Special Issue on Machine Learning for Signal Processing

Machine learning is devoted to the design of methods and algorithms able to learn from empirical data. This approach is especially important in signal and image processing, where sets of sensors, usually large and heterogeneous, provide large amounts of data, usually noisy and corrupted with various sources of interference. From a methodological point of view, machine learning is concerned with multi-dimensional and statistical signal processing, especially with problems such as detection, estimation, and optimization. In addition to classical supervised or unsupervised learning, reinforcement learning and semisupervised learning, machine learning methods include Bayesian modeling, Markov models, support vector machines, and kernel methods. It spans a broad area of applications, such as adaptive filtering, pattern recognition, scene analysis in computer vision, data mining, robot control, data fusion, blind and semi-blind source separation, sparse component analysis, brain-computer interfaces, hyperspectral images, and cognitive radio. This special issue has been designed following the IEEE international workshop Machine Learning for Signal Processing which was held in Grenoble (France) in September 2009. The papers have been extended and rereviewed. In addition to theoretical contributions in signal detection, pattern recognition and classification, blind source separation, learning theory, Bayesian learning and modeling, and to applied contributions in speech and audio processing, biomedical application and communications, we have featured three Special Sessions, one on Braincomputer Interfaces, the second on Machine Learning in Remote Sensing Data Processing, and the third one on Learning in Markov Models. The 19 papers of this special issue reflect the current trends in Machine Learning.