Visual Commensence for Scene Understanding
Lead Research Organisation:
University of Glasgow
Department Name: College of Medical, Veterinary, Life Sci
Abstract
Abstracts are not currently available in GtR for all funded research. This is normally because the abstract was not required at the time of proposal submission, but may be because it included sensitive information such as personal details.
People |
ORCID iD |
| Philippe Schyns (Principal Investigator) |
Publications
Bjornsdottir RT
(2024)
Social class perception is driven by stereotype-related facial features.
in Journal of experimental psychology. General
Chen C
(2024)
Cultural facial expressions dynamically convey emotion category and intensity information.
in Current biology : CB
Chen C
(2018)
Distinct facial expressions represent pain and pleasure across cultures.
in Proceedings of the National Academy of Sciences of the United States of America
Daube C
(2021)
Grounding deep neural network predictions of human categorization behavior in understandable functional features: The case of face identity.
in Patterns (New York, N.Y.)
Heaven D
(2020)
Why faces don't always tell the truth about feelings.
in Nature
Ince R
(2017)
Measuring Multivariate Redundant Information with Pointwise Common Change in Surprisal
in Entropy
Ince RA
(2021)
Bayesian inference of population prevalence.
in eLife
Jack RE
(2017)
Toward a Social Psychophysics of Face Communication.
in Annual review of psychology
| Description | We have developed a new methodology to achieve a deeper interpretability of deep networks. Specifically, using information theoretic measures, we can now visualize the information that is represented at each layer of a deep network. From this understanding, we can better estimate the information transformation function that are performed across layers. Furthermore, we have using Generational Autoencoders to compare the representations constructed on the hidden layers with those of several other models (i.e. classic ResNet DeepNetwork, an engineered generative model and an ideal observer model. |
| Exploitation Route | Others users of deep networks might use our methodologies to better understand why deep networks fail to generalize--cf. adversarial testing. |
| Sectors | Aerospace Defence and Marine Creative Economy Digital/Communication/Information Technologies (including Software) |
| URL | https://arxiv.org/abs/1811.07807 |