What is Deep Learning? 

Artificial intelligence is any technique that enables computers to mimic human behavior. Deep learning is a sub-category of A.I.

The best description of deep learning I’ve come across derived from a paper by Yann LeCun, Yoshua Bengio, and Geoffrey Hinton, in which they said,

“Deep learning allows computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction.”

Ok, that may not be so clear to everyone, myself included, so let’s try to break it down further. Deep learning is a machine learning technique that was designed to mimic how the human brain functions. Lex Fridman provides a more accessible definition, “deep learning is representation learning: the automated formation of useful representations from data.” In 2013, Andrew Ng, who worked on the Google Brain Project, described the concept as “using brain simulations in hopes to…make learning algorithms easier to use…and make revolutionary advances in machine learning and AI”. 

Basically, deep learning is a type of machine learning based on the way the human brain functions.

What Can Deep Learning Do:

During an introductory course at MIT, instructors identified the following uses:

• Facial Recognition

• Image Classification

• Natural Language Processing

• Self-Driving Cars

• Handwriting Transcription

• Digital Assistants

• Search, ad, and social recommendations

• Medical Diagnosis

A Timeline (derived from Ark Invest’s chart)

The 1970s — commercial software came onto the scene with the founding of such companies as Microsoft and Oracle.

The 1980s — advances in programming made software reusable and helped expand its uses 

The 2000s — the internet boom grew the market for software from millions to billions

2012 — a team from the University of Toronto won the ImageNet Challenge with AlexNet, a deep neural network architecture, setting the stage for software 2.0

2020 — deep learning was powering most large-scale internet services such as search, social media, etc. 

Ark’s Forecast — the next decade will see deep learning create important software that enables drug discovery, self-driving vehicles, etc.

But why now?

We are living in the age of big data, and these algorithms require massive amounts of data. Processing such large amounts of data requires a lot of processing power, and chip technology has progressed significantly in recent years. Additionally, open-source tools have made building and deploying models much more accessible and streamlined. When capability meets accessibility, technology prevails. 

Processing Power 

As AI models grow larger and larger, training will require more computing power, massively expanding the need for high-powered processors. The type of information being processed influences the computing requirements; for example, conversational AI requires approximately ten times more resources than vision does, further driving the need for advanced chips in the coming years. Companies like TSMC, Nvidia, and Samsung stand to benefit from the surge in chip demand, provided they can rise to the occasion and create chips that are powerful enough to do the heavy lifting required by deep learning.

Speaking of Language (pun intended) 

Learning languages is difficult, believe me. I spent several semesters learning Italian and came away knowing how to order a drink and not much more (the important things, right?). AI agrees, but 2020 was a pivotal year because, for the first time, AI systems were able to not only understand the language but generate language with incredible, human-like accuracy. In other words, you don’t even know for certain that I, James Vermillion, actually wrote what you’re reading…spooky!

GPT-3

Yes, it sounds like a Star Wars character, and yes, it’s pretty much as awesome. OpenAI’s GPT-3 can understand language to the point of being able to translate jargon-heavy legal speak into plain English. Not only that, it can diagnose diseases, write code in many languages, and much more. Check out this example:

Robot chauffeurs?

It wasn’t long ago when self-driving cars seemed pretty far fetched, but the idea of robot-taxis doesn't seem far off anymore. That’s largely due to automakers pouring resources into self-driving systems, most notably Tesla and Waymo. The carmakers along with several tech giants are approaching the task with different approaches, but they all require powerful computing to get the job done. I for one can’t wait for the day when I can hop in my car, take a nice nap and wake up at a destination several hours away. I’m guessing the airlines aren’t so enthusiastic about such possibilities though.

What’s Next?

Those are just a few examples of how deep learning is and will change the world around us. And as impressive as those examples may be, the most exciting part is deep learning is that it’s still in its infancy, and like Gin Rummy so eloquently said “there are also unknown unknowns; things we don’t know that we don’t know,” and that’s exactly what makes deep learning so intriguing. It’s impossible to even predict where this technology will take us, because of those MF unknown unknowns! I’m excited to continue to learn more about deep learning and how it can help solve human problems, boost the standard of living, and change the world in ways we can’t yet imagine.

***This article is provided for general information and illustration purposes only. Nothing contained in the material constitutes tax advice, a recommendation for purchase or sale of any security, or investment advisory services. I encourage you to consult a financial planner, accountant, and/or legal counsel for advice specific to your situation. Reproduction of this material is prohibited without written permission from James Vermillion, and all rights are reserved.**