The Basics: AI

“Hey Alexa, add oranges to my shopping cart.”

“Ok, adding oranges to your shopping cart.”

This is a classic example of AI, a technology that can and will impact every aspect of your life. AI is not something to be feared although because its sole purpose is to help humans with trivial tasks like adding oranges to their shopping carts. AI, alongside cryptocurrencies, blockchain, gene editing, and more, are taking over the 21st century. These topics can be incredibly difficult to grasp, especially with all the breakthroughs and companies that are emerging from the shadows. I have asked quite a few of my friends to help me explain the 50 top, up-and-coming, topics! In The Basics, an all-new article series, you can learn about new technologies every day! A new article will be published every Wednesday. Now, back to the incredible technology!

For the kick-off article, I thought I would write an article series about AI, the innovator of our future.

First things first, Artificial intelligence, or more commonly known as AI is a fast-growing technology that has been around for a couple of decades. It has and will disrupt our life in every aspect imaginable. AI relates to the notion of creating computerized systems “intelligent” by utilizing a series of algorithms that will allow us to program session making a computer that can respond to different situations in appropriate manners.

One of the most common sub-groups of AI is machine learning or ML. A well-known fact about ML is that AI can narrow the scope of the creation of intelligent machines and it clearly says so. Artificial intelligent systems like Alexa or Siri have home automation devices and cars that will utilize machine learning to complete their tasks. The magic of machine learning is that you allow a machine to learn and adapt without the help of explicit programming. There are also many sublayers to machine learning.

Inside of Machine Learning, there are 5 main types of learning methods. These learning methods are very important while learning AI.

Reinforcement Learning. Another subgroup inside of Artificial Intelligence is RL, more commonly known as Reinforcement learning. Reinforcement learning explains different training methods for various machine learning models. It instructs a software algorithm on ways to conduct itself in various environments buy using a statistical concept called cumulative reward. Basically, we give the machine model a positive or negative cue on how it complies and action. Slowly and with moderation, these machine adapt to depending on the reward system instead of a supervisor.

Supervised Learning. The alteration to reinforcement learning is SL, more commonly known as supervised learning. Supervised learning is another machine learning algorithm that involves the sequential comprehension of data through machine learning algorithms by feeding the algorithm labeled data allowing it to comprehend a task by knowing an input and its respective output, and the relation between the two.

Unsupervised Learning. This form of learning is a full 180 compared to supervised learning, as you can clearly see by the titles. This is where the data is unlabeled, or that there is no corresponding output of data. The machine learning model is constructed to find triangulations between the data points. This is also known as relational statistics.

Semi-supervised learning. Semi-supervised learning is an amalgamation of the past two machine learning methods, supervised and unsupervised learning. This form of learning provides a basis for different applications suck as clustering, regression, and optimization. This occurs because it can pass a measure of labeled data alongside large amounts of unlabeled data through the machine model for training purposes.

Deep Learning. The last and final learning model is called DL, or more commonly known as Deep Learning. Deep Learning is a much more advanced paradigm that essentially builds off the basics and principles of machine learning to help structure the machine learning algorithms into formatted layers. It also uses connections via mediums of linear algebra to form ANN, or an Artificial Neural Network, computation substructure. ANN and other neural networks are formed using an input layer as a matrix, passed with wights to a neuron, which then utilizes a combinatorial calculation within it, then an activation function occurs, and finally, it concludes with the given output calculated. This process is called the multistep perceton model; however, there are many types of neural networks that can accomplish a variety of functions like CNN’s or Computer Vision.

For the past couple of months, Aditya Mittal, a close friend of mine has been working on numerous AI projects after having little to no knowledge about AI. One of his projects involved him creating something, a new “innovation” per says. Here is the overview, how it relates to AI, and the breakthrough that he found!

1. “Overview: Isn’t it incredible how humans can intuitively sense how a person is like from a few pictures or a video of them? Think about it, you generally gravitate to individuals that “look” nicer or more homely. But how exactly do we do this and is it possible to replicate this using a computer? My project aims to answer these two questions by using machine learning, specifically neural networks. By analyzing specific facial features in still images and videos, a comprehensive personality can be predicted by the neural network.
2. How It Relates to AI: My project currently contains two parts: what features can be used to determine personality, and how can we use those features for prediction. The former was discovered by doing extensive research on how facial features predict personality; by referencing books from the CIA and renowned psychologists, I was able to compile a list of facial features that could provide telltale signs of a person’s personality.

The latter uses a combination of two neural networks: one that extracts specific facial features (eyes, eyebrows, noses, and ears) and another that can predict personality from a combination of these facial features. Images are passed into the first neural network and condensed to a set of 68 points. Then, these 68 points are further condensed into a prediction of a personality.

1. What the “Breakthrough” Was: The moment that the computer was able to achieve ~60% accuracy (still working on optimizing further!), I knew that computers could be as good at identifying personality as humans. I realized that our analysis of a human’s face condenses down to a few key points, as well as analyzing the face. Our ability to generalize from taking predictions from a variety of facial features is what makes us excellent predictors of personality. My next step is to see how hyperparameters can be tuned further, and whether more facial features can be added to create a stronger personality classifier.”

Check out his LinkedIn: https://www.linkedin.com/in/adityamittal-/
Check out his Medium: https://adityamittal307.medium.com/

AI itself is very fascinating, but without a different field interjected in it, it cannot go very far. I recommend looking into things like AI in Healthcare or AI in Cybersecurity.

See you all next Wednesday with a brand-new article in The Basics!