Pattern recognition
AssigneeBritish Telecommunications public limited company
InventorGraham D Tattersall
Life of this patent
Estimating the term…
Abstract
Input values are digitally represented and the bits used to address memories. A composite (e.g. the sum) of the memory outputs is formed, and update control is used during “learning” to modify the memory contents. The digital representation is formed by an encoder which encodes input values using a redundant code, which preferably has a Hamming distance to signal distance relationship which has a relatively steep initial slope and is closer to being monotonic than is the relationship for a simply binary code.
Claims22
A pattern recognition apparatus comprising:
- an input port for receiving a set of input digital signal values representing a pattern of physical occurrences distributed over time or space to be recognized;
- means for encoding the input digital signal values into coded input signal values to form a digital representation of the received set of input digital signal values;
- a plurality of addressable digital signal storage memory devices each having at least one address line for addressing locations therein, the address lines being connected to receive bits from said digital representation;
- and
- means connected to receive digital data signals from said memory devices for producing a composite digital output signal dependent on data signals read from the addressed locations of the memory devices;
- wherein the means for encoding employs a redundant code having a normalized coded input signal value Hamming distance to normalized input signal value distance relationship with a mean slope greater than 1 for input signal value distances up to a predetermined level of at least 10% of the maximum input signal value distance.
A pattern recognition apparatus comprising:
- an input port for receiving a set of input digital signal values representing a pattern of physical occurrences distributed over time or space to be recognized;
- means for encoding the input digital signal values to form a digital representation of the received set of input digital signal values;
- a plurality of addressable digital signal storage memory devices each having at least one address line for addressing locations therein, the address lines being connected to receive bits from said digital representation;
- and
- means connected to receive digital data signals from said memory devices for producing a composite digital output signal dependent on data signals read from the addressed locations of the memory devices;
- wherein the means for encoding uses a code employing a number of bits which is:
- i) greater than that which a binary code would require to represent an equal number of signal levels;
- and
- ii) less than that which a bar code would require to represent an equal number of signal levels;
- and being selected to have a higher monotonicity than would such a binary code.
A method of training a pattern recognition device which is arranged to receive a set of digital input signal values representing a pattern of physical occurrences distributed over time or space to be recognized comprising the steps of:
- encoding said digital input signal values into coded input signal values to form a digital signal representation of said pattern using a redundant code which has a normalized Hamming distance to normalized input signal value distance relationship with a mean slope greater than 1 for input signal value distances up to a predetermined level of at least 10% of the maximum input signal value distance;
- and
- using the coded input signal values to recognize the pattern of physical occurrences.
A method of training a pattern recognition device which is arranged to receive a set of digital input signal values representing a pattern of physical occurrences distributed over time or space to be recognized comprising the steps of:
- encoding said digital input signal values using a code employing a number of bits which is:
- i) greater than that which a binary code would require to represent an equal number of signal levels;
- and
- ii) less than that which a bar code would require to represent an equal number of signal levels;
- and being selected to have a higher monotonicity than would such a binary code;
- and
- using the coded input signal values to recognize the pattern of physical occurrences.
A pattern recognition apparatus comprising:
- an input port for receiving a set of digital input signal values representing a pattern of physical occurrences distributed over time or space to be recognized;
- means for encoding the digital input signal values into coded input signal values to form a digital representation of the received set of input digital signal values;
- a plurality of addressable digital signal storage memory devices each having at least one address line for addressing locations therein, the address lines being connected to receive bits from said digital representation;
- and
- means connected to receive digital data signals from said memory devices for producing a composite digital output signal dependent on data signals read from the addressed locations of the memory devices;
- wherein the means for encoding employs a redundant code which has a normalized coded input signal value Hamming distance to normalized input signal value distance relationship having a mean slope greater than 2 for input signal value distances up to a predetermined level of at least 20% of the maximum input signal value distance.
A pattern recognition apparatus comprising:
- an input port for receiving a set of input digital signal values representing a pattern of physical occurrences distributed over time or space to be recognized;
- means for encoding the input digital signal values into coded input signal values to form a digital representation of the received set of input digital signal values;
- a plurality of addressable digital signal storage memory devices each having at least one address line for addressing locations therein, the address lines being connected to receive bits from said digital representation;
- and
- means connected to receive digital data signals from said memory devices for producing a composite digital output signal dependent on data signals read from the addressed locations of the memory devices;
- wherein the means for encoding uses a code employing a number of bits which is:
- i) greater than that which a binary code would require to represent an equal number of signal levels;
- and
- ii) less than that which a bar code would require to represent an equal number of signal levels;
- and being selected to have a monotonicity at least 1.5 times that of a binary code.
A method of training a pattern recognition device which is arranged to receive a set of digital input signal values representing a pattern of physical occurrences distributed over time or space to be recognized comprising the steps of:
- encoding said digital input signal values into coded input signal values to form a digital signal representation of said pattern employing a redundant code which has a normalized coded input signal values Hamming distance to normalized input signal value distance relationship having a mean slope greater than 2 for input signal value distances up to a predetermined level of at least 20% of the maximum input signal value distance;
- and
- using the coded input signal values to recognize the pattern of physical occurrences.
A method of training a pattern recognition device which is arranged to receive a set of digital input signal values representing a pattern of physical occurrences distributed over time or space to be recognized comprising the steps of:
- encoding said digital input signal values by employing a number of bits which is:
- i) greater than that which a binary code would require to represent an equal number of signal levels;
- and
- ii) less than that which a bar code would require to represent an equal number of signal levels;
- and being selected to have a monotonicity at least 1.5 times that of a binary code;
- and
- using the coded input signal values to recognize the pattern of physical occurrences.
Dependencies are inferred from the claim text.