C07: Soft Computing: Current Trend of Machine Learning in Computer Vision

Paper Submission for C07

SMC2018:C07 submission site (external site)

Abstracts

Although the history of Neural Networks backs to 1940s, however, since past few years it attracted researcher around the world and became very popular and successful in various application domains, including image classification, natural language processing, as well as data representation for more general AI tasks, e.g. state encoding in the game of Chinese Go. Addition of more training layers to the neural network architecture resulted in deep networks and eventually resulted in the rebranding of ‘Neural Networks’ to ‘Deep Learning’. However, many of the core concepts of deep neural networks were available in early 80’s and 90’s, however, in recent few years that neural networks came in the lime light and have witnessed many success. Among many factors that changed overtime, the most important ones were the availability of massive labelled datasets and GPU computing facilities.
Computer Vision (CV) is a science that enables complete analysis of useful information from single or multiple images. It has numerous applications in several disciplines starting from neuro-biology to signal processing. It draws support from fields like machine learning and deep learning since most probabilistic models and fuzzy algorithms have been developed there. In turn, it lays solid foundation for Computational Intelligence, enabling the system to gather firsthand information, opposed to data fed in by the user, and thus, shares a symbiotic relationship with this field.

One major challenge in Computer Vision is practical implementation of a filter that provide a noise-free desired output that can be further analysed by expert systems as required. Filters analysed by the CNN method reveal that the very first layers learn the low-level features as the top levels understand the high-level semantics. CNN method is a deep-learning based model and is a strong, yet simple application of machine learning and computational intelligence in solving a grass-root level issue in Computer Vision. The design and development of such algorithms involve complicated mathematics, preventing direct mathematical modelling of the filter. However, using Computer Intelligence allows the implementation of various such solutions, whose modelling is not possible with human intelligence. Computer Vision and Computer Intelligence go hand-in-hand with each other. Computer Vision broadens the scope for the system to implement its learning, while Computational Intelligence methods enhance the ability of Computer vision to analyse the image captured.

This special session aims to bring together the current research progress on Deep Learning theories and applications. Special attention will be devoted to handle advanced issues of network architecture design, real-time performance criteria for various applications and diverse application areas.

Session Chairs