Artificial Neural Networks - Artificial intelligence:
Decision trees, while strong, are a easy representation method. While graphical on the surface, they may be seen as disjunctions of conjunctions, and hence are logical representation, and we call such type of methods symbolic representations. In this lecture, we see at a non-symbolic representation method also known as Artificial Neural Networks. This term is often reduced to Neural Networks, but this annoys neuron-biologists who deal with actual neural networks (inside our human brains).
As the name shows, ANNs have a biological inspiration, and we concisely look at that first. Following this, we see in detail at how data is represented in ANNs, then we see at the easiest type of network, two layer networks. We see at perceptions and linear units, and talk about the boundaries that such easy networks have. In the next lecture, we talk about multi-layer networks and the back- propagation algorithm for learning these networks.
Biological Motivation
In our conversation in the very first lecture about how people have reacted the question: "How are we going to have an agent to work intelligently", one of the answers was to realize that intelligence in individual humans is resulted by our brains. Neuro - scientists have told us that the brain is made up of architectures of networks of neurons. At the most essential level, neurons may be seen as methods which, when provided some input, will either fire or not fire, depending on the character of the input. The input to fix neurons arises from the senses, but in common, the input to a neuron is a set of outputs from other neurons. If the input to a neuron goes over a fix threshold, then the neuron will fire. In this way, one neuron firing will influence the firing of various other neurons, and information may be stored in terms of the thresholds set and the weight assigned by every neuron to every of its inputs.
Artificial Neural Networks (ANNs) are constructed to mimic the behavior of the brain. Some ANNs are built into hardware, but the wide majority are simulated in software, and we focus on these. It's important not to get the analogy too far, because there actually isn't much similarity between artificial and animal neural networks. In particular, while the human brain is predictable to contain around 100,000,000,000 neurons, ANNs usually contain less than 1000 comparable units.
Moreover, the interconnection of neurons is much superior in normal systems. Also, the method in which ANNs store and manipulate information is a gross overview of the way in which networks of neurons work in normal systems.