摘要:
An improved iterative learning machine having a plurality of multi-input/single-output signal processing units connected in a hierarchical structure includes a weight coefficient change control unit which controls weight change quantities for those multi-input/single-output signal processing units having iteratively reduced errors thereby increasing the learning speed, contrary to conventional learning machines which perform a learning operation in order to minimize a square error of multi-input/single-output signal processing units in the highest hierarchy of the hierarchical structure.
摘要:
A learning machine with multi-input single-output circuits connected in hierachical structure, wherein product-sums of the output signals of input and output signal registers and the weights are obtained by the parallel processing of a plurality of product-sum computing units with a plurality of input and output signal registers being connected in cascades, thereby to render to scale down the circuit of the learning machine with the sigmoid function computing unit being one in number.
摘要:
A recognition apparatus is provided with a plurality of recognition units organized in a multilayered hierarchical structure. Each of the recognition units includes a signal input section, a quantizer for performing a quantization according to a signal inputted from the signal input section, and a path selecting section for performing a selection of paths according to an output from the quantizer. The path selecting section includes a path input section having at least one path input terminal, a path output section having at least one path output terminal, a load distribution selecting section for selecting a load distribution, and a load setting section for changing the strength of connection between the path input terminal and the path output terminal according to the output of the quantizer by the use of the load distribution selected by the load distribution selecting section.
摘要:
An iterative learning machine uses, as a direction of changing iterative weight, a conjugate gradient direction in place of the conventional steepest descent direction, thereby saving time. Learning rates are set dynamically. Error calculation for plural learning rates, with respect to a certain weight changing direction, are accomplished by storing a product-sum of the input signals and weights in a hidden layer and a product-sum of the input signals and the weight changing direction in the hidden layer. When the learning falls into a non-effective state where further iteration does not effectively reduce an error, the weights are adjusted in order to restart the learning.
摘要:
A learning machine has plural multiple-input single-output signal processing circuits connected in a hierarchical structure. The learning machine sets a threshold value, which is a evaluation standard for change in weight coefficients, high during the early part of the learning process and enables rough learning to progress without changing the weight coefficients for those multiple-input single-output signal processing circuits for which errors are sufficiently small. On the other hand, the learning machine gradually reduces the threshold value as learning progresses and advances learning by a non-linear optimization method (including a conjugate gradient method, a linear search method, or a combination of conjugate gradient and linear search methods) during the later part of the learning process, and thereby improves the learning speed.
摘要:
The invention is a learning type waveform recognizer that has a sequential signal extractor that extracts changes in the values of an input waveform signal at each point on the transition axis, a learning and recognition section that learns the relations between the changes extracted from the input signal and classified waveform signals and recognizes the degree of matching with the classified waveform at each point on the transition axis based on the learned relations, and an integrated recognition section that integrates the recognized results at points in an interval and performs an overall judgment by identifying the waveform of the input waveform signal.
摘要:
A recognizing and judging apparatus for a learning and recognizing processing to be effectively performed in a short period of time, the apparatus including a plurality of recognition units in a multi-layered hierarchical network structure with one or more path output terminals of the recognition units of an upper layer being connected with one or more path input terminals of the recognition units of a lower layer. The recognition unit comprises a vector signal input structure for inputting a plurality of input signals for showing characteristics of an object to be recognized, a path signal transmitting information computing structure for converting an input signal vector from the vector signal input structure into a path signal transmitting information and a path signal transmitting structure having one or more path input terminals for inputting a path signal and one or more path output terminals for outputting the path signal to transmit the path signal from the path input terminals to the path output terminals according to the path signal transmitting information.
摘要:
There is disclosed a character recognition machine adapted to recognize Japanese characters such as kanjis and kanas. The machine comprises a character string storage portion, a character extraction portion, a character recognition portion, and a language processing portion. A character string to be recognized is stored as an image in the storage portion. The character extraction portion comprises a network consisting a plurality of interconnected operators each of which has numerous inputs and outputs. An evaluation function which assumes its minimum value when a character extraction produces the best results is calculated by the operators simultaneously so as to minimize the value of the function. The character recognition portion calculates degrees of similarity of a character pattern to various character categories, the character pattern being applied from the character extraction portion. The language processing portion receives these degrees of similarity and selects a character category which seems to provide the most correct combination of characters in terms of vocabulary and grammar.
摘要:
There is disclosed a character recognition machine adapted to recognize Japanese characters such as kanjis and kanas. The machine comprises a character string storage portion, a character extraction portion, a character recognition portion, and a language processing portion. A character string to be recognized is stored as an image in the storage portion. The character extraction portion comprises a network consisting a plurality of interconnected operators each of which has numerous inputs and outputs. An evaluation function which assumes its minimum value when a character extraction produces the best results is calculated by the operators simultaneously so as to minimize the value of the function. The character recognition portion calculates degrees of similarity of a character pattern to various character categories, the character pattern being applied from the character extraction portion. The language processing portion receives these degrees of similarity and selects a character category which seems to provide the must correct combination of characters in terms of vocabulary and grammar.
摘要:
An input/display integrated information processing device has: image display circuit for displaying document and image information; image input circuit through which information of an image such as a picture, a photograph, documents drawn or written on a sheet presented by the user can be an input; a visual coincidence structure for integrally combining a display screen of the image display circuit with an input screen of the image input circuit so as to establish visual coincidence of screens as viewed from the user; contact information detection circuit for, when a finger of the user, the sheet presented by the user, or the like makes contact with the contact information detection circuit, detecting information of a contact position, the contact information detection circuit being disposed on the whole of the input screen of the image input circuit or the display screen of the image display circuit; image extraction circuit for automatically extracting only a required portion from image information input through the image input circuit; image information storage circuit for storing the image information; image synthesize circuit for combining the image with an image which is already displayed and stored; and image information control circuit for controlling input/output operations of the image information.