Summary
The visual world around us is a source of rich semantic information that guides our higher-level cognitive processes and actions. To tap into this resource, the brain?s visual system engages in complex, intertwined computations to actively sample, extract, and integrate information across space and time. Surprisingly however, the integrative nature of vision hardly plays a role in the way we approach it in experimentation and computational modelling. Instead, higher-level vision is commonly treated as a largely bottom-up categorization process.
TIME proposes a new approach. It will allow us to study vision in a more natural setting and as a process that is (a) geared towards semantic understanding instead of label-based categorisation, (b) naturally intertwined with active information sampling and (c) expanding across multiple timeframes, including network dynamics that unfold within and across eye fixations. This will be accomplished by an ambitious, three-step work program that combines cutting-edge non-invasive human brain imaging performed while participants visually explore tens of thousands of rich human-annotated natural scenes, the development of novel multivariate analysis techniques, and large-scale computational modelling using a new bio-inspired deep learning framework for active vision that closes the sensory-motor loop. Using this interdisciplinary approach, TIME will establish, for the first time, when, where, and how visual semantic understanding emerges in the brain as it actively samples and integrates information from the world in a continuously updating and dynamic decision process. These ground-breaking developments both in experimentation and deep neural network modelling build towards a fundamental paradigm shift in how we study, model, and understand vision, yielding new insights into its complex neural processes operating in more natural, ecologically valid conditions, as well as a closer alignment between biological and synthetic vision.
TIME proposes a new approach. It will allow us to study vision in a more natural setting and as a process that is (a) geared towards semantic understanding instead of label-based categorisation, (b) naturally intertwined with active information sampling and (c) expanding across multiple timeframes, including network dynamics that unfold within and across eye fixations. This will be accomplished by an ambitious, three-step work program that combines cutting-edge non-invasive human brain imaging performed while participants visually explore tens of thousands of rich human-annotated natural scenes, the development of novel multivariate analysis techniques, and large-scale computational modelling using a new bio-inspired deep learning framework for active vision that closes the sensory-motor loop. Using this interdisciplinary approach, TIME will establish, for the first time, when, where, and how visual semantic understanding emerges in the brain as it actively samples and integrates information from the world in a continuously updating and dynamic decision process. These ground-breaking developments both in experimentation and deep neural network modelling build towards a fundamental paradigm shift in how we study, model, and understand vision, yielding new insights into its complex neural processes operating in more natural, ecologically valid conditions, as well as a closer alignment between biological and synthetic vision.
Unfold all
/
Fold all
More information & hyperlinks
| Web resources: | https://cordis.europa.eu/project/id/101039524 |
| Start date: | 01-07-2022 |
| End date: | 30-06-2027 |
| Total budget - Public funding: | 1 499 455,00 Euro - 1 499 455,00 Euro |
Cordis data
Original description
The visual world around us is a source of rich semantic information that guides our higher-level cognitive processes and actions. To tap into this resource, the brain?s visual system engages in complex, intertwined computations to actively sample, extract, and integrate information across space and time. Surprisingly however, the integrative nature of vision hardly plays a role in the way we approach it in experimentation and computational modelling. Instead, higher-level vision is commonly treated as a largely bottom-up categorization process.TIME proposes a new approach. It will allow us to study vision in a more natural setting and as a process that is (a) geared towards semantic understanding instead of label-based categorisation, (b) naturally intertwined with active information sampling and (c) expanding across multiple timeframes, including network dynamics that unfold within and across eye fixations. This will be accomplished by an ambitious, three-step work program that combines cutting-edge non-invasive human brain imaging performed while participants visually explore tens of thousands of rich human-annotated natural scenes, the development of novel multivariate analysis techniques, and large-scale computational modelling using a new bio-inspired deep learning framework for active vision that closes the sensory-motor loop. Using this interdisciplinary approach, TIME will establish, for the first time, when, where, and how visual semantic understanding emerges in the brain as it actively samples and integrates information from the world in a continuously updating and dynamic decision process. These ground-breaking developments both in experimentation and deep neural network modelling build towards a fundamental paradigm shift in how we study, model, and understand vision, yielding new insights into its complex neural processes operating in more natural, ecologically valid conditions, as well as a closer alignment between biological and synthetic vision.
Status
SIGNEDCall topic
ERC-2021-STGUpdate Date
09-02-2023
Images
No images available.
Geographical location(s)
