US5420964A
Expired / Lapsed
US
⏱️ 0.0 years remaining
System, for learning an external evaluation standard
Assignee: Fujitsu Ltd
•
Filing: 1992-10-14
•
Granted: 1995-05-30
•
Expires: 2012-10-14
Abstract
Abstract This invention pertains to neural network system for learning an external evaluation standard and for learning the evaluation from the outside for the processing result, in a system capable of internal evaluation of the correspondence between external information and the processing result of its own system for the input information. It purports to learn the external evaluation as the internal evaluation standard of the internal evaluation time. The learning system comprises an internal evaluation unit for evaluating an evaluation input pattern including input information at a first point in time and input information inputted at a point in time for the processing result of its own system for the input information according to the internal evaluation standard at a system execution time; and an evaluation desired pattern memory unit for making the external evaluation correspond with the evaluation input pattern and for memorizing it as an evaluation desired pattern for having the internal evaluation unit learn the external evaluation standard. The system is configured to have an internal evaluation unit to learn the evaluation desired pattern at a learning time of the system. This system is applicable in a robot control system.
Inventors
- 👤 Tamami Sugasaka
- 👤 Kazushige Saga
- 👤 Minoru Sekiguchi
- 👤 Shigemi Nagata
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-31 23:13
CPC Classifications
G; G06; G06N; G06N20/00
Claims Analysis
Total Claims: 13
Claim 1
Independent Claim
214 words
A system for learning an external learning standard, said system having an internal evaluation standard and a function for evaluating a correspondence between external information and external output, produced from processing the external information according to input information of the internal evaluation standard, and for learning an evaluation standard of an external evaluation by using the external output and the internal evaluation standard, said system comprising: neural network processing means for processing an input pattern based on a desired pattern as input information of the internal evaluation standard, independently of the external evaluation and providing an output pattern; input/output pattern corresponding means for comparing the input pattern with the output pattern, and for outputting a corresponding relation as an input/output pattern; evaluation means for judging whether the input/output pattern is correct by using a reinforcement signal as the external evaluation supplied from outside the system and the input/output pattern from the input/output pattern corresponding means, and for outputting the judged input/output pattern when an evaluation result of the judging is correct; and desired pattern table means for storing the judged input/output pattern whose evaluation result outputted from said evaluation means is correct, and for outputting the judged input/output pattern as an evaluation desired pattern at a learning time of said neural network processing means.
Claim 2
Independent Claim
214 words
A system for learning an external learning standard, said system having an internal evaluation standard and a function for evaluating a correspondence between external information and external output, produced from processing the external information according to input information of the internal evaluation standard, and for learning an evaluation standard of an external evaluation by using the external output and the internal evaluation standard, said system comprising: neural network processing means for processing an input pattern based on a desired pattern as input information of the internal evaluation standard, independently of the external evaluation and providing an output pattern; input/output pattern corresponding means for comparing the input pattern with the output pattern, and for outputting a corresponding relation as an input/output pattern; evaluation means for judging whether the input/output pattern is correct by using a reinforcement signal as the external evaluation supplied from outside the system and the input/output pattern from the input/output pattern corresponding means, and for outputting the judged input/output pattern when an evaluation result of the judging is correct; and desired pattern table means for storing the judged input/output pattern whose evaluation result outputted from said evaluation means is correct, and for outputting the judged input/output pattern as an evaluation desired pattern at a learning time of said neural network processing means.
Claim 3
Dependent Claim
25 words
The system for learning the external evaluation standard according to claim 1 further comprising means for adding noise to said output pattern of said system.
Claim 4
Dependent Claim
96 words
The system for learning the external evaluation standard according to claim 1, further comprising: input means for receiving the external information and for converting the external information to the input pattern; a memory for storing past external information and for outputting the past external information to said input means; noise generating means for generating noise; adder means for adding the output pattern from said neural network processing means to the generated noise, producing a noise added output pattern; and output means for converting the noise added output pattern from said adder means to the external output.
Claim 5
Dependent Claim
224 words
The system for learning the external evaluation standard according to claim 3, wherein said evaluation means comprises; a short term memory for temporarily storing the input/output pattern outputted from said input/output pattern corresponding means; evaluation input means for creating an evaluation input pattern including a current input pattern and a past input pattern by using a current input/output pattern and a past input/output pattern outputted from said short term memory means; neural network internal evaluation means for receiving the evaluation input pattern including input information at a first time and input information at a second time after the first time, and for evaluating the evaluation input pattern in accordance with the internal evaluation standard and for outputting an evaluation signal indicating whether the evaluating is correct; evaluation output means for outputting the past input/output pattern stored in said short term memory means to said desired pattern table means when the evaluation signal indicates a correct evaluation outputted from said neural network internal evaluation means; and evaluation desired pattern memory means for receiving the reinforcement signal as the external evaluation for the external output from said output means, for making the reinforcement signal correspond with the evaluation input pattern, and for storing the evaluation input pattern as the evaluation desired pattern for having said internal evaluation means learn the evaluation standard of the external evaluation.
Claim 6
Dependent Claim
82 words
The system for learning the external evaluation standard according to claim 4, wherein said neural network internal evaluation means comprises means for receiving the evaluation input pattern including the current input pattern and the past input pattern, obtained as a result of said input means converting the external information at a current time and past external information stored in said memory, and the output pattern as an output from said adder means at a same past time as the past input pattern.
Claim 7
Dependent Claim
88 words
The system for learning the external evaluation standard according to claim 4, wherein said neural network internal evaluation means comprises means for receiving the evaluation input pattern including the current input pattern and the past input pattern, the current and past input pattern obtained as a result of said input means converting the external information at a current time and the external information converted at a past time and stored in said memory, respectively, and the output pattern from said adder means corresponding to the past input pattern.
Claim 8
Dependent Claim
189 words
The system for learning the external evaluation standard according to claim 3, wherein: said input means includes means for converting the inputted external information to the input pattern and for sending the input pattern to said input/output pattern corresponding means; said evaluation means includes means for evaluating the input/output pattern according to the internal evaluation standard, and for sending the input/output pattern to said desired pattern table means when the evaluation result is correct; said output means includes means for converting the output pattern outputted from said adder means to the external output and for providing the external output outside of the system; and the system further comprises means for changing an external environment using the external output for judging whether the external output should be terminated due to an environmental change, for repeating actions, in an after-conversion of the external information to the input pattern, when the external output should not be terminated and for sending the desired pattern, stored in said desired pattern table means, to said neural network processing means when the external output should be terminated, so that learning using a back propagation method occurs.
Claim 9
Dependent Claim
230 words
The system for learning the external evaluation standard according to claim 7, further comprising: short term memory means for storing the sent input/output pattern at a third time of the input/output pattern according to the internal evaluation standard by said evaluation means; evaluation input means for determining whether an input of the external information is an initial input, for terminating processing if the external information is the initial input and for creating an evaluation input pattern if the external information is not initial external information; neural network internal evaluation means for evaluating the evaluation input pattern; evaluation output means for sending the input/output pattern in said short term memory means to the desired pattern table means when said evaluation is correct; and evaluation desired pattern memory means for receiving the reinforcement signal, wherein said evaluation input means includes means for providing the evaluation input pattern to the evaluation desired pattern memory means; wherein said evaluation desired pattern memory means includes means for creating a correspondence between the reinforcement signal and the external evaluation with the evaluation input pattern and for storing the correspondence as the evaluation desired pattern; and wherein the evaluation desired pattern stored in said evaluation desired pattern memory means is used by said neural network internal evaluation means to perform a learning using the back propagation method when a terminating condition of the external output is met.
Claim 10
Dependent Claim
63 words
The system for learning the external evaluation standard according to claim 7, wherein: said neural network processing means includes means for receiving the input pattern including a current input pattern and at least one past input pattern obtained as a result of said input means converting the external information at a current time and the external information at at least one past time.
Claim 11
Dependent Claim
63 words
The system for learning the external evaluation standard according to claim 3, wherein: said neural network processing means includes means for receiving the input pattern including a current input pattern and at least one past input pattern obtained as a result of said input means converting the external information at a current time and the external information at at least one past time.
Claim 12
Dependent Claim
50 words
A system for learning the external evaluation standard according to claim 3, connected to a robot, said input means comprising means for receiving the external information which includes outputs from a plurality of visual sensors attached to the robot, wherein the external output indicates direction of movement of the robot.
Claim 13
Dependent Claim
56 words
A system for learning the external evaluation standard according to claim 1, said system being connected to a robot and further comprising an input means comprising means for receiving the external information which includes outputs from a plurality of visual sensors attached to the robot, wherein the external output indicates direction of movement of the robot.
References Cited (Backward Citations)
US4884216A
Cited
US4933871A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1998-05-29 | 1998-11-28 | 1999-05-30 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2002-05-29 | 2002-11-28 | 2003-05-30 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2006-05-29 | 2006-11-28 | 2007-05-30 | $7700.00 | Window Closed / Lapsed |