IMaLeVICS: New VIPLab project
We propose this «Interpretable Machine Learning for Vision, Imaging, and Color Sciences (IMaLeVICS)» project to obtain systems able to solve specific problems in each of the fields and also to analyze what the systems have learned to interpret the knowledge that has been extracted from data. And therefore, to study if this knowledge can be included in the current scientific models in the state of the art for the corresponding problems. This approach has led to interesting results so far.
The main objective of our VIP Lab group is to contribute to the better understanding of how digital images are perceived by the human visual system through building new models based on data and interpreting the knowledge that the models have extracted from the data to contribute to the development of more advanced computational vision models.
The IMaLeVICS team
For the proper development of the project, we incorporate new team members that complement the VIP Lab team in terms of knowledge of more interpretable machine learning models and vision science along with more facilities to acquire new data.
Thus, we count on a multidisciplinary team with knowledge in mathematical modelling, vision, color and imaging sciences, and interpretable artificial intelligence models, and also with access to both data resources as well as laboratories where more data can be obtained if needed.
The main target of the project
A simple search in scopus database shows more than 23.000 machine learning or deep learning publications in the vision, imaging, and color fields, with dozens of conferences taking place around the globe per year about these topics, like the successful LIM conference presented by Prof. Samuel Morillas. However, almost all of these publications are related to deep neural networks that have very low interpretability and, apart from solving the particular problem, they are trained for, are of lesser use to develop the knowledge in the field.
Therefore, the main target of the project is to use interpretable machine learning methods trained from data to provide insight into a series of problems in the vision, imaging, and color sciences. In particular, we aim at targeting the following series of problems where the researchers in the team are already working with other approaches and for which we have extensive amounts of data that are of critical importance for this approach to be successful:
- Determination of relevant color in images
- Processing of endoscopy imaging
- Subjective image and video quality evaluation
- Image classification
- Gaze patterns in art viewing
- Display characterization
We requested this grant from the Generalitat Valenciana in late 2022, and the project will be executed from 2023 to 2025. If you want to know more about it, contact us.