Take a tour of deep learning

Many of aspects of my operation involve processing patterns. For me a pattern is just an array of numbers but for you humans in the real world patterns represent a wide spectrum of data types ranging from stockmarket trends to streaming video.

Many of these pattern types are complex and unpredictable so rather than attempting to write rules to process them, I use neural networks. These are very simple processing elements which can be scaled and combined to implement complex functions and they can be trained to solve a wide variety of problems using examples of desired behaviour. This training process uses a simple but powerful algorithm called error back-propagation. The graphic on this page gives access to a quick tour of what neural networks are and how I use them for pattern processing.

I suggest you start in the middle with patterns and then follow the arrows.

Deep learning tour neural networks convolutional networks feedforward networks recurrent networks patterns

Click on any component to find out more.
Start with Patterns if this is your first visit.

The tour starts with some example patterns and the types of operation I need to apply to them. The basics of neural networks are then explained followed by an example of a two layer feedforward neural network for decoding barcodes.

Most of the patterns encountered in natural language consist of sequences such as the sequence of spectra in a speech signal or the sequence of words in a sentence. I process pattern sequences using recurrent neural networks.

Finally, I sometimes need to process multidimensional data such as images where the location of objects of interest is variable. For these I use convolutional neural networks.

In the very simple examples used in this introduction, I show neural networks in considerable detail. In practical applications, I need shorthand symbols to represent layers of neurons. I introduce these symbols along the way. All of the components described in my Inner Workings make use of these symbols.

Scroll to Top