ML & Machine Learning

Neural Networks
The above image shows educational version of software for pattern recognition, which I created in 2007. The neural networks I used is rather very simple - Hopfield neural network. Yet, despite simplicity, results can be powerful, in scenarios, where use of such neural networks makes sense, like for instance - pattern recognition. As such neural networks are similar connections between neurons in brain. Neurons can be connected with each other on many ways, through many layers, having number of inputs and outputs designed for solving specific problems. 

In the above example Hopfield Neural Network was used for pattern recognition, specifically for symbols and alphabet characters recognition. It had two modes, one for training and another one for recognition of unexpected "damaged" or incomplete characters based on 7x7 matrix (49 inputs, and 49 outputs). In the image we can see that the network recognised partly "damaged" cross symbol.

There is no neural networks (except human brain), which can fit all possible usage scenarios. For this reason some neural networks are better suited for one category of problems, whereas other for different problems. There are ready to use, Machine Learning platforms including pre-trained neural networks selected for solving specifically the problem you need to solve. 

Machine Learning
Machine Learning systems include Neural Networks as one of many possible components. In general, Machine Learning is supposed to "learn" like humans, to be able to solve problems or make predictions, and neural networks are "trained" in various ways, depending on their complexity.

For the educational pattern recognition application described here every "lesson" or attempt to recognise symbol properly, there were thousands of multiplications required form making the 49x49 matrix brain to be able to "remember". So, with every lesson, the numbers in the 49x49 matrix were modified in such way, that the "brain" was able to recognise unknown shape/pattern/symbol with higher probability. The point is, that even for simple machine learning system, the amount of memory operation is rather high, and it requires much energy, or huge amounts of energy for advanced generative AI systems.