The University of Amsterdam (UvA) and Bosch are entering into a close collaboration in the field of deep learning. Ten PhD students and postdocs will start working at the Delta Lab on the Science Park in Amsterdam. 'Delta' here stands for 'Deep Learning Technologies Amsterdam'.
Artificial intelligence is going to increasingly change our world, argue the UvA and Bosch. Machines and devices will be able to learn from experiences and adapt their actions accordingly. Deep learning is central to this.
What is deep learning?
Deep learning is part of machine learning, a research field within artificial intelligence focused on the development of algorithms and techniques that enable computers to learn. For example, machine learning can be used to teach a machine to distinguish between apples and pears. This is possible by specifying in programming code the characteristics by which an apple and a pear can be recognised, such as colour, shape and texture. The system can then be trained by having it process a 'training set' of images showing these characteristics.
In this process, it is therefore necessary to manually specify features that allow distinguishing apples from pears. However, it is also possible to deploy Artificial Neural Networks to let machine learning systems figure out these features independently. Artificial Neural Networks are based on the nerve cells, also known as neurons, present in the human brain. Here, the artificial neurons form several 'layers', each of which performs a relatively simple calculation and passes the output value to the next layer of neuron. The neurons on this layer revisit the value, perform a new calculation on it and then pass on the output value again. By each performing a relatively simple calculation, the layers of neurons can collectively perform a complex calculation.
Multiple layers
An Artificial Neural Network thus consists of multiple neurons, which form several separate layers to which the output value of a computation is passed each time. The more layers an Artificial Neural Network contains, the 'deeper' this network is. The deployment of this type of network is also known as deep learning. Those who want to read more details about deep learning can read, among others, the report 'Deep Learning: Methods and Applications' by Li Deng and Dong Yu of Microsoft Research, the research arm of IT giant Microsoft.
"Deep learning allows you to discover hidden connections in data and make predictions based on data," explains Max Welling, a professor at Uva's Institute of Computer Science. "The applications are numerous, think for example of fully automated cars, robotics, smart homes and the Internet of Things." In their research projects, the PhD students and postdocs will develop mathematical models and algorithms that enable machines to learn from information and experiences. Welling points out that the demand for smart devices and systems is high, so the importance of deep learning is also increasing. The results produced by the Delta Lab should lead directly to Bosch applications and products.
Learning from massive amounts of data
Arnold Smeulders, also a professor at the UvA's Institute of Computer Science, adds: "The amount of sensory data is greatest within computer vision. This is why this subfield of artificial intelligence is one of the most exciting targets for deep learning and why the intersection of deep learning and computer vision is so important. It poses sharp challenges to deep learning to learn from the massive amount of data. There are so many places in the world where the results could be used: in smart cameras, in self-driving cars, in home care. I actually wouldn't know where not."
By entering into a collaboration, the UvA and Bosch aim to let fundamental and applied research reinforce each other by exchanging knowledge. This should lead to more knowledge and innovations. Led by UvA professors Max Welling and Arnold Smeulders, the ten PhD students and postdocs will work together with researchers from the Bosch Center for Artificial Intelligence at Bosch's research campus in Renningen, Germany. Bosch researchers will also regularly come to Amsterdam for the research.
By: Wouter Hoeffnagel
Photo: Shutterstock
