Source: Deep Learning on Medium
What is AI?
Inspired by nature’s own intelligence, artificial intelligence mimics cognitive behaviors, such as learning and problem-solving. It is comprised of algorithms enabled by constraints, exposed by representations that support models targeted at thinking, perception, and action. Phew! That’s a mouthful. Let’s break that sentence down:
Comprised of algorithms — when building AI, we build artificial neural networks. These neural networks, inspired by biology, are a method of building algorithms that are able to learn and independently find connections within the data. As per classical programming, each neuron has a set of input and output values.
Enabled by constraints, exposed by representations — when building neural networks, we have activation and loss functions that allow us to adjust and perform calculations to find connections within the data. An activation function calculates the weighted sum of its input, adds bias and then decides whether it should be fired or not, similar to how a human brain fires neurons. A loss function or error function, on the other hand, is a measure of how good a prediction model does with respect to the expected outcome.
Support models targeted at thinking, perception, and action — the goal of these activation and loss functions are to train faster, reduce overfitting and make better predictions. This allows our systems and devices to make data-driven actions based on thinking and perception.