we have described the neural network, you had to figure out that the key factors are affordability, awareness, and perceived quality. To build a machine learning system using a neural network, you only need to give it the inputs and the output. The network figures out all of the concepts in the middle layer by itself, and most of the time, they are hard to interpret by a human. However, their performance is significantly higher than traditional ML models. You may sacrifice explainability to reach a higher level of performance, which is usually unachievable by humans.