PyLissom : a tool for modeling computational maps of the visual cortex in PyTorch
The fields of neuroscience and artificial intelligence (AI) have a long and intertwined history. Better understanding of biological brains could play a vital role in building intelligent machines. However, communication and collaboration between the two fields has become less commonplace [79...
Guardado en:
| Autor principal: | |
|---|---|
| Otros Autores: | |
| Formato: | Tesis de grado publishedVersion |
| Lenguaje: | Inglés |
| Publicado: |
Universidad de Buenos Aires. Facultad de Ciencias Exactas y Naturales
2019
|
| Materias: | |
| Acceso en línea: | https://hdl.handle.net/20.500.12110/seminario_nCOM000623_Barijhof https://repositoriouba.sisbi.uba.ar/gsdl/cgi-bin/library.cgi?a=d&c=aextesisg&d=seminario_nCOM000623_Barijhof_oai |
| Aporte de: |
| Sumario: | The fields of neuroscience and artificial intelligence (AI) have a long and intertwined history. Better understanding of biological brains could play a vital role in building intelligent machines. However, communication and collaboration between the two fields has become less commonplace [79]. Computational tools that integrate approaches to neuroscience and machine learning, in accessible and documented form, are very scarce in the literature. The availability of these tools could be fruitful for the interaction between the neuroscience and machine learning communities, and the emergence of new ideas and collaborations. Self-organized neural networks with lateral connections (LISSOM) have been proposed in the literature as a computational model of maps in the visual cortex in primates [84]. These networks were implemented by a group of the University of Edinburgh and the University of Texas in a computational system called Topographica [71]. The use case of the Topographica software has been the neuroscience community. The Topographica software has been used successfully by some researchers to validate computational models in neuroscience. However, due its design, Topographica use has been restricted to neuroscience, and it is very difficult to extend and adapt its code for machine learning uses. In this thesis, LISSOM networks are implemented with a hybrid use case for the machine learning and the neuroscience communities. The software developed in this work, named PyLissom, allows on one hand to build hierarchical models of the visual system, and on the other hand, be used for machine learning applications, since it can combine LISSOM neural networks with other type of artificial neural networks. PyLissom has a modern software design, is implemented in PyTorch and can use GPU optimization. |
|---|