Conjunctions in biological neural architectures for visual pose estimation / Tom´As Maul
The current thesis is concerned with how biological systems solve the computational problem of visual pose estimation. Four levels of analysis are traversed in order of decreasing abstraction: computational, algorithmic, implementational and formational. As each level is traversed, a biological plau...
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2006
|
| Subjects: | |
| Online Access: | http://studentsrepo.um.edu.my/10973/ http://studentsrepo.um.edu.my/10973/1/Tomas_Maul.pdf |
| _version_ | 1848774263729815552 |
|---|---|
| author | Tom´As, Maul |
| author_facet | Tom´As, Maul |
| author_sort | Tom´As, Maul |
| building | UM Research Repository |
| collection | Online Access |
| description | The current thesis is concerned with how biological systems solve the computational problem of visual pose estimation. Four levels of analysis are traversed in order of decreasing abstraction: computational, algorithmic, implementational and formational. As each level is traversed, a biological plausibility argument is gradually strengthened. At the algorithmic level, several approaches for solving the pose estimation problem are compared in terms of their neural implementability. A highly parallel approach based on simple inter-map conjunctions, or correspondence distributions, is chosen and tested on synthetic and real patterns. The accuracy and robustness of the approach are demonstrated in relation to various critical environmental factors. At the implementational level, the algorithm is translated into various artificial neural architectures. Several maximum value networks are investigated and compared in this context. Combinatorial issues regarding the numbers of nodes and connections are analyzed. The analyses suggest that the architectures can satisfy biological constraints. The spatial arrangement of nodes in different architectures is optimized via an elastic network, with the goal of minimizing the total wiring length between nodes, revealing novel and interesting design principles, some of which correlate with several aspects of biological neural maps. Other revealing links to biological findings are discussed, such as the computation of conjunctions at the level of dendritic branches. Following this, at the formational level, various local mechanisms are investigated in the context of the biological development of the proposed neural architectures. It is shown that simple local rules, together with visual experience, such as that provided by dynamic images, are sufficient for the development of the neural architectures. The generalization of inter-map conjunctions is discussed in the context of other visual functions and sensory modalities. Some pointers towards methodologies for uncovering direct evidence of inter-map conjunctions are also provided. The general hypothesis supported by the thesis states that at least some biological neural systems are likely to be using inter and intra-map conjunctions for efficiently solving computational problems such as visual pose estimation.
|
| first_indexed | 2025-11-14T13:55:32Z |
| format | Thesis |
| id | um-10973 |
| institution | University Malaya |
| institution_category | Local University |
| last_indexed | 2025-11-14T13:55:32Z |
| publishDate | 2006 |
| recordtype | eprints |
| repository_type | Digital Repository |
| spelling | um-109732020-02-03T17:10:13Z Conjunctions in biological neural architectures for visual pose estimation / Tom´As Maul Tom´As, Maul QA75 Electronic computers. Computer science The current thesis is concerned with how biological systems solve the computational problem of visual pose estimation. Four levels of analysis are traversed in order of decreasing abstraction: computational, algorithmic, implementational and formational. As each level is traversed, a biological plausibility argument is gradually strengthened. At the algorithmic level, several approaches for solving the pose estimation problem are compared in terms of their neural implementability. A highly parallel approach based on simple inter-map conjunctions, or correspondence distributions, is chosen and tested on synthetic and real patterns. The accuracy and robustness of the approach are demonstrated in relation to various critical environmental factors. At the implementational level, the algorithm is translated into various artificial neural architectures. Several maximum value networks are investigated and compared in this context. Combinatorial issues regarding the numbers of nodes and connections are analyzed. The analyses suggest that the architectures can satisfy biological constraints. The spatial arrangement of nodes in different architectures is optimized via an elastic network, with the goal of minimizing the total wiring length between nodes, revealing novel and interesting design principles, some of which correlate with several aspects of biological neural maps. Other revealing links to biological findings are discussed, such as the computation of conjunctions at the level of dendritic branches. Following this, at the formational level, various local mechanisms are investigated in the context of the biological development of the proposed neural architectures. It is shown that simple local rules, together with visual experience, such as that provided by dynamic images, are sufficient for the development of the neural architectures. The generalization of inter-map conjunctions is discussed in the context of other visual functions and sensory modalities. Some pointers towards methodologies for uncovering direct evidence of inter-map conjunctions are also provided. The general hypothesis supported by the thesis states that at least some biological neural systems are likely to be using inter and intra-map conjunctions for efficiently solving computational problems such as visual pose estimation. 2006 Thesis NonPeerReviewed application/pdf http://studentsrepo.um.edu.my/10973/1/Tomas_Maul.pdf Tom´As, Maul (2006) Conjunctions in biological neural architectures for visual pose estimation / Tom´As Maul. PhD thesis, University of Malaya. http://studentsrepo.um.edu.my/10973/ |
| spellingShingle | QA75 Electronic computers. Computer science Tom´As, Maul Conjunctions in biological neural architectures for visual pose estimation / Tom´As Maul |
| title | Conjunctions in biological neural architectures for visual pose estimation / Tom´As Maul |
| title_full | Conjunctions in biological neural architectures for visual pose estimation / Tom´As Maul |
| title_fullStr | Conjunctions in biological neural architectures for visual pose estimation / Tom´As Maul |
| title_full_unstemmed | Conjunctions in biological neural architectures for visual pose estimation / Tom´As Maul |
| title_short | Conjunctions in biological neural architectures for visual pose estimation / Tom´As Maul |
| title_sort | conjunctions in biological neural architectures for visual pose estimation / tom´as maul |
| topic | QA75 Electronic computers. Computer science |
| url | http://studentsrepo.um.edu.my/10973/ http://studentsrepo.um.edu.my/10973/1/Tomas_Maul.pdf |