Springing into AI - Part 1: 1000 feet overview

 Artificial Intelligence (AI)....We live in the world where technology trends adapts in one form or another through our needs as humanity evolves. However the field of AI isn't a recent discovery. It has existed for years and recently gained even more visibility through its offering aimed to make our lives easier, depending on how we adopt it. 

    The first part of the series will focus on providing a high level overview of sorts into the world of AI so we have clarity on its eco-system before we start to get our hands dirty with playground of how we can use it for our custom applications.  

Ecosystem

The eco system of Artificial Intelligence is composed of various sub-sections. Figure below should provide you a concise summary of its elements.

 

Let's have a summarized break down. Note: There exists plethora of material on web to dive into every topic, I won't be covering it all, as my grand children will have their children by the time I am done writing "The end":

• Artificial Intelligence: systems designed to mimic human intelligence
• Machine Learning: algorithms that improve output automatically through experience

          These operate on data performing learnings:

• Reinforcement Learning: training algorithms based on trial and error approach
• Supervised Learning: operate on labelled datasets
• Classification: helps identify particular entity. e.g.: pen or pencil, etc. In a typical workflow, the AI model in use is trained on a known data set with ability to also handle input that it may not be aware of for edge cases. 
• Regression: dependency between variables to produce output, e.g.: stock forecasting 
• Unsupervised Learning: operate on un-labelled datasets  
• Clustering: groups certain entities together based on certain pattern, e.g. using k means clustering, genetic algorithms such as Population Based Incremental Learning (PBIL)
• Association: helps analyze relationship between entities, e.g.: grocery store (milk, butter, bread, etc.)

• Deep Learning: complex construct of network(s) working towards a goal

It is in deep learning, where we will be focusing on majorly. for the series, as one of the sections from it is that of Generative AI. The magic that many have seen today and so blindly impressed by it for its ability to do content generation (text, video, audio), analysis, etc. depending on the use case. 

       It would be a crime if I don't mention the word Artificial Neural Networks (ANN) in here, as under the hood there is a vast amount of network that believe it or not model its architecture based on how our brain functions being constituted of neurons, synapses, and the integration of many of these billions, trillions act together in harmony to provide us sense of information we so adapt ourselves to learning on daily basis. 

       The above image shows a very basic composition of a neuron drawing inspiration from how a human brain works. The AI systems that generative AI use in its Large Language Model (LLM) is composed of billions and billions of these neurons working together towards an aimed output. In summary for a neuron: 

• Input: Each input is a composition of a data point that we present to the network. These inputs can vary in its use case. Example for a classification of a pen, we may present its feature, such as shape, color, dimensions, pen type, etc. 
• Weight: When a network is initially constructed, these weights are initialized with a random value. As the network is trained and evolved, these weights are adjusted in the network memory to store its final trained values.
• Bias: The bias acts as a confidence to tilt the result of the neuron towards favour of the output that we expect. 
• Activation Function: The activation function can be any function such as for example, sine, cosine, or a gaussian curve that evaluated the computation of input, weight, and bias to product an output of a neuron. 
• Output: The output is what we expect the network to provide. Example for a classification system where we want to identify that the features we present let AI decide whether the input is a pen or not can be in the form of a binary 1 and 0 with 1 indicating the input we presented is a pen, and 0 vice versa.  As the network is trained, a process of Back Propagation takes place where at each cycle of training network, the weights are adjusted via a feedback loop from output back into the network, so it can be fine tuned to provide correct output.

       As an example, for my masters degree, I had done a classification system using ANN where the system captures the state of different type of signals that transformers emit when nearing corrosion or breakdown. These signals are usually in the form of partial discharges withe ach discharge having its own shape representation.  The use of a trained ANN since it followed a typical supervised and unsupervised learning was trained to understand and classify these signals into the type of defect transformer is emitting (if any) and empower electrical service engineers to take action for it's maintenance on it thereby preventing heavy losses. 

     This should conclude our summarized theory of a 1000 feet overview of the AI. Hopefully it provides enough information to solidify you with interest to gauge into world of it to dive deeper where relevant for your learning. In the next learning we will look at Generative AI and LLM's and their working before we start getting our hands dirty finallyyyy with coding :). Stay tuned.