What is deep learning?
Deep learning is a type of machine learning that can recognize complex patterns and make associations in a similar way to humans. Its abilities can range from identifying items in a photo or recognizing a voice to driving a car or creating an illustration. Essentially, a deep learning model is a computer program that can exhibit intelligence, thanks to its complex and sophisticated approach to processing data.
Deep learning is one kind of artificial intelligence (AI), and it is core to how many AI services and models function. Large language models (LLMs) such as ChatGPT, Bard, and Bing Chat, and image generators such as Midjourney and DALL-E, rely on deep learning to learn language and context, and to produce realistic responses. Predictive AI models use deep learning to gain conclusions from sprawling collections of historical data.
How does deep learning work?
Usually, using a computer program requires precise inputs for obtaining the correct outputs. Deep learning, in contrast, can take arbitrary or imprecise data and produce a relevant output. For example, a traditional computer program might be able to tell if two digital portraits are exactly the same. A deep learning model might be able to recognize similarities in the portrait's subjects, even if the portraits themselves are different.
Deep learning relies on large data sets and lots of computational power — and as the availability of those two things has increased, deep learning models have become more sophisticated. Today, big data collections and GPU-powered cloud computing services make deep learning more accessible to developers and everyday users than ever before.
What is the difference between machine learning and deep learning?
Machine learning is a type of computer program that can learn without explicit instructions. Deep learning is a specialized kind of machine learning, just as a jet is a specialized kind of airplane. Both involve letting a computer program learn on its own from a set of data. However, deep learning can do more, just as a jet is more powerful than a propeller plane or a glider.
Deep learning can also learn from unlabeled data, while more basic machine learning models may require more context about the data they are fed in order to "learn" correctly. Finally, deep learning models are built using neural networks. Machine learning models may be built on neural networks, but this is not always the case.
How is deep learning used?
Deep learning already has a plethora of applications in the world today, and new uses are still being discovered. Current use cases include:
Voice assistants
Self-driving cars
Predictive models
Image creation
Conversational AI chatbots
Medical research
(Techniques like low-rank adaptation can help developers quickly adapt a deep learning model for a new use case.)
What is unsupervised learning?
In the field of machine learning, unsupervised learning is a way to identify patterns and associations in a large data set without any context as to what the data set contains. In contrast, supervised learning provides example inputs and outputs to a model. Deep learning can use supervised learning for training models, but its ability to learn unsupervised sets it apart from other types of machine learning.
Imagine a machine learning model is fed examples of news articles, with an indication of what topic each article is about. After sufficient training, this model might be able to "write" an article on a given topic. This is supervised learning.
Now, imagine a deep learning model is fed a series of example news articles, with no guidance as to what each article is about. Such a model, if it is powerful enough, might be able to write an article on a given topic, with the topic alone provided as the input. This is unsupervised learning.
What is unlabeled data?
Unlabeled data is data without classifications, tags, or labels. Unlabeled datasets can contain any arbitrary data and can take any form: random photos, video compilations, long lists of file names, log data, or a combination of all of the above. The news articles provided without context (from the previous example) would be an example of unlabeled data.
Deep learning models are able to contextualize and "understand" unlabeled data. And typically, the more data they are fed, the more sophisticated the models become.
Unlabeled data and object storage
Unlabeled data is often unstructured as well. Unstructured data does not follow any particular format, and thus can contain any type of digital information. Object storage is often used for saving unstructured data of this kind. Such data collections can grow indefinitely, and object storage is a highly scalable, fairly cost-effective way to store them.
Deep learning models grow more effective when they are given large data collections to learn from, even when that data is unlabeled and unstructured. Object storage is therefore an important resource for deep learning models.
What is a neural network?
A neural network is a type of machine learning architecture based on how a human brain functions. Neural networks are a collection of nodes; each node is its own processing unit. Data that is statistically significant gets passed along from one node to the next.
These nodes are spread across at least three layers: an input layer, a hidden layer, and an output layer. Usually there are several nodes in each layer. There can be multiple hidden layers, and deep learning models tend to have many.
Think of a neural network as a team working together to solve a problem. Each member of the team is responsible for one aspect of the problem, and once their role is fulfilled, they hand it off to the next team member. Finally, the team arrives at a full solution together.
Neural networks have existed for decades, but modern-day deep learning uses more layers than neural networks of the past. The deep learning models of today also have access to vastly more compute power and data than ever before, enabling developers to accelerate the advancement of AI technology.
How does Cloudflare enable the construction of deep learning models?
Cloudflare helps enable developers to easily build AI applications that can be accessed from anywhere with minimal latency. Cloudflare Workers AI provides access to serverless GPUs on Cloudflare's global network for running advanced machine learning models. And Cloudflare R2 is object storage with no egress fees for more cost-effective storage of large data sets, which can be used to train deep learning models.
Learn all about Cloudflare for AI.
FAQs
What is deep learning?
Deep learning is a subset of machine learning that identifies complex patterns and makes human-like associations. Deep learning models can train themselves to make these associations using unlabeled data, which makes them more powerful than other types of machine learning.
Why is deep learning considered foundational to AI?
Deep learning forms the foundation for many AI services, such as large language models (LLMs) and image generators. Deep learning is what enables those types of models to learn language, understand context, and generate realistic or usable outputs.
What are some applications of deep learning?
Deep learning powers many AI-based services, including voice assistants, self-driving cars, predictive models, image creation, and natural language processing (NLP). Deep learning can also assist with research and data analysis.
How does deep learning differ from traditional machine learning?
Deep learning is a specialized subset of machine learning with more advanced capabilities. Deep learning models are capable of unsupervised learning — they can train themselves on large data sets without any context or labels for the data within.
What is unsupervised learning in deep learning?
Unsupervised learning is when a model identifies patterns and associations in data without needing prior context or labeled examples. Deep learning excels at this, which enables deep learning models to find structure and patterns in raw, unlabeled data.
What is unlabeled data and how does deep learning use it?
Unlabeled data has no classifications or tags. Any arbitrary type of data or file can be in an unlabeled data set. Deep learning models can analyze and interpret unlabeled data when shown enough of it.
Why is computational power important for deep learning?
Deep learning relies on processing very large datasets and needs significant computational resources in order to do so. Advances in data availability and GPU-powered computing have made deep learning more practical.
What is a neural network?
A neural network is a machine learning architecture inspired by how the human brain works. It consists of interconnected nodes across input, hidden, and output layers. Each node processes and passes along significant data to nodes in the next layer.
What are the layers in a neural network and what do they do?
Neural networks have an input layer, one or more hidden layers, and an output layer. Each layer’s nodes process data and pass relevant information forward to ultimately produce an output.