Deep Learning Simplified
With over 20+ years of experience as senior executive leader, Manuj Desai is currently responsible for process innovation, pricing and under writing applications, InsureTech automation, decision tools development, digitization, catastrophic modeling and predictive modelling.
Taboo’s exists in the technology market around concepts like Analytics, Data Visualization, Machine Learning and Deep Learning, due to insufficiency in understanding the concepts or having issues relating to the concepts as they are still in the nascent stages. Let us take a few steps back to understand these fancy terms.
The example that we all can relate to is driving a car. When an individual learns to drive, in most countries, the first exam they give is a written test after which they get a learner’s license also called learner’s permit. The individual then starts to learn the tricks needed to drive a car either from a driving school or an adult, someone who has driven a car before.
When “Lessons are learned from someone who has driven the car in the past”, we would equate this to having ready-made formulas provided to us.
After the first few rounds of understanding the ABC of the car i.e. the “Accelerator, Break and Clutch (in the case of manual cars)” and the workings of the gear box (again in the case of manual cars), the individual starts to maneuver the car in normal city traffic. The first few dry runs in the traffic are always horrendous for anyone; we would call this the test conditions.
With practice, one starts to gain confidence on driving, which I like to call as experience of driving. Initial runs are always on the cautious side while one starts to understand the environment. After which comes a “beautiful” memory of someone coming too close to your car or your car accidently
touching someone else’s. We start to build upon experience on what to do and what not to do while driving, coaching our brain to calculate the distance, speed, relativity, acceleration limits, deceleration capabilities along with understanding the limits of the car we are driving which includes parameters like torque, horsepower, engine size, fuel efficiency, etc.
Analytics in this case is the “creating the magic formulas” and the necessary calculations of multiple parameters in your brain to ensure that your car runs smoothly in the environment it is. Using Mathematics (mainly probability, statistics & trigonometry) and understanding the laws of physics, knowingly or unknowingly, our brain starts to create the necessary formulas after understanding the visual dashboards created from the data of the car. The data that our brain collects (we would call it as parameters) are speed, torque, acceleration and the gear motion every time we drive the car.
The technology world we live in today relates a lot to what we humans have seen and can relate to
Machine Learning is the experience that our brain, “our machine” gains, based on what we learn. There is supervised machine learning and unsupervised machine learning. The supervised is the case when we learn under the supervision of someone who has driven the car in the past, in this case the formulas are provided to us and we have to operate our car under the guidelines of those formulas only.
Unsupervised is when we start to drive on our own and the calculations we start to make in our brains on what works and what does not. When you start to update the calculations and formulas in your brain depending on your understanding of the situation, that is called, machine learning or in simple terms your machine is learning from your experience of the situation to update the formulas accordingly.
Deep Learning is the years of unsupervised learning that we gain after being on the road on our own. After I learned my driving, I was very comfortable driving on any road, any terrain having a clear notion that I can drive; until I hit the icy conditions of USA. A frozen road was a concept that my brain “machine” was not used to, meaning I did not have any formula in my machine for that situation. While there was documentation about how not to apply breaks on an icy road, my machine still did not register the impact until I was in that situation. When on an icy road, my car slipped and took two turns on a freeway before coming to a stop was the time I had to update my machine to change the formulas I had stored in my brain. This time it was not just the update of a formula it was creating new formulas for the situation faced. When we adapt to the environment and new formulas get created based on how we understand the environment, we start to enter the world of deep learning. Deep learning includes coaching my neural network in terms of reflex action to react in a particular situation under the new formulas created. Neural network is simplified in the example to equate back to the human neurons.
The technology world we live in today relates a lot to what we humans have seen and can relate to. The article helps to simplify some of the jargons used around us to demystify our understanding.