History of Artificial Neural Networks

Deep learning to find cats: Google’s neural networks

Recent reports about Google’s new neural nets are very interesting: it’s a biologically (i.e. human brain) inspired machine learning method using many thousands of computer processors to recognise an object which has been exposed to the artificial neural network millions of times. For us humans, we take object recognition for granted. It’s almost effortless to tell what an object placed in front of us is because we are very good at "common sense" reasoning and somehow the human memory can easily link all kinds of information together, including the very different views of the same object, to generate a concept of that object. To do this with computers, artificial neural network researchers have been working hard for decades to create a simple representation of real world objects. Although thousands or even millions of training data are needed for a neural network model to achieve this, the vast supply of free Internet resources (i.e. texts, pictures and videos) makes the gathering of training data a lot easier.

The real breakthrough in Google’s work relies on the use of millions of YouTube videos. Recognising sensible objects within those videos from scratch (i.e. unsupervised learning, meaning no hints/help from a human teacher) was only made possible due to the recent increases in processor speed, low costs of processing power and millions of (cat) videos uploaded by YouTube users. In the past, recognising objects from thousands of individual images was difficult enough; you would think that doing it from millions of videos may still sound impossible? Actually, we speculate there even more reasons to work with videos. Imagine how recognising cats from individual images works: One individual image gives the computer (with its artificial neural network) a particular view of a particular type of cat, while the next image may give a completely different view for another type of cat. In this way, the inconsistency of individual images actually increases the difficulty of forming the target pattern of a cat in the neural network. However, since a video is constructed by motion pictures (i.e. continuous movements/changes are displayed in a long sequence of images), it provides multiple contexts for the same object (such as a moving cat) to allow easier recognition from the neural network. Google’s new work in this area is great (but not magical) and it will rely heavily on a huge amount of hardware resources to analyse millions of videos in order to reach a reasonable learning rate (i.e. high accuracy of object recognition), so don’t expect it to recognise everything just yet!

Source: DelivApp.com

You might also like:

Marcus Hutter - History of Artificial Intelligence (no music)
Marcus Hutter - History of Artificial Intelligence (no music)
Avi Peled vid hys.mpg
Avi Peled vid hys.mpg

You think outside the box!

2011-07-13 08:44:32 by polaris_8

I don’t think your ideas are crazy. In fact what you suggest is something that actually would change the human race forever on this planet. However remember as it stands the reason why humanity has not yet made the jump into the future is because we are held back by our energy issues. Mainly our use of oil as an energy source is holding everything back. Which in turn is also about control by the monetary system we currently have set up as a means for the human race to survive. This includes big banks, wall street, the stock market, and governments. And as we speak our middle class is becoming extinct due to the shift of the global market place as we compete for jobs in cheaper overseas markets. In essences’ we are becoming a third world country simple because big companies, banks and...

Continues . . . .

2009-03-04 10:40:19 by jungle-red

Turning to economic woes, he cited the slide in major stock indexes, the decline in U.S. gross domestic product and Washington’s bailout of banking giant Citigroup as evidence that American dominance of global markets has collapsed.
“I was there recently and things are far from good,” he said. “What’s happened is the collapse of the American dream.”
Panarin insisted he didn’t wish for a U.S. collapse, but he predicted Russia and China would emerge from the economic turmoil stronger and said the two nations should work together, even to create a new currency to replace the U.S. dollar.
Asked for comment on how the Foreign Ministry views Panarin’s theories, a spokesman said all questions had to be submitted in writing and no answers were likely before Wednesday.

Medical Device Veteran and Author, Rudy Mazzocchi is Honored as One of Top …  — PR Web
The event will take place on February 12-13th at the Computer History Museum in Silicon Valley, and will honor the 50 Top Innovators in Life Sciences, Cleantech and Software/IT. As a Co-founder and/or ... As recently described in The 2012 Best of ...

Related posts:

  1. Fast Artificial Neural Networks
  2. Usage of Artificial Neural Networks
  3. Models of artificial neural network