User Tools

Site Tools


10a.002.TAU_WP4 - Computationally Efficient Graph-Embedded Subspace-Learning Methods

Project - Summary

In this project, we propose a novel subspace learning framework for obtaining a mapping along with data description in low-dimensional feature space. The proposed framework presents the problem in the form of graph embedding. It includes the previously proposed subspace one-class techniques as its special cases and provides further insight into what these techniques optimize. The framework allows for incorporating other meaningful optimization goals via the graph preserving criterion. It reveals a spectral solution and a spectral regression-based solution as alternatives to the previously used gradient-based technique. We combine the subspace learning framework iteratively with Support Vector Data Description (SVDD) applied in the subspace to formulate Graph-Embedded Subspace Support Vector Data Description (GESSVDD). We experimentally analyzed the performance of newly proposed different variants. We demonstrate improved performance against the baselines and the recently proposed subspace learning methods for one-class classification. In this project, we also present a novel dataset that captures facial expressions and the associated physiological signals, such as heart rate (HR), electrodermal activity (EDA), and skin temperature (TEMP), under different stress levels. The data was collected from 20 participants at different sessions for 26 hours. The data includes seven different signal types, including both computer vision and physiological features that can be used to detect stress. This part of the project is carried out in collaboration with the University of Louisiana at Lafayette.

Project - Team

Team Member Role Email Phone Number Academic Site/IAB
Moncef Gabbouj PI 358 40 073 6613 Tampere University
Fahad Sohrab Researcher 46 962 9962 Tampere University
Paul Ervi Project Mentor N/A Dead Set Bit

Project - Novelty of Approach

We formulated subspace learning for one-class classification in the graph embedding framework and discussed the novel insights obtained from this formulation. In particular, we showed that subspace learning for SVDD applies a weighted Principal Component Analysis (PCA) over the support vectors and outliers to define the projection matrix. We discussed how this information could be combined with other data relationships in the optimization process via an adaptable graph. We also formulated a novel GESSVDD with gradient-based, spectral, and spectral regression-based solutions and different adaptable graphs.

This work also presents a multimodal stress-emotion dataset containing stress data from wearable devices and emotion data extracted from facial expressions through a video feed. We analyze human expressions and biometric signals under different stress levels. The dataset presented is the first effort toward identifying stress from non-wearable devices in general and facial expressions in particular. The current study is inspired by our initial effort to use facial expressions to study satisfaction [1]. The dataset is the first of its kind, where facial features and physiological features are combined for stress measurement.

Project - Deliverables

1 Mathematical modelling
2 Identifying and pre-processing datasets along with setting up experimental protocols.
3 Implementation of graph-embedded subspace learning method, initial experiments, analysis and comparison with other similar methods.
4 Efficient implementation and visualization of data in shared subspace.
5 Verification, finalizing and publishing results.

Project - Benefits to IAB

The developed methods will be useful for stress-detection, anomaly detection, abnormal behavior/event detection, self-monitoring (wellbeing). Companies actively developing empathic building technologies can exploit this research's output to optimize the different models distributed throughout the empathic building.

Project - Documents

projects/year10/10a.002.tau_wp4.txt · Last modified: 2022/10/03 15:01 by sally.johnson