Interactive Learning For Physiologic Signal Analysis

Emotion is a psycho-physiological process triggered by conscious and/or unconscious perceptions of an object or situation.It plays and important role in human communication and can be expressed either verbally through emotional vocabulary or by expressing nonverbal cues such as intonation of voice,facial expressions and gestures.On the other hand,affective computing is currently one of the most active research topics,furthermore,having increasingly intensive attention.This strong interest is driven by spectrum of promising applications in many areas such as virtual reality,smart surveillance,perceptual interface,etc.Affective computing concerns multi-disciplinary knowledge background such as psychology,cognitive,physiology and computer sciences.Interactive learning is a set of methods that learn by interacting with data.Under this context,active learning,online learning,reinforcement learning can also be denoted as interactive learning methods.By continuously communicating with data,the goal can be represented specifically,and the model is able to avoid and modify bias generated during learning process.We present new solutions for emotion analysis through combining interactive learning with physiologic data.The main contributions of this thesis are as follows:1.A Hessian regularized active learning method for physiologic emotion analysis is proposed.Compared with traditional Laplacian support vector machine(SVM),Hessian SVM utilizes more information.As a semi-supervised manifold method,it can also use unlabeled samples and explore the embedding geometric structure of data.In this thesis,we present a Hessian regularization based active learning model and apply it on physiologic dataset.The result shows that our algorithm perform better than Laplacian based method.2.A Laplacian online learning method for physiologic signal analysis is proposed.For emotion related sequential data,online learning method can be used to predict future value with data obtained.On the other hand,Laplacian regularized methods exploit data structure better.By combining Laplacian regularization and online learning,we develop a Laplacian online learning method for physiologic signal analysis,and achieve better performance.3.A reinforcement online learning method for physiologic signal analysis is proposed.We build an online learning method that borrow the idea from reinforcement learning.This method can be advantageous based on both online learning and reinforcement learning,thus improving emotion analytical effectiveness.