The engineers knew they were on to something massive when they decided to put a camera on a mobile phone. Fast-forward to 2017: amateur photography and selfie culture are at an all-time high, and the trend is likely only to grow in the future. The latest innovation in this space is the use of artificial intelligence and neural networks to create some of the most psychedelic pictures on the web: Welcome to Project DeepDream.
Overview
Lately, you might have found social media flooded with images that could best be described as ‘trippy,’ ‘psychedelic’ or ‘outright crazy.’ These pictures are courtesy Google’s Project DeepDream, or Prisma or in some cases, both.
Both applications take ordinary pictures and deliberately convert them into over-processed images with many layers of effects.
What Is It?
Google’s DeepDream is primarily a computer vision software which utilises convolutional neural networkto deeply analyse images and find patterns within them, which could be later manipulated to create desired effects.
DeepDream was originally designed for the 2014 ImageNet Large-Scale Visual Recognition Challenge (ILSVRC) and made available to the public in July 2015.
Soon after that, Google published this software and made the DeepDream code open source; enabling anyone to use the system and make their image manipulation software and apps for the web, desktop, and mobile phones. One such app is, the much loved, Prisma.
What Are Neural Networks and AI?
The concepts that lie at the core of these capabilities are Neural Networks and Artificial Intelligence. Let’s elaborate on these.
One of the key elements of Artificial Intelligence is its ability to learn, adapt to the situation and make decisions accordingly, without human interference.
Neural Network is a tool of Artificial Intelligence, which is focused on developing systems and machines which can be ‘trained’ over a period into recognising patterns and predicting models, like a biological brain.
The ‘Brains’ Behind It
Even though Neural Network is designed to train itself and make decisions like a biological brain would, it does so by using mathematical algorithms and not biological neurons.
In that case, why is it still called ‘neural’ network? That is because the system is based on an extensive collection of computational neural units, each with its function, and connected in the form of layers, similar to how biological neurons are connected by axons.
How Does It Work
We know that Neural Networks (and DeepDream) operate in layers. Each layer of the network, starting from the most basic and moving up, analyses the picture and adds its set of effects.
For Example, if a part of the cloud in a picture looks a bit like fish, the basic layers of the network will analyse that and make it look ‘more’ like a fish. As the image is passed on to consecutively more complex layers, that part of the cloud starts closely resembling a fish and eventually looks exactly like one.
What Are Its Applications?
DeepDream is in its early stages, and there’s a lot of work to be done and a lot to be understood about the functioning of the network.
As it continues to improve, Artificial Intelligence will significantly see a boost in its applications as its ability to concentrate on more important features and ignore the unimportant ones will improve.
And as AI improves, so will a robots’ ability to understand the world around them and to better blend in the society. The future looks bright.
Applications like DeepDream and Prisma are classic examples of taking a seemingly mundane piece of technology and doing something extraordinary while also engaging the masses. With an increase in its user-base, feedback from users and overall development of the technology, Artificial Intelligence is a step closer to being an integral part of our everyday lives.
References:
https://en.wikipedia.org/wiki/Artificial_neural_network
https://research.googleblog.com/2015/06/inceptionism-going-deeper-into-neural.html
http://mastersofmedia.hum.uva.nl/blog/2015/10/30/googles-deepdream-algorithms-on-lsd/
LSI Keywords:
Google DeepDream Project
Neural Network technology
DeepDream Algorithm
DeepDream working process
Neural Network layers
