. A new model which can recognize rotated, distorted, scaled, shifted and noised patterns is proposed. The model is constructed based on psychological experiments in a mental rotation. The model has two types of processes: (i) one is a bottom-up p...
A rotation-invariant neocognitron is constructed by extending the neocognitron which can recognize translated, scaled and/or distorted patterns from training ones. In constructing the model two technical methods during the learning, a "threshold-c...
A popular approach to understand our human visual "functions" is performing the computational simlutaions of its linear models created with a focus on input-output relations of cells. However, due to a lot of simulation time for a huge amount of c...
Numerical simulation for the linear model of visual neurons is the most important approach to understand our visual system from computational viewpoints. We attempt to parallelize the time-consuming simulation on a cluster computer system. We achi...
IEICE technical report. Neurocomputing 109(461) 7-12 Mar 2010
It is believed that in primary visual cortex(V1), neurons efficiently encode natural images by the sparse outputs of these neurons. However, such outputs of V1 cells are not sufficiently sparse because natural images have a spatial structure like ...
IEICE technical report. Neurocomputing 109(461) 391-396 Mar 2010
We propose a simple but novel mathematical framework dealing with binocular vision; using complex functions, {left image} + i{right image}. This framework expands the solutions of mathematical problems on visual functions into complex functions. I...
IEICE technical report. Neurocomputing 108(480) 141-146 Mar 2009
A novel model for spatio-temporal receptive fields of V1 simple neurons is proposed by introducing fractional order derivatives. We consider the computational role of V1 simple neurons, and derive physiologically acceptable and theoretically optim...
IEICE technical report. Neurocomputing 108(480) 37-42 Mar 2009
Mathematical model of the brain developed so far have targeted specific brain areas such as, cells, networks and/or phenomena, including microscopic and macroscopic levels. However, in order to understand the whole brain system, integration of suc...
IEICE technical report. Neurocomputing 106(590) 1-6 Mar 2007
A visual model for filling-in at the blind spot is proposed. Standard regularization theory is employed to derive a visual model deductively. First, some problems of the diffusion eqation, which is frequently used for depth perception, are pointed...
IEICE technical report. Neurocomputing 105(658) 55-60 Mar 2006
Neurophysiological experiments have revealed that post-synaptic effect conveyed by long-range horizontal connections (LHC) in V1 can be excitatory or inhibitory, and the effect highly depends on the activity of pre-synaptic neurons (non-linearity ...
IEICE technical report. Neurocomputing 104(760) 1-6 Mar 2005
A visual model for object detection is proposed. The computational principles of active contours and level-set method are utilized to construct the proposed model. A visual model for object detection have been proposed by Satoh (IEICE Tech. Rep. V...
Technical report of IEICE. HIP 104(525) 1-6 Dec 2004
A visual model for object detection is proposed. The dynamics of the proposed model is based on the computational principle of active contours, which is a fundamental method for various image processing. A redefined level set function is utilized ...
IEICE technical report. Neurocomputing 103(734) 85-90 Mar 2004
A computational model for working memory which is controled by a reward system is proposed. The control system of working memory is formulated by use of a temporal difference model which estimates future reward. Through the computational experimen...
IEICE technical report. Neurocomputing 103(734) 91-96 Mar 2004
From the recent physiological data, it is shown that acetylcholine controls the synaptic transmission in the hippocampus and induces the theta rhythm and the rhythm of sleep and waking. It is also suggested that acetylcholine plays a very importan...
IEICE technical report. Neurocomputing 103(733) 61-66 Mar 2004
Multilayer perceptron is widely used as high-performance discrimination circuit. Learned knowledge is expressed as connection weights and output values in hidden and output layer. However it is difficult to gain the comprehensible knowledge from t...
IEICE technical report. Neurocomputing 103(732) 117-122 Mar 2004
In figure-ground reversal perception for ambiguous figures, figure-ground relationship reverses depending on the position and the range of our attention. Our attention also changes depending on figural features. We propose a neural network model t...
Yamazaki Susumu, Ito Makoto, Miyake Shogo, Satoh Shunji
IEICE technical report. Neurocomputing 102(731) 125-130 Mar 2003
We construct a model of a radial-arms maze task for rats in which a higher function of memory is required. We formulate a radial-arms maze task as a reinforcement learning problem. In this model, agent can successfully learn depnding on past memor...
IEICE technical report. Neurocomputing 102(730) 79-84 Mar 2003
Principles of visual grouping and figure-ground relationship about contours of objects have been identified as closure, symmetry, convexity, parallelism, proximity and so on. These principles should be considered in order to construct an abject re...
IEICE technical report. Neurocomputing 102(627) 37-42 Jan 2003
A model for visual attention based on scale-space theory is proposed by employing results given by neurophysiological experiments and cognitive phenomena of humans on the vision. The major parts of the model are (a) a model of early vision, and (b...
MORISHIMA K., SATOH S., SHIMOMURA M., MIYAKE S., ASO H.
IEICE technical report. Neurocomputing 101(736) 95-102 Mar 2002
A modular-type neural network is proposed in order to solve pattern recognition problems. Neurons in the hidden layer convert an input space to a curved space by use of polynominal functions, and a neuron in the output makes discriminant surfaces ...
IEICE technical report. Neurocomputing 100(686) 55-62 Mar 2001
A new recognition model for occluded patterns and texture patterns is proposed. The model includes two types of cells; cells that detect the luminance of input patterns and ones which have Gabor-like receptive fields. Generally, features of unknow...
IEICE technical report. Neurocomputing 100(686) 63-70 Mar 2001
We present a new neural network model for pattern extraction which is able to efficiently recognize patterns allocated in a plane. The model consists of two units, a saccade unit with low-resolution and a gazing unit with high-resolution. Both uni...
IEICE technical report. Neurocomputing 100(686) 109-116 Mar 2001
The method of feature extraction in usual pattern recognition depends heavily on objects and is heuristic. In order to get high performance in pattern recognition, feature extraction is an important process. There are only few studies which discus...
IEICE technical report. Neurocomputing 99(685) 137-144 Mar 2000
Neocognitron is a neural network model which can recognized noised, distorted and shifted patterns, and several extended models based on the neocognitron are proposed. However, the structures of these models become more and more complex in the pro...
IEICE technical report. Neurocomputing 98(674) 231-238 Mar 1999
A new neural network model which can recognize rotated, distorted, scaled, shifted and noised patterns is proposed. The model is constructed based on psychological experiments in a mental rotation. The model has two types of processes. One is a fe...
Satoh S., Kuroiwa J., Aso H., Miyake S., Inawashiro S.
IEICE technical report. Neurocomputing 95(598) 263-270 Mar 1996
The neocognitron is a hierarchical neural network to recognize patterns after self organizing, and it is robust for deformation and/or shifting of patterns for recognition. However, it is difficult to recognize greatly rotated patterns. We propose...
