Nvidia and MIT are getting closer to "Computer, improving & # 39; image cleaning

One of the main features of any computer system used in movies or crime series is the ability to improve and add information in a degraded image, and thanks to the work of Nvidia, MIT, and Aalto University of Helsinki, reality and fiction are a step closer to the meeting

Detailed in an article [PDF] the Noise2Noise system could be used to clean photography in low light and astronomy, magnetic resonance imaging and delete text. "It is possible to learn to reproduce signals without ever observing their own, to a performance sometimes exceeding the training by using own copies," the paper writes.

[The neural network] is up to the cutting edge methods that use own examples – using precisely the same training methodology, and often without appreciable disadvantages in terms of time or performance.

"Of course, there is no free lunch – – we can not learn to detect features that are not present in the input data – but this also applies in training with clean targets. "

The researchers said that the neural network was working with 4,936 images of 256×256 pixel resolution of 50 subjects, and had 500 random images of 10 different subjects, during 13 hours on a GPU Nvidia Tesla P100.

The work is presented at the International Conference on Machine Learning to be held in Stockholm this week. Nvidia also announced overnight that the Autonomous Car Partnership between Daimler and Bosch had chosen Pegasus as the platform.

Launched in October last year, Pegasus is the latest iteration of the Nvidia Drive PX platform and contains a pair of Xavier system. The system is rated as capable of 320 trillion operations per second, and the company says it will be able to achieve level 5 autonomy with it.

age

Nvidia aims to make neural networks work faster and more efficiently

As data increases and models become more widespread, in-depth learning is once again "completely blocked by the material".

The Chinese technology giant's research unit claims that its deep neural network model is the first to beat humans in the Stanford Answering Dataset, but is listed first alongside Microsoft on the last

The goal of the next generation of chips is artificial intelligence, networking of neurons and the support of high quality photos and videos. create an AI, deep learning system to allow robots to learn from human demonstration (TechRepublic)

The document detailing the method is presented at a conference in Brisbane , Aus tralia.

Nvidia Clara's new supercomputer designed to act as a hospital treatment center

Nvidia's CEO and founder, Jensen Huang, said that science needed overloaded computers, that's why his company builds them