Artificial Intelligence, Machine Learning, Deep Learning: Optimizing Video Streaming

When discussing Artificial Intelligence, there is often some confusion with Machine Learning, and where Deep Learning comes in. In spite of these terms sometimes being used interchangeably, they are not the same thing. In this post, we will examine the differences between AI, Machine Learning, and Deep Learning, and how they each play a different role in video streaming.

Artificial Intelligence, Machine Learning, and Deep Learning: What Are They?

Artificial Intelligence is a machine simulating human thinking. Popularized as a concept in the 1950’s, most modern AI is considered to be narrow AI – which is focused on particular tasks. We are still some years away from general AI, as often depicted in science fiction (whether friendly like C-3PO, or less so, like HAL 9000,) although assistants like Alexa and Siri show promise.

Machine Learning is a subset of AI, in which a machine uses external information to make independent decisions. This is typically done by using observed data to set new rules, or issue predictions. And of course, the more information available to a machine, the better the results will be.

Deep Learning is a subset of Machine Learning (and a smaller subset of AI) that utilizes neural networks, which were designed to mimic a human brain. These neural networks connect thousands of nodes together through a variety of layers which can focus on different parts of a problem. These layers facilitate the large amount of computing power required for more complex tasks.

Main Uses

Machine Learning is a subset of Artificial Intelligence, and Deep Learning is a subset of Machine Learning. All Deep Learning is Artificial Intelligence, but not all Artificial Intelligence is Deep Learning; the terms are used to describe different ways of implementing AI.

As general AI, indistinguishable from human thought, does not exist yet, all of today’s AI implementations are task-focused. Often it’s a question of setting a few rules, then letting a program run based on those rules, often learning from mistakes.

A basic example of how Machine Learning is used in the real world is an email spam box. Over time, the spam filter will adjust its rules based on observations from the user and other sources. If emails with particular keywords in the subject line (ex: MIRACLE HAIR GROWTH SERUM) are consistently marked as spam by a user, the software will start filtering emails with those subjects into the spam folder automatically.

A limitation found with a lot of Machine Learning is that it requires a fair degree of human intervention; this is where Deep Learning comes in. Deep Learning replaces a significant amount of human intervention with processing power and can be “trained” how to learn things as how to recognize cats in videos. This is done by having the machine continually fine tune its algorithm for identifying correct answers. Over time a deep learning algorithm can even become even more accurate than humans. Deep learning algorithms are ideal for sifting through huge amounts of data, especially visual data.

A fascinating aspect of Deep Learning is the fact that within its different nodes, it can evaluate different parts of the problem at hand. Each node can assign a certain levels of importance to its observations and decisions. An example would be for a machine looking for a cat. One node observes a four-legged creature – it might be a cat, it might be a dog. However, another node observes the same creature meowing – the observation from the second node is far more important, and gives more weight to the decision.

Working Together to Improve Video Streaming

AI technology, including Machine Learning and Deep Learning, is already making a mark in video streaming. The technology has reached maturity where it can offer clear benefits.

One of the clearest examples of where AI can improve video streaming is with something called “content-aware encoding.” An AI-enabled video encoder can automatically optimize the bitrate and picture quality of a video – it will do this by running the calculations on what the picture quality would be at a variety of bitrates, choosing the lowest possible bitrate that won’t affect picture quality. This AI will use Deep Learning to first recognize the kind of video content being streamed, whether it’s a sports broadcast, a “talking head” interview, or children’s cartoon. The AI will then apply the rules refined by Machine Learning, by narrowing the calculations it needs to perform as it has already learned that fast-paced sports match requires higher bitrates than a children’s cartoon.

By optimizing bitrates in ways that simply could not have been calculated by hand, video streamers are saving significant bandwidth, and money. And as AI, Machine Learning, and Deep Learning continue to evolve, one can’t help but be excited about the future in video streaming.