• Exploration of the temporal dynamics of emotion attributes, and incorporate it in the prediction system.

• Development of novel machine learning/deep learning frameworks to capture the temporal dependencies of emotions and predict time-varying emotion states. Some key techniques applied including but not limited to: filtering methods (i.e., Kalman filter), Hidden Markov Models, Long Short-Term Memory network (LSTM), Neural Ordinary Differential Equations and Stochastic Differential Equations etc.

• Investigation and modelling of the uncertainty of continuous time series labels in the application of emotion recognition that reflects the variations of human perceptions.

• Development of comprehensive emotion representations with time-varying label distributions using probabilistic frameworks, such as Monte Carlo algorithms, stochastic process etc.

• Exploration of the relationships among different emotion representations among interval, ordinal and categorical scales, with specializion in ordinal scale for less ambiguous emotion labels.

• Development of novel framework to convert emotion labels from interval to ordinal with the integration of rater individual information and emotion dynamics.

Development of emotion recognition frameworks that predict ordinal emotions using audio, video cues, and the fusion of the two modalities.

Investigation of relative salience of different modalities (i.e., audio and video) in predicting emotion static and dynamic aspects using the developed frameworks.