My Encounter with Neural Net: The Exaggerated and Acclaimed “Brain” behind Artificial Intelligence
Until my encounter with a Neural Network, I was thinking of it as a huge and complex black box that perceives things, thinks as well as makes decision the way we humans do, but now I know.
Most people, especially those that only hear and read about Artificial Intelligence (AI), think that Neural Network (the “brain” of artificially intelligent system) works pretty much the same way the human brain does. This impression about Neural Network, also known as Artificial Neural Network, may have originated from the way most professionals (in both industry and academia) and authors (of books, blog posts and articles) try to portray it, making people to think as if Artificially Intelligent system is something much comparable to a Super Human!
Until my encounter with a Neural Network, I was thinking of it as a huge and complex black box that perceives things, thinks as well as makes decision the way we humans do. And also this has been the perception of most people to the extent that some are even beginning to think that computer is a “machine that can think” as opposed to the well-known notion of it being a “machine that can’t think and act on its own until told (via instructions or code) to do so”. But having been demystified to me, I feel the need also to share with you guys the true nature of Neural Network. In this article, I will be explaining the components of Neural Network bit-by-bit.
As pointed out earlier, most definitions of Neural Network are tweaked so as to manipulate our minds towards believing that it actually works the same way our brain does. But nevertheless, I found this definition fair enough to just call spade a spade:
Artificial neural networks (ANN) or connectionist systems are computing systems vaguely inspired by the biological neural networks that constitute animal brains. The neural network itself is not an algorithm, but rather a framework for many different machine learning algorithms to work together and process complex data inputs –Wikipedia
In reality, an Artificial Neural Network is nothing other than a predictive model that makes use of Mathematical and Statistical formulae in making guesses based on the probability of event happening or not. It takes in huge amount of labeled data, process the data, identify a recurring pattern in the data, and use this to make predictions. One may ask then, “Is that not the same way our brain works”? No! It’s not, and you’ll understand why by understanding the composition of the Neural Network as well as its modus operandi.
Neural Network is made up of three distinct layers of Nodes connected by parameters known as Weights. Each of the layers performs a unique task which adds up together to give the Network the ability to make predictions. These layers are namely, Input Layer, Hidden Layer and Output Layer.
And How do they work? Continue reading! J
Input Layer
This is the first layer in the Neural Network and its task is just simply to grab the data, which must have been converted to numbers, and then multiply it with the Weights, add a Bias and then pass the result to the next layer (Hidden layer). Weights are numerical values connecting each of the nodes to another node in the next layer (i.e it represents connection between layers), while a Bias is also a single numerical value added to the result before passing it to the hidden layer (later I’ll explain the purpose and functions of the Weights and Bias).
Hidden Layer(s)
Hidden layer is where most of the mathematical manipulations of the data take place. The function of this layer is to fetch the resulting data from the input layer, and then apply a mathematical function upon the data, this results into a single probability value that can be fed into the next layer (which may be another Hidden layer or an output layer). A Neural Network may be so simple enough to have a single hidden layer (which is most unlikely in the real world applications of Neural Network) or complex enough to have thousands or even a millions, yes millions of hidden layers, each of them repeating same task of applying the same mathematical function and then add a Bias. The idea of having a multiple of hidden layers is to refine the probability so that the prediction/guess of the network can be so close to reality. The mathematical function used by this layer is, as per as the current standard is concerned, one of the following these three functions: Sigmoid function, Tangential Hyperbolic function and Rectified Linear Unit (ReLU).
After using one of the above functions (Threshold not included) to refine the probability, the result is then passed to the output layer.
Output Layer
This layer mostly fetches the probabilities supplied by the hidden layer, find a mean value and then make the prediction. A Neural Network may have a single output unit/Node if it predicts one out of two objects/classes, it may also have a multiple output units/Nodes corresponding the number (more than two) of objects/classes to be predicted.
It’s obvious that there is a huge difference, indeed very huge difference between our brain and the Neural Network. I now come to realize that, while our brain is fully aware of the context of the data we use in making predictions, the Neural Network know nothing other than apply the formula to the bunch of data to produce a probability/guess that we term as prediction. While our brain reasons with the data it comes in contact with, the Neural Network makes no sense out of the data, it rather takes in input, applies a formula and produces an output –nothing more than that!
Since it makes no sense out of the data, one may ask: How does it “learn” to make a better prediction? This will the topic for another day, as I will write comprehensively how the Weights and Biases do help the Neural Network in making a better prediction as it “learns”.
If you like what you read, don’t hesitate to lend me a clap as you’ll really encourage me to do more. I like sharing what I know, so please give me a Follow so as to receive more of what I share, for me to know that I’m not alone.
