A novel markovian framework for integrating absolute and relative ordinal emotion information. 

IEEE Transactions on Affective Computing, 2022

This paper explores the representation and prediction of emotions using ordinal scales, distinguishing between Absolute and Relative ordinal emotion labels. It introduces a Dynamic Ordinal Markov Model (DOMM) that leverages both types of labels to enhance speech-based ordinal emotion prediction.

Multimodal affect models: An investigation of relative salience of audio and visual cues for emotion prediction. 

Frontiers in Computer Science, 2021

How do people perceive emotions through various cues, such as speech and visual cues? Are there differences when perceiving static and dynamic aspects of emotion perception?

Belief Mismatch Coefficient (BMC): A Novel Interpretable Measure of Prediction Accuracy for Ambiguous Emotion States , ACII 2023 

Best Paper Award 🏆

Is there an interpretable and quntitative way to measure the prediction accuracy of an ambiguity-aware emotion prediction system? In this paper, we propose a novel measure called the ”Belief Mismatch Coefficient (BMC)”, that quantifies the differences in the belief that emotional states are perceived from certain regions within the arousal/valence space when comparing a predicted distribution. 

Constrained Dynamical Neural ODE for Time Series Modelling: A Case Study on Continuous Emotion Prediction , ICASSP 2023

Top 3% paper recognition 🏆

This paper addresses the challenge of incorporating dynamical constraints into deep learning models for time series prediction. It introduces Constrained Dynamical Neural Ordinary Differential Equation (CD-NODE) models, which describe the time series as a dynamic process governed by an ODE, with the rate of change influenced by the time series and input features.

From Interval to Ordinal: A HMM based Approach for Emotion Label Conversion, interspeech2023.

Ordinal labels are often derived from interval labels, typically converted into either absolute ordinal labels (AOL) or relative ordinal labels (ROL), but not both. This paper presents a novel method for mapping continuous interval labels to continuous ordinal labels, accounting for inter-rater ambiguity and rater consistency.

A Novel Sequential Monte Carlo Framework for Predicting Ambiguous Emotion States , ICASSP2022

Most automatic emotion recognition systems only consider the average rating, ignoring annotator disagreements. In this paper, we propose a Sequential Monte Carlo framework that models perceived emotions as time-varying distributions to incorporate ambiguity.