When we talk about the Internet of Things, (IoT) we are basically talking about the ways in which things – machines, phones, wearable devices – communicate and connect with each other. Over the last decade IoT has enabled smarter cities where real and virtual are connected to improve citizen’s lives. This is hugely important as more and more people move into cities. Smarter cities utilising IoT technologies are required to accommodate the needs of the increasing urban populations and preserve natural resources.
Urban data is captured by sensors and is analysed using machine learning techniques.
Deep learning is a popular approach to data analytics achieving human-like multimedia (video, image, text) classification and prediction. However, current deep learning models are computationally too expensive to run on embedded systems. Camille Ballas’s research focuses on how to adapt the existing state-of-the art algorithms to fit the limitation of IoT ecosystems, as well as exploring deep continuous learning methods to take advantage of the continuous flow of urban data sensing in smart cities.
A novel technique called deep reinforcement learning, based on an evolutionary computing approach, reduces the need for hand-labelled data. An agent is created to learn control policies from the environment based on observed experience. This method associated with deep neural network has shown cutting edge results for gaming applications beating human performance (Atari games, AlphaGo). The machine is basically learning from its own experience.
IoT ecosystems are suitable for reinforcement learning because the data they are collecting is continuous. A prediction regarding a future event can be verified at the event time, such as traffic jam, congested city area, pollution spike.
Deep learning has a key role to play in IoT. Exploiting the continuous learning of such powerful models can accelerate the development of smart city.