A guide to machine learning in search: Key terms, concepts ...
DISCLAIMER: This blog is fully automated, unmonitored, and does not reflect the views of Garett MacGowan. The ML model may produce content that is offensive to some readers.
This blog post was generated with a (potentially) real article title as the only prompt. A link to the original article is below.
Photo by Stephen Phillips - Hostreviews.co.uk on Unsplash
Generated: 5/2/2022
A guide to machine learning in search: Key terms, concepts ...
Machine learning is at the heart of AI. It has been since the beginning. The only difference is that, today, machine learning is a thing. As a consequence, Google’s own definition of AI has changed. In the company’s own view, “AI encompasses a set of innovations in machine learning, other technologies, and high-level understanding of data.”
To that end, machine learning has seen remarkable growth at Google. It has become the go-to tool for image recognition, speech and language processing. It powers Google Search and serves as the platform for Google Assistant. Its technology is also part and parcel of Google Lens, a technology search users can use to find information with their camera lens. If machine-learning algorithms are the heart of AI, machine learning itself is the brain that drives it.
In the last year or two, the search team has devoted particular attention to machine learning. The change of direction was in part to help build a platform to develop advanced machine learning algorithms. However, given Google’s long and deep history in searching, machine learning has also been central to search for a long time. With that in mind, here are some ideas on the subject.
Machine learning (ML) is the science of getting software to learn. Google already has a history of doing this with great success – for example, with Google Voice and Google Now, the company has achieved some impressive feats in giving its users context-based, personalized services. Machine learning, which is related to artificial intelligence (AI), is more than that, though. Machine learning has come to have a life of its own.
For search, it is not just about giving personalisation a new shot in the arm, it’s about achieving true general intelligence in search. At the heart of true general intelligence are algorithms that are able to process information like a human – and the ability to learn is how search would achieve that. It is that ability that search engines like Google search have always relied on; machine learning provides the key that will allow search to go from being just a collection of useful tools to becoming an integral feature of our world.
Machine learning and search have been entwined from the early days of computer science: Search is what we use to find information online. We are a tool for the machine-learning algorithms we use to learn about the world. The two have become so closely linked that it is hard to imagine one without the other. Machine learning is the foundation on which artificial intelligence rests.
So how does search fit into the bigger picture of machine learning? Here is the short answer: The way search engines work today rely, by and large, on algorithms that learn what we want to find by analyzing the things that we search for. For example, Google has built an algorithm that extracts information from the content of pages and can tell that the page contains links, images or videos, or information about an issue, or about a person in some cases; in fact, it has to extract information on a wide range of topics from text-only pages. In search, Google also uses the most popular questions asked on other search engine site pages to improve results. But the search engine has made a point of noting that it does not understand these pages. It uses machine learning to learn what they are about, not humans.
This is a very small part of where machine learning has to be integrated; it is just a brief glance at the power that can be applied to any aspect of a search engine or website.
We’re also at a point in time where the search engine is being augmented by the internet of things, where it is possible to learn about people and about objects in the physical world in the same way that it is possible to learn about information on websites and with people. This is the key area of innovation in machine learning at the moment, and it is very much the subject of the search team’s focus.
When we talk about machine learning, one of the things that comes up is the fact that the search engine is changing. Search is now becoming an engine of knowledge – in the best cases, a “learning engine” or, in the least case, a collection of tools people use to find information.
It’s important to keep that in mind. Google is in the business of using information and technology to make the world a better place – both for its users and for the people that drive Google’s own growth. That will always be where the search team is looking. The search team itself started life with a small team of people that, early on, were working on image search; it grew out of that, and is now a larger group of people focused on many things, including machine-learning development. Google has been using many types of machine learning. In a very general sense, machine learning comprises the technologies that allow you to train a computer to make good judgments.
However, much of the work the team has been doing is related to learning algorithms specifically. What is machine learning? The term covers several areas of technology that together are able to make machines learn from the world; it covers technology that is more than that, and is in the broadest sense related to artificial intelligence. What is “machine learning”? Is it something that goes beyond artificial intelligence? The answer is that it does apply to artificial intelligence. The difference is, “machine learning algorithms” can be taught on a “case-by-case basis”, unlike intelligence.
The search team is also interested in a broader question: What defines a learning algorithm? That question raises questions about the way that we think about algorithms. What sets apart machine learning algorithms from other types of algorithms? It is a common misconception that machine learning is some magic that a search engine can do in addition to algorithms that analyze the search information. This is wrong, because many common types of algorithms can be expressed in some form as a search-like expression – for example, algorithms that search for duplicated elements or for links. Machine learning, as we have seen, involves learning about various types of information in the world from an analysis of case studies.
With that in mind, and to make it easy to follow, we are going to take a look at key terms and ideas relating to the field of machine learning. This is a brief definition, not an exhaustive one. The hope is to have created a basic understanding of some of the things that are going on, as we explore the different aspects – “machine learning in search” – in future articles.
Key terms
The term “machine learning” can be loosely defined a number of ways. An algorithm that learns is a common way of capturing it. The term is a loose and imprecise one, but is still commonly used to mean this. It is important to understand that, when we talk about “machine learning in search”, we are talking more about the processes around searching. To the best of my knowledge, there is currently no “search engine that learns for your company”, so machine learning is an important aspect of search but it is not the main topic; that is more about “machine learning in search” and how the search engine can be used.
Machine learning in search generally refers to the development or use of machine learning algorithms, or sometimes just machine learning. In those contexts we can talk about algorithms like neural networks, which are used in the same way that statistics were used over the years to learn a statistical model. Neural networks don’t have to be neural in a real sense, but they usually are, and they are used in a way that they can learn.
Machine learning in general can be associated with algorithms which work by “trying out a series of models on small chunks of data, and identifying which of those models best predicted new chunks of data”. So we might have a model that helps a computer understand a language, or it can be used for image recognition.
When we talk about machine learning in particular, as something specific, we are typically talking about supervised learning, also known as “pattern recognition”. What’s important is that we shouldn’t be thinking about these terms as if one or the other is more important, because both are equally useful. What’s important is that we understand the context in which those terms or concepts are being used.
A very basic question often posed is: How does a search engine apply machine learning? A search engine does this on two levels. At the simplest level, the searcher types in a query, and the engine returns a list of things that it thinks are relevant (hopefully). How that happens is different on a regular search engine from how it happens on a machine-learning based search engine. However, on both, the core of the process is “how the learner works”.
This is the first key aspect, and we’ll get into how to define the different forms of search engine later.
Machine learning is a broad area, and also as broad as what we know about searching. It is about applying algorithms to problems in a way that takes advantage of the way they learn that we want to understand. A common approach across machine learning is the use of neural models, but the approach isn’t limited to that. There are a number of ways that people are building machine learning systems that are specific to the problem at hand. That makes machine learning in search something a little different from machine learning as a whole.
It is also important to understand the various fields of machine learning that are subsets of the concept. The most recognizable machine learning subsets are supervised learning, reinforcement learning and unsupervised learning.
Machine learning is at the heart of AI. It has been since the beginning. The only difference is that, today, machine learning is a thing. As a consequence, Google’s own definition of AI has changed. In the company’s own view, “AI encompasses a set of innovations in machine learning, other technologies, and high-level understanding of data.”
To that end, machine learning has seen remarkable growth at Google. It has become the go-to tool for image recognition, speech and language processing. It powers Google Search and serves as the platform for Google Assistant. Its technology is also part and parcel of Google Lens, a technology search users can use to find information with their camera lens. If machine-learning algorithms are the heart of AI, machine learning itself is the brain that drives it.
In the last year or two, the search team has devoted particular attention to machine learning. The change of direction was in part to help build a platform to develop advanced machine learning algorithms. However, given Google’s long and deep history in searching, machine learning has also been central to search for a long time. With that in mind, here are some ideas on the subject.
Machine learning (ML) is the science of getting software to learn. Google already has a history of doing this with great success – for example, with Google Voice and Google Now, the company has achieved some impressive feats in giving its users context-based, personalized services. Machine learning, which is related to artificial intelligence (AI), is more than that, though. Machine learning has come to have a life of its own.
For search, it is not just about giving personalisation a new shot in the arm, it’s about achieving true general intelligence in search. At the heart of true general intelligence are algorithms that are able to process information like a human – and the ability to learn is how search would achieve that. It is that ability that search engines like Google search have always relied on; machine learning provides the key that will allow search to go from being just a collection of useful tools to becoming an integral feature of our world.
Machine learning and search have been entwined from the early days of computer science: Search is what we use to find information online. We are a tool for the machine-learning algorithms we use to learn about the world. The two have become so closely linked that it is hard to imagine one without the other. Machine learning is the foundation on which artificial intelligence rests.
So how does search fit into the bigger picture of machine learning? Here is the short answer: The way search engines work today rely, by and large, on algorithms that learn what we want to find by analyzing the things that we search for. For example, Google has built an algorithm that extracts information from the content of pages and can tell that the page contains links, images or videos, or information about an issue, or about a person in some cases; in fact, it has to extract information on a wide range of topics from text-only pages. In search, Google also uses the most popular questions asked on other search engine site pages to improve results. But the search engine has made a point of noting that it does not understand these pages. It uses machine learning to learn what they are about, not humans.
This is a very small part of where machine learning has to be integrated; it is just a brief glance at the power that can be applied to any aspect of a search engine or website.
We’re also at a point in time where the search engine is being augmented by the internet of things, where it is possible to learn about people and about objects in the physical world in the same way that it is possible to learn about information on websites and with people. This is the key area of innovation in machine learning at the moment, and it is very much the subject of the search team’s focus.
When we talk about machine learning, one of the things that comes up is the fact that the search engine is changing. Search is now becoming an engine of knowledge – in the best cases, a “learning engine” or, in the least case, a collection of tools people use to find information.
It’s important to keep that in mind. Google is in the business of using information and technology to make the world a better place – both for its users and for the people that drive Google’s own growth. That will always be where the search team is looking. The search team itself started life with a small team of people that, early on, were working on image search; it grew out of that, and is now a larger group of people focused on many things, including machine-learning development. Google has been using many types of machine learning. In a very general sense, machine learning comprises the technologies that allow you to train a computer to make good judgments.
However, much of the work the team has been doing is related to learning algorithms specifically. What is machine learning? The term covers several areas of technology that together are able to make machines learn from the world; it covers technology that is more than that, and is in the broadest sense related to artificial intelligence. What is “machine learning”? Is it something that goes beyond artificial intelligence? The answer is that it does apply to artificial intelligence. The difference is, “machine learning algorithms” can be taught on a “case-by-case basis”, unlike intelligence.
The search team is also interested in a broader question: What defines a learning algorithm? That question raises questions about the way that we think about algorithms. What sets apart machine learning algorithms from other types of algorithms? It is a common misconception that machine learning is some magic that a search engine can do in addition to algorithms that analyze the search information. This is wrong, because many common types of algorithms can be expressed in some form as a search-like expression – for example, algorithms that search for duplicated elements or for links. Machine learning, as we have seen, involves learning about various types of information in the world from an analysis of case studies.
With that in mind, and to make it easy to follow, we are going to take a look at key terms and ideas relating to the field of machine learning. This is a brief definition, not an exhaustive one. The hope is to have created a basic understanding of some of the things that are going on, as we explore the different aspects – “machine learning in search” – in future articles.
Key terms
The term “machine learning” can be loosely defined a number of ways. An algorithm that learns is a common way of capturing it. The term is a loose and imprecise one, but is still commonly used to mean this. It is important to understand that, when we talk about “machine learning in search”, we are talking more about the processes around searching. To the best of my knowledge, there is currently no “search engine that learns for your company”, so machine learning is an important aspect of search but it is not the main topic; that is more about “machine learning in search” and how the search engine can be used.
Machine learning in search generally refers to the development or use of machine learning algorithms, or sometimes just machine learning. In those contexts we can talk about algorithms like neural networks, which are used in the same way that statistics were used over the years to learn a statistical model. Neural networks don’t have to be neural in a real sense, but they usually are, and they are used in a way that they can learn.
Machine learning in general can be associated with algorithms which work by “trying out a series of models on small chunks of data, and identifying which of those models best predicted new chunks of data”. So we might have a model that helps a computer understand a language, or it can be used for image recognition.
When we talk about machine learning in particular, as something specific, we are typically talking about supervised learning, also known as “pattern recognition”. What’s important is that we shouldn’t be thinking about these terms as if one or the other is more important, because both are equally useful. What’s important is that we understand the context in which those terms or concepts are being used.
A very basic question often posed is: How does a search engine apply machine learning? A search engine does this on two levels. At the simplest level, the searcher types in a query, and the engine returns a list of things that it thinks are relevant (hopefully). How that happens is different on a regular search engine from how it happens on a machine-learning based search engine. However, on both, the core of the process is “how the learner works”.
This is the first key aspect, and we’ll get into how to define the different forms of search engine later.
Machine learning is a broad area, and also as broad as what we know about searching. It is about applying algorithms to problems in a way that takes advantage of the way they learn that we want to understand. A common approach across machine learning is the use of neural models, but the approach isn’t limited to that. There are a number of ways that people are building machine learning systems that are specific to the problem at hand. That makes machine learning in search something a little different from machine learning as a whole.
It is also important to understand the various fields of machine learning that are subsets of the concept. The most recognizable machine learning subsets are supervised learning, reinforcement learning and unsupervised learning.