2013/08/10 by Jayanta Dutta, Jayanta K. Dutta, Bonny Banerjee +2
Biochemistry, Genetics and Molecular Biology · Computer Science · Engineering · Mathematics · Neuroscience · #Artificial Intelligence (cs.AI) #Artificial intelligence #Artificial neural network #Cluster analysis #Competitive learning #Computer Vision and Pattern Recognition (cs.CV) #Computer science #Deep learning #FOS: Biological sciences #FOS: Computer and information sciences #Hebbian theory #I.2 #I.4 #I.5 #Image Processing Techniques and Applications #Invariant (physics) #Layer (electronics) #Learning rule #Machine Learning (cs.LG) #Mathematics #Neural Networks and Applications #Neural and Evolutionary Computing (cs.NE) #Neural dynamics and brain function #Neurons and Cognition (q-bio.NC) #Neuroscience #Pattern recognition (psychology) #Pooling #Receptive field #Simple cell #Unsupervised learning #Visual cortex #cs.AI #cs.CV #cs.LG #cs.NE #q-bio.NC
paper · pdf · doi:10.48550/arxiv.1308.2350
published in arXiv (Cornell University) (Cornell University)
arxiv created 2013/08/10 · openalex publication_date 2013/08/10 · arxiv updated 2020/12/14 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/06
Learning features invariant to arbitrary transformations in the data is a\nrequirement for any recognition system, biological or artificial. It is now\nwidely accepted that simple cells in the primary visual cortex respond to\nfeatures while the complex cells respond to features invariant to different\ntransformations. We present a novel two-layered feedforward neural model that\nlearns features in the first layer by spatial spherical clustering and\ninvariance to transformations in the second layer by temporal spherical\nclustering. Learning occurs in an online and unsupervised manner following the\nHebbian rule. When exposed to natural videos acquired by a camera mounted on a\ncat's head, the first and second layer neurons in our model develop simple and\ncomplex cell-like receptive field properties. The model can predict by learning\nlateral connections among the first layer neurons. A topographic map to their\nspatial features emerges by exponentially decaying the flow of activation with\ndistance from one neuron to another in the first layer that fire in close\ntemporal proximity, thereby minimizing the pooling length in an online manner\nsimultaneously with feature learning.\n