 Frame synchronization in space time block coded transmit antenna diversity for WCDMA |
| These problems are resolved by a circuit is designed with a correction circuit coupled to receive a ... |
|
| Composite code match filters |
| The invention in one aspect is a receiver stage of a mobile terminal, such as a cellular phone and ... |
|
| Spread spectrum receive apparatus |
| OF THE INVENTION The embodiments of the present invention will be described below with reference ... |
|
| Multiple channel quadrature communication system and method |
| Accordingly, an object of the present invention is to eliminate the need for separate modulation (... |
|
| Method and apparatus for assigning codes |
| According to one aspect of the present invention, a code assigning apparatus used in a radio ... |
|
| Method for using a non-orthogonal pilot signal with data channel interference cancellation |
| The present invention is directed to a method for encoding/decoding data channels in a system ... |
|
| Microcell load measurement using feedback control |
| OF THE PRESENT INVENTION The present invention relates to a method and system for dynamically ... |
|
|
Neural network apparatus
| Details |
Inventors: Togawa, Fumio; Ueda, Toru; Aramaki, Takashi; Ishizuka, Yasushi;
Assignee: Sharp Kabushiki Kaisha (Osaka, JP)
Primary Examiner: Boudreau; Leo H.
Assistant Examiner:
Attorney, Agent or Firm: Morrison & Foerster
When performing learning for a neural network, a plurality of learning vectors which belong to an arbitrary category are used, and self-organization learning in the category is carried out. As a result, the plurality of learning vectors which belong to the category are automatically clustered, and the contents of weight vectors in the neural network are set to representative vectors which exhibit common features of the learning vectors of each cluster. Then, teacher-supervised learning is carried out for the neural network, using the thus set contents of the weight vectors as initial values thereof. In the learning process, an initial value of each weight vector is set to the representative vector of each cluster obtained by clustering. Therefore, the number of calculations required until the teacher-supervised learning is converged is greatly reduced. |
|
DETAILED DESCRIPTION The neural network apparatus of this invention, which overcomes the above-discussed and numerous other disadvantages and deficiencies of the prior art, comprises self-organization learning means for performing self-organization learning in a category to which a plurality of weight vectors are allocated and a plurality of learning vectors belong, said self-organization learning means having modification means for modifying the contents of a predetermined number of said weight vectors which are in the vicinity of one of said learning vectors, toward said one learning vector, said apparatus further comprising teacher-supervised learning means for performing teacher-supervised learning against said weight vectors, using said modified contents as an initial value.
In the above configuration, said predetermined number may be one, or alternatively two or more.
Preferably, said self-organization learning means performs self-organization learning in all of categories to which said plurality of learning vectors belong.
In preferred embodiments, said apparatus comprises a Kohonen type neural network.
The apparatus may comprise: output nodes allocated to said category; and an input node through which said learning vectors are input.
In preferred embodiments, said apparatus comprises a perceptron type neural network.
The apparatus may comprise: intermediate layer nodes allocated to said category; and input layer nodes through which said learning vectors are input.
The method according to the invention comprises the steps of: performing self-organization learning in a category to which a plurality of weight vectors are allocated and a plurality of learning vectors belong, while modifying the contents of a predetermined number of said weight vectors which are in the vicinity of one of said learning vectors, toward said one learning vector; and performing teacher-supervised learning against said weight vectors, using said modified contents as an initial value.
Both the self-organization learning step and teacher-supervised learning step may be conducted substantially simultaneously
|
|
Fault Detection 
