By Ankur Saxena
How smart are you when it comes to the nuances of Artificial Intelligence (AI)?
When you read about the future of AI, it can seem like there are a lot of buzzwords being thrown around in the media. Differentiating among AI, machine learning and deep learning technologies can be confusing, especially when terms are being used interchangeably. Let’s begin by clearing things up with a few definitions.
First, there is AI, which refers to intelligence exhibited by machines in the form of human cognitive functions like visual perception, speech recognition, decision-making, and language translation.
One of the approaches in the field of AI is machine learning, wherein software “learns,” or extracts insights from data, without being explicitly programmed on how or where to look.
Finally, there is deep learning (within the field of machine learning), which uses algorithms inspired by the structure of the human brain to make predictions about comprehensive networks and data sets.
Confused? Think about AI as the umbrella. All machine learning is AI, but not all AI is machine learning. Similarly, all deep learning is machine learning but not vice-versa.
Going a step deeper, to enable software to make complex predictions without being explicitly programmed to do so, we need to perform two tasks. First task is feature extraction or feature engineering, which refers to ascertaining the features to analyze from the data. Second task is data optimization, i.e. weighing the variables in the available data to make accurate predictions.
In a simple linear regression analogy, determining the independent variables x and y is feature extraction and allocating the weights b and c is data optimization.
z = a + b*x + c*y
Now that we’ve addressed the basics, let’s explore what makes each AI subset so important and unique.
Machine learning is like an engine built to make predictions for a specific use case, using algorithms that “learn” through data training. Algorithms first make predications based on real data and domain examples that already have known outcomes. The algorithms then track the difference between its prediction and the real output, constantly adjusting the weights of the inputs to improve the accuracy of its calculations. Thus, the quality of prediction improves with more training.
For all its promise of automating model building, machine learning still requires humans to perform feature specification (while it takes care of data optimization). Deep Learning solves for both data optimization and feature extraction through a layered approach involving more training data and additional computation.
To understand what is special about deep learning, consider using a computer to identify cats in a collection of random images. A machine learning algorithm would require humans to specify defining characteristics of cats, then train the algorithm on labeled data of cat images. A deep learning approach would determine cat features – as well as the weights of these features – and be able to classify the images all on its own.
About the Author
Ankur Saxena joined Cisco Corporate Development in 2016 and is currently focused on investment and acquisition opportunities in Internet of Things and Artificial Intelligence & Machine Learning.