Machine Learning- Making AI Way Less Artificial and Even More Intelligent

Suraj Bansal
Nov 1 · 7 min read

Artificial intelligence (AI) has experienced exponential growth, creating an unprecedented parallel between AI and human capabilities. From understanding your facebook content preferences to speech recognition and image classification, the possibilities with AI are endless. Voting AI as your presidential elect- AI that cures cancer-AI-powered drug discovery. Concepts and ideas that we can’t even imagine integrating into our lifestyle will become an exciting reality of the future.

Now consider that researchers and AI enthusiasts alike are developing artificial intelligence-based approaches to world-wide problems like cancer treatments, preservation of the environment and educational inequity.

Yeah- not every day some 16-year old kid tells you that technology can power solutions to humanity’s most prevalent problems.

These ludicrous technologies are powered by the principles of machine learning! Machine learning algorithms enable the machine to identify data points and trends to learn from its environment and reconfigure its approach to reaching optimal task performance. This concept essentially emulates the way humans grow and adapt through the interdisciplinary use of our 5 senses.

There are 3 main branches of machine learning

  1. Supervised Learning
  2. Unsupervised Learning
  3. Reinforcement Learning

Supervised learning describe algorithms where the machine has an assigned teacher that can provide answers and confirmations to the machines attempts but cannot provide justification or explanations for why something may be correct or incorrect. They are learning from labeled data.

Imagine you’re back in grade 1, learning addition and subtraction and you’re given 100 math problems but your teacher never taught you the mathematical approaches! You guess each answer and submit the worksheet. The teacher shows you which 20 questions you guessed correctly, and which 80 questions you got wrong. Naturally, you’re confused. Rather than looking for consistencies, the first-grader would probably continue to make the same mistake.

Well- that’s okay. Over time the child’s cognitive function would grow and develop and he would recognize trends and derive the rules himself. What’s fascinating about machine learning revolves around its unprecedented speed and accuracy- machines are able to identify patterns and trends right away and decide which approach will increase accuracy and precision.

Supervised learning can be separated into two algorithmic systems.

Classification allows the machine to find discrete values and evaluate accuracy- like whether the product is classified as apple or banana

  • Binary classification refers to predicting responses with only two possible output values
  • Multi-class classification enables the model to predict multiple classes

Regression models outputs that have continuous values with calculating error values- the smaller the error margin; the greater the regression model’s accuracy

  • Determining the weight or value of the product (all continuous values)

Let’s consider the following example to put the concept of supervised learning into perspective.

Meet Freddy! Freddy was given an abundant basket of fruits and was built with the purpose of differentiating apples and bananas. Freddy makes guesses and the labels reveal whether he’s right or not- each attempt prompting the machine to alter its models. Freddy continues to train himself and slowly become more accurate until the machine feels ready to face unlabelled data and can confirm whether given objects are apples or bananas with pretty decent accuracy. Freddy would continue to model relationships and dependencies between the output and input features so that we can predict output values for new data based on these relationships which were acquired from previous data sets.

Now that Freddy has reached big boy status, he prepares himself to face real environments in which he no longer has the labels. This trains his capabilities in real life situations since you aren’t given labels or confirmation in practice. 75% of the data is used for trained data points where the machine has been fed inputs and outputs and 25% of the data was used for testing data points where it was fed only inputs.

Supervised learning algorithms are incredibly valuable- these seemingly complex principles and algorithms have already been integrated into your lifestyle without you even knowing! Email filtrations systems, facial recognition and decision support machines are only some applications of supervised learning.

Let’s revisit Freddy- except this time, Freddy wasn’t given the labels and had no idea whether the guesses he was making were true or false since there was no teacher to confirm nor deny his approximations. Could Freddy still train his models to build accuracy and differentiate the apples from bananas? Heck yeah!

These models are built using unsupervised learning algorithms and learn with unlabelled data. The neural network begins to identify patterns, structures and abnormalities without actually needing to understanding the purpose of the data. This concept can be super valuable in situations where human experts aren’t sure what they should be looking for.

Now you’re probably thinking hold up- machines can learn about data that it literally cannot even understand? That’s impossible.

Sounds like something totally unheard of right? Well, consider that thousands and thousands of years ago, humanity couldn’t understand why weather changes occurred. Yet, through trial and tribulation, we were able to deduce weather patterns and comprehend their implications. Similar to humans, the machine uses unsupervised learning to investigate trends and patterns and develop its own understanding of the data.

Unsupervised learning can be separated into two algorithmic systems.

Clustering allows machines to identify the inherent groupings of the data and places each data point into an appropriate cluster

  • This can be useful when identifying consumer buying behaviors

Association finds relations between the parameters of large data sets

  • This can be useful to identify that people who buy product X usually purchase product Y as well

Coming back to our friendly machine, Freddy- if he was given that same basket of fruits without labels, he would use unsupervised learning to analyze the shape, mass and other characteristics of the fruits. Over time Freddy would be able to cluster the fruits into apples and bananas after evaluating the characteristics without actually knowing that they are apples and bananas!

Supervised and unsupervised learning have integrated themselves into our lifestyle to an unimaginable extent. Although their capabilities for data analysis surpass that of humans, they are unable to interact with the environment. Reinforcement learning manifests methods similar to point systems with a primary objective. This form of machine learning teaches the agent to maximize reward and minimize punishment by reacting with its environment. What’s unique about reinforcement learning is that the agent has no definitive models- rather they only require the following:

  1. Representation of their environment
  2. Reward and punishment count
  3. Knows what actions/functions it can execute

Through practice, the agent will learn what actions and behaviours are conducive to maximizing its reward and it will continue to iterate its approach until they have achieved minimum risk and maximum reward. Simple reward feedback is required for the agent to learn its behavior; this is known as the reinforcement signal.

Reinforcement learning can be modeled in the following example of an old classic: Pacman (aka my childhood)!

Pacman (the agent) has functional movement and was built with the purpose of reaching max reward. Here are the representations of positive and negative reward counts in the game.

  • Small circle == 10 points
  • big circle == 100 points
  • berry == 1000 points aka jackpot
  • ghost == -500 points

After testing the environment, Pacman will associate relative levels of gratification with these values and learn how to reconfigure its environment to avoid the ghosts and collect the rewards.

Key Takeaways and Highlights

  • Machine learning has revolutionary capabilities- this technology will change the world
  • Machine learning principles power the capabilities of artificial intelligence. There are 3 primary methods by which machines learn
  • Machine learning has revolutionary capabilities- this technology will change the world
  • Supervised learning- machines train with labeled data for classification and regression
  • Unsupervised learning- machines train with unlabelled data for clustering and identifying patterns and trends of the dataset
  • Reinforcement learning- machines interact with their environment and learn from reward and punishment counts

I hope you enjoyed this article and learning the fundamentals of machine learning! Please support my efforts by doing the following!

  1. Bless my article with some claps!
  2. Connect with me on LinkedIn
  3. Check out my portfolio
  4. Stay updated on my journey and subscribe to my newsletter!

Data Driven Investor

from confusion to clarity, not insanity

Suraj Bansal

Written by

Student passionate about interdisciplinary approaches to innovation and healthcare through leveraging disruptive technologies like AI!

Data Driven Investor

from confusion to clarity, not insanity

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade