This is how neural networks are enabling smarter AI products
This decades old technology is finally getting its time in the sun
It is for fact that AI and machine learning have become an important part of the future. The consumer demand for “smart products” is all time high and over $15.7 trillion will be contributed to global economy by 2030 from AI. China alone plans to create a $150 billion AI sector by 2030.
Neural networks have become a particularly hot area in the AI domain lately. This new approach to computing can help us build truly “smart” products. And here’s how.
What Are Neural Networks?
Howard Rheingold once said: “The neural network is this kind of technology that is not an algorithm, it is a network that has weights on it, and you can adjust the weights so that it learns. You teach it through trials.”
The quote above gives a very elementary explanation of the concept of neural networks. However, to learn more, let’s take a look at the human brain. In the brain, a neural network is a set of interconnected neurons. The neurons connect with one another via synapses. When people learn things, the synapses become more effective. This changes the amount of influence that neurons have on one another.
In an artificial neural network, the neurons are replaced by units. Units fall in one of three categories. These are input units, hidden units, and output units. Every neural network has these. The input units receive information, the hidden units determine information will be processed and what will be learned, and the output units indicate the response the network has to the data. In order to validate their conclusions, neural networks must receive feedback. This is done through backpropagation.
Challenges And Roadblocks to Training Neural Networks And Potential Use Cases
It may be helpful to think of neural networks like pets. They need to be fed data, to be trained, and to receive feedback on their behaviors (aka the conclusions they draw). The deeper and more complex a neural network is, the more challenging training it can be. When there are multiple networks interconnected, that can be difficult as well.
Take a grocery store as an example. Imagine how long it would take a neural network just to learn the stores inventory – between 50,000 – 150,000 different items on sale. To train the netwok to recognize different goods, each will need to be turned into labelled data – a group of training images (between 2000-5000 per product, with descriptive characteristics, shot under multimple conditions. Producing that kind of data could take thousands of man hours, not to mention significant costs.
One way to get around this is to use 3D simulation and synthetic data. The Neuromation team is offering to do just that. Let’s get back to the example of the grocery store. With their technology, a 3D model of a store shelf can be created with items accurately labeled. Then from their millions of images of items generated. This data can include products from different angles and in different lighting.
Previously, efforts have been made to crowd source this work. Unfortunately, it turned out to be costly. It was also quickly discovered that humans aren’t really cut out for this type of work. They simply don’t have the speed and accuracy required, nor do they have the ability to set aside human biases.
Also discuss: e27 discussions, is Artificial Intelligence an existential risk to humanity?
Neural Networks And Smarter AI
Deep learning which is a combination of large neural networks, big data, and high performing computer systems. The potential for deep learning to have a major impact on AI is amazing. However, this isn’t just technology for the future. There are companies that are using deep learning and neural networks right now.
For example, Deep Genomics is using deep learning to develop an AI platform that can be used by geneticists and others to develop gene therapies. The idea is that human biology is ultimately too complex to understand. Therefore neural networks and other technologies will be necessary for advancements.
There’s also Affectiva. This offshoot of MIT’s media labs uses deep learning to identify emotions from still pictures and videos. Their technology can also help to identify emotion from sound clips as well. Their technology is already available to app developers and market researchers. They even offer an emotions as a service product which is a cloud based service that provides analysis of images and video.
For those of us who aren’t gene scientists, market researchers, or app developers, there’s the Arsenal. This is a little device that attaches to your DSLR camera. The arsenal analyzes what your camera is focused on, and does a comparative analysis with a database of thousands of professional photographs. Then, it adjusts your camera settings to those which will result in the highest quality picture.
This decades old technology is finally getting its time in the sun. Neural networks will continue to play a major role in the improvement of AI technology. This isn’t something that’s only happening in the lab. Real products and services are available now, and more will be seen in the future. Gaming, photography, healthcare, retail, and manufacturing are just a sampling of the industries that will be improved thanks to the use of deep learning. New AI technology will continue to bring exciting developments in the future.
—-
Featured Image: 123RF Stock Photography
The post This is how neural networks are enabling smarter AI products appeared first on e27.
