- 5th November 2019
- Posted by: Manolis
- Category: Artificial intelligence, machine learning
AI deep learning is still in a nascent stage…
Artificial Intelligence – deep learning is a subset of machine learning. The man who coined the phrase ‘machine learning’, IBM developer Arthur Samuel, once described it as a “field of study that gives computers the ability to learn without being explicitly programmed.” At the time he was teaching IBM’s systems to play checkers…
Where machine learning algorithms work well on datasets that have up to a few hundred columns, unstructured datasets like images or videos have so many features that traditional methods of training them are unfeasible.
As Datarobot explains: “Deep learning algorithms learn progressively more about the image as it goes through each neural network layer. Early layers learn how to detect low-level features like edges, and subsequent layers combine features from earlier layers into a more holistic representation. For example, a middle layer might identify edges to detect parts of an object in the photo such as a leg or a branch, while a deep layer will detect the full object such as a dog or a tree.”
Artificial Intelligence Deep Learning: Layered Complexity
Artificial intelligence deep learning systems, in short, have a series of layers through which data is passed and given a judgement or ‘weight’ at each layer. Layers in the network are tiled grids and each tile is a neuron. Each system will have an input and an output layer, but it may have a host of hidden layers in between, the more layers the more ‘potentially’ complex the neural network is.
If you take the image of a cat and feed it to a neural network it will chop it up into its defining attributes such as fur, tail and slit-shaped pupils. Crucial to success is the amount of data or images you feed it related to cats.
For instance, if you feed the network five images of cats that were all taken outside, then the network might assume that green grass is a key component in the thing we call a cat. Just like Plato’s cave allegory, it can only know what it is shown.
However, if you show it a million images of cats in different settings, shapes, and colors, an artificial intelligence deep learning system can, via the layered process, train itself to understand what a cat is. If done correctly it can identify a cat in images that are blurry or ambiguous in nature. Of course, it’s not just image detection that a neural network can be trained to do: here is a video that shows how a car can be trained to race around a track using a very simple version of a deep neural network. It demonstrates the rinse and repeats the learning process of these systems.
Artificial Intelligence Deep Learning
This type of AI training has made impressive strides when it comes to image and language models. In the medical field, networks are trained to spot cancers in brain scans, while virtual assistants such as Siri and Alexa use it to understand language.
Currently, the system is proving to be a critical component in finding Exoplanets in space, which are just on the very edge of what we can detect with current technology. Writing in a paper that details how NASA is using the Kepler Space Telescope with deep learning models, its researchers wrote: “A deep image classification model might first detect simple edge features, which can then be used to detect curves and corners, and so on, until the model’s final feature layer can discriminate between complex objects.”
“Our neural network model is able to accurately distinguish the subtle differences between transiting exoplanets and false positives like eclipsing binaries.”