Patrick Biddiscombe explains how machine learning and deep learning differ. Machine learning uses human-defined features on labeled data; deep learning uses multi-layer neural networks on unstructured data. Deep learning powers chatbots and image recognition but demands far more data and computing power.
Chapters
How smart can a machine really be? Could it ever replace the incredibly complex and organic machine between our ears? And if it did, what would that mean for us? Could we quit our day job, pack things up, and move to the nearest beach? Well, maybe not yet, but artificial intelligence is reshaping how marketers are going about their business. If you haven't already, check out our video that dives into how AI is changing the world for people everywhere. In this video, let's dig a little bit deeper and see why computers are advancing so quickly and how that's affecting us and marketers in particular. I want to talk about two programs, machine learning and deep learning. Both of these are playing a huge role in closing the gap between the human brain and the computer. How does that affect marketers like you and me? Let's break this down and look at what these terms mean, how they relate to each other, and what they're actually used for. In a nutshell, machine learning is an algorithm or statistical model that enables a machine to improve its performance through experience. Now, in traditional programming, a coder explicitly writes the rules for the algorithm to follow. But machine learning can absorb information without being explicitly programmed. And it's a way of training algorithms using data, allowing the machine to learn how to make decisions with constant human intervention. Machine learning leverages structured and labeled data to make predictions and create outcomes. It's the process of allowing machines to carry out tasks independently. Over time, it builds up knowledge about that task or object over time. However, machine learning does require an initial human input to distinguish features between objects. Before you unleash your machine learning program on the world, you need to make sure it understands the parameters of that world. Take the classic example of cats and dogs. As a human, which I assume you are, you can look at a dog next to a cat and clearly see which is which. For a computer, it's not so simple. Both have whiskers, four legs, and a tail. Initially, a coder needs to train the machine on the fundamental differences between the two animals. The different features, eye shape, posture, running style, size, and all the other small details that add up to make a dog a dog and a cat a cat. Over time, the machine will learn the difference to a reliable degree, and from there, it can be allowed to to run solo. Off it goes into machinedom. Now, imagine that applied to, say, detecting spam in your inbox. You probably know a frustrating email when you see one. All caps headline, lots of exclamation marks. An account you've never interacted with before. Google has trained a machine learning algorithm to recognize emails with these features and send them to your junk folder. It's not 100% perfect, but it catches a lot of nonsense from cluttering your inbox. So if you're a marketer who writes a lot of emails, you better make sure those headlines are strong and your content is on point. Machine learning is raising the bar on all of us. Okay, so what's deep learning then? Just as machine learning is a subset of AI, deep learning is a subset of machine learning. AI is machine mimicking human behavior. Machine learning is a way to train AI through algorithms. Deep learning is a type of machine learning that mimics the human brain. Let's talk about neurons for a minute. Your brain is full of them, about 86 billion. They make up your nervous system. Each neuron sends and receives chemical signals and electrical impulses that transmit information. Deep learning uses an artificial neural network to pick out distinct features between objects without human intervention by using the same process as the neurons in your brain. To do this, you'll need a huge amount of data to train the machine. So while deep learning is highly accurate, it takes a long time to process due to the high volume of data. If you have less than ideal hardware and lots of labeled data, machine learning is more appropriate as it's less complex and less labor intensive. But both are ways to train models and classify data, depending on the processing power and data you have available. So how does deep learning actually work? Deep learning programs are built on at least three layers of neurons. The first layer of neurons is called the input layer. This captures a piece of information about the subject. This is then passed down through the subsequent layers, called channel neurons, into the final output layer. Each neuron has a value attached. For example, a dog's paw will trigger a reaction, while cats will not. If an activation threshold is reached that matches the value of the neuron, the appropriate neuron in the output layer activates. This process repeats and adjusts to create a trained network. And because the algorithm only needs unstructured data, text, and images, it doesn't need human input to characterize objects. By observing repeat patterns, deep learning can cluster inputs and discover hidden patterns and groupings without human intervention. This is known as unsupervised learning. Take a chatbot, for example. By analyzing thousands of customer interactions, deep learning can figure out the common keywords and responses that lead to a resolution. Over time, the chatbot program will learn how to not only give useful information, but also how to respond in a way that feels relatively natural and almost human. And the great thing about deep learning is that the more interactions it has, the more effective it gets. So if you're a marketer whose company uses chatbots, deep learning provides an amazing look into your customers attitudes towards your business. These are the kinds of insights that can form super targeted messaging, better UX and smoother customer journeys. AI is the grandfather of smart computing. Machine learning is the sun and deep learning is the grandson. Both are valuable ways of training programs to ease human administration and improve internal and external facing experiences. Machine learning is leading to more accurate illness diagnosis, wine quality predictions and stock price predictions, while deep learning is creating self driving cars, faster customer support, and even safer surgical procedures. So are computers going to replace your brain? Well, not today, but they're already doing a pretty incredible job of doing human based tasks as well as if not better than people today. And they're only getting smarter. Ultimately, both models are here to help us do amazing things and make the things we already do a whole lot better. If you want to keep learning about the weird world of AI and computing, subscribe to the channel so you don't miss a thing. Now if you'll excuse me, I'm going to go deep, learn how to make some cookies. You know what I'm saying? Your boy's a little bit hungry. I'll see you next time. I can find this client info.
Have you heard of HubSpot? HubSpot is a CRM platform so it shares its data across every application. Every team can stay aligned. No out of sync spreadsheets or dueling databases. HubSpot grow better.
