| Bangert T and Izquierdo E (2016), "Defining the Neural Code", In Neural Information Processing Systems (NIPS 2016), 30th Annual Conference on. Barcelona, Catalonia, 5-10 December 2016. Exhibit of Rejected Papers. Barcelona, Catalonia, December, 2016. , pp. 1-13. NIPS.
[Abstract] [BibTeX] [DOI] [URL]
|
| Abstract: While the cellular physiology of neurons has been studied in great detail and is reasonably well defined, the underlying nature of their activity is far from understood and continues to be much debated. In the same way that Boolean logic and theoretical Von Neumann machines have provided the theoretical basis for computation and helped stimulate the development of practical computing, we suggest that it may be helpful to study the nature of neural function more abstractly; specifically, without taking into consideration the inherent complexity of the underlying cellular mechanics. It is commonly assumed by those studying the functionality of complex natural neural systems that their activity of is equivalent to computation. It is asserted (by Church's thesis) that the underlying principles of computation are fundamentally universal and equivalent, and therefore that everything a neural system (no matter how complex) is able to do can be done by any other system capable of general computation. The way computation is done by neural systems might well be very different from well-known models of computation (such as Von Neumann machines) but by Church's thesis they may be considered equivalent. If the core activity of complex neural systems may be considered computation and we accept that the neuron is the atomic unit through which this computation is expressed then an understanding of neural activity must be preceded by a formal understanding of the neuron itself and of the means by which the neuron communicates with other neurons. We suggest that to effectively study and develop a greater understanding of neural function it may be helpful to design minimal but fully functional artificial organisms in which the complex cellular physiology is abstracted and simplified but where the need to function in a complex environment is retained. |
BibTeX:
@inproceedings{bangert2016Defining,
author = {Bangert, Thomas and Izquierdo, Ebroul},
title = {Defining the Neural Code},
booktitle = {Neural Information Processing Systems (NIPS 2016), 30th Annual Conference on. Barcelona, Catalonia, 5-10 December 2016. Exhibit of Rejected Papers},
publisher = {NIPS},
year = {2016},
pages = {1--13},
note = {google scholar entry: Neural Information Processing Systems (NIPS 2016), 30th Annual Conference on. Barcelona, Catalonia, 5-10 December 2016.},
url = {http://mmv.eecs.qmul.ac.uk/TB/mmv/pdf/DefiningNeurons.pdf},
doi = {10.13140/RG.2.2.28471.73121}
}
|
| Bangert T and Izquierdo E (2014), "A Unifying Colour Model for Natural Visual Systems", April, 2014.
[Abstract] [BibTeX] [URL]
|
| Abstract: Although the ancestors of mammals had the same visual sensor arrangement as the ancestors of birds, colour vision in most mammals is very limited. Humans belong to a small subset of mammals that have re-developed some of the colour ability lost by their immediate ancestors. They have done this by diverging one of the two colour sensors available to mammals into a third semi-independent colour sensor. The most common sensor arrangement for animals with complex visual systems is, however, four colour sensors. Animals such as birds not only have four fully independent colour sensors but each sensor has a coloured oil droplet that acts to spectrally restrict the light that falls onto the sensor. Because of this difference in the sensor arrangement, it has been suggested that it is difficult (or ``impossible'') for us to know the perception of colour in animals such as birds. We propose a general colour model that aims to unify our understanding of colour vision in a wide variety of organisms, thereby allowing us to have some insight into how animals such as birds perceive colour. |
BibTeX:
@techreport{bangert2016Unifying,
author = {Bangert, Thomas and Izquierdo, Ebroul},
title = {A Unifying Colour Model for Natural Visual Systems},
year = {2014},
note = {sumbitted initally to Vision Research},
url = {http://mmv.eecs.qmul.ac.uk/TB/mmv/pdf/bangert2014_aUnifyingColourModel.pdf}
}
|
| Bangert T (2012), "Color: an algorithmic approach". Thesis at: Queen Mary University of London., October, 2012. , pp. 1-93.
[Abstract] [BibTeX] [URL]
|
| Abstract: The fundamentals of colour vision were set out in the mid-19th century but have been split between the empirical observation that the underlying hardware responsible for vision was based upon three classes of physical sensors and the perceptual finding that colour consisted of variations of four underlying indivisible primaries, organized into two opponent pairs (blue-yellow and red-green). One of the great advances in the understanding of colour vision was developing an understanding of the mechanism of opponency that makes up the first layer of the neural circuitry that resides directly behind the sensor array of the human visual system. Two opponent colour channels were found, precisely as predicted by the study of perception. Despite the fact that the neural processing circuitry of the visual sensor array consists of only two or three layers of neurons, little further progress has been made to decipher the functionality of subsequent layers. As a result there is little agreement on the nature of the information that is produced by the neural systems that lie directly behind the sensors (at the front of the brain) which is sent to the visual system at the rear of the brain. In this thesis it is proposed that the failure to understand the nature of this information stems mainly from two factors: (1) a need to compensate for an inherent deficiency in the sensor array specific to our evolutionary history (2) the success of the paradigm under which colour is a property of perception rather than information structured by underlying function. In this thesis a paradigm of colour as functional information of an artificial computational visual system is proposed, a simplified artificial colour sensor processing system is presented and parallels are drawn between how this system processes information and how the human visual system is known to process information. It is suggested that understanding the computational requirements of functional colour processing might be helpful in understanding the complex functionality that resides directly behind the sensor array of the human visual system. |
BibTeX:
@mastersthesis{bangert2012color,
author = {Bangert, Thomas},
editor = {Izquierdo, Ebroul},
title = {Color: an algorithmic approach},
school = {Queen Mary University of London},
year = {2012},
pages = {1--93},
url = {http://mmv.eecs.qmul.ac.uk/TB/mmv/pdf/Theses/ThomasBangert(tb300)_MScThesis.pdf}
}
|
| Bangert T (2012), "Color: an algorithmic approach". September, 2012.
[BibTeX] [URL]
|
BibTeX:
@misc{bangert2012color_Presentation,
author = {Bangert, Thomas},
title = {Color: an algorithmic approach},
school = {Queen Mary University of London},
year = {2012},
note = {Presentatation given for MSc Viva and internally to research group},
url = {http://www.eecs.qmul.ac.uk/~tb300/pub/ColourVision.pptx}
}
|
| Bangert T (2008), "TriangleVision: A Toy Visual System", In Artificial Neural Networks (ICANN 2008), 18th International Conference. Prague, Czech Republic, September 3-6, 2008. Proceedings, Part I.. Prague, Czech Republic, September, 2008. Vol. 5163, pp. 937-950. Springer.
[Abstract] [BibTeX] [DOI] [URL]
|
| Abstract: This paper presents a simple but fully functioning and complete artificial visual system. The triangle is the simplest object of perception and therefore the simplest visual system is one which sees only triangles. The system presented is complete in the sense that it will see any triangle presented to it as visual input in bitmap form, even triangles with illusory contours that can only be detected by inference. |
BibTeX:
@inproceedings{bangert2008trianglevision,
author = {Bangert, Thomas},
editor = {Kůrková, Vra and Neruda, Roman and Koutník, Jan},
title = {TriangleVision: A Toy Visual System},
booktitle = {Artificial Neural Networks (ICANN 2008), 18th International Conference. Prague, Czech Republic, September 3-6, 2008. Proceedings, Part I.},
publisher = {Springer},
year = {2008},
volume = {5163},
pages = {937--950},
note = {google scholar entry: Artificial Neural Networks (ICANN 2008), 18th International Conference. Prague, Czech Republic, 3-6 September 2008.},
url = {http://mmv.eecs.qmul.ac.uk/TB/mmv/pdf/TV.pdf},
doi = {10.1007/978-3-540-87536-9_96}
}
|