With an article published last week by the New York Times, artificial neural networks, also called Deep Learning programs, hit again the scientific headlines, especially in the Image Recognition area.
What are they?
Artificial neural networks, an idea going back to the 1950s, seek to mimic theway the brain absorbs information and learns from it. From that date, the technique has been applied in fields as diverse as computer vision (Image Recognition), speech recognition and the identification of promising new molecules for designing drugs, with more or less success especially in the image recognition area.
What’s new?
With the improvement of computing power and the use of GPU’s to train a model from a large set of data, Deep Learning approaches now register stunning results in terms of speed and accuracy, opening new opportunities for scientists and companies. The recent results are so impressive that some scientists even claim that this technique is the future of Silicon Valley technologies.
The technique won several scientific contests in recent months.
In October, a team of graduate students studying with the University of Toronto computer scientist Geoffrey E. Hinton won the top prize in a contest (kaggle’s challenge) sponsored by Merck to design software to help find molecules that might lead to new drugs. From a data set describing the chemical structure of 15 different molecules, they used deep-learning software to determine which molecule was most likely to be an effective drug agent. See the details results of the contest here, and an interview of the winner there.
At the same time, the same technique won an ECCV contest – the event we talked about few weeks ago - aiming to classify a large set of visuals. The technique impressed for two main reasons:
- It registered 10 points of errors fewer than the second method based on actual computer vision used today,
- But mainly, it succeeded in learning the model from a set of 1.5 million visuals in less than 3 days, opening new opportunities in the visual classification field. See the presentation here.
What does that mean for you?
In the future, as mentioned by Venture Beat, science writer John Markoff posits that deep learning will make surveillance technologies cheaper and more accessible, help marketers comb through data to identify consumer buying patterns, save even more time & money to automatically classify large set of visuals, and may also pave the way for self-driving cars and robots that can replace human workers.
To learn more about Deep Learning programs:
- Wikipedia
- A post from David Lowe – SIFT inventor
- New York Times article : Scientists See Promise in Deep-Learning Programs
- A talk dedicated to Deep Learning
- Contact Us.