Method and apparatus with data processing
AssigneeSamsung Electronics Co., Ltd.
InventorsIhor Vasyltsov, Wooseok Chang, Youngnam Hwang
Life of this patent
Estimating the term…
Abstract
A processor-implemented data processing method includes: normalizing input data of an activation function comprising a division operation; determining dividend data corresponding to a dividend of the division operation by reading, from a memory, a value of a first lookup table addressed by the normalized input data; determining divisor data corresponding to a divisor of the division operation by accumulating the dividend data; and determining output data of the activation function corresponding to an output of the division operation obtained by reading, from the memory, a value of a second lookup table addressed by the dividend data and the divisor data.
Claims27
A processor-implemented data processing method, the method comprising:
- normalizing input data of an activation function, the activation function comprising a division operation;
- based on a precision of the activation function, selecting a first lookup table from a first set of first lookup tables;
- determining dividend data corresponding to a dividend of the division operation by reading, from a memory, a value of the selected first lookup table addressed by the normalized input data;
- determining divisor data corresponding to a divisor of the division operation by accumulating the dividend data using an accumulator;
- based on the precision of the activation function, selecting a second lookup table from a second set of second lookup tables;
- and
- determining output data of the activation function corresponding to a quotient output of the division operation obtained by reading, from the memory, a quotient value from the selected second lookup table, the quotient value addressed by a column-index value determined by the dividend data and addressed by a row-index value determined by a determined number of most significant bits of the accumulator, wherein the second lookup table comprises a column-index corresponding to dividends, a row-index corresponding to divisors, and quotient values, including the quotient value, indexed by the column-index and the row-index.
A data processing apparatus, the apparatus comprising:
- a memory storing instructions;
- and
- one or more processors configured by the instructions to:
- normalize input data of an activation function, the activation function comprising a division operation;
- based on a precision of the activation function, select a first lookup table from a first set of first lookup tables;
- determine dividend data corresponding to a dividend of the division operation by reading, from the memory, a value of the selected first lookup table addressed by the normalized input data;
- determine divisor data corresponding to a divisor of the division operation by accumulating the dividend data using an accumulator;
- based on the precision of the activation function, select a second lookup table from a second set of second lookup tables;
- and
- determine output data of the activation function corresponding to a quotient output of the division operation obtained by reading, from the memory, a quotient value from the selected second lookup table, the quotient value addressed by a column-index value determined by the dividend data and addressed by a row-index value determined by a determined number of most significant bits of the accumulator, wherein the second lookup table comprises a column-index corresponding to dividends, a row-index corresponding to divisors, and quotient values, including the quotient value, indexed by the column-index and the row-index.
A processor-implemented data processing method, the method comprising:
- based on a precision of an activation function, selecting a first lookup table from a first set of first lookup tables;
- determining dividend data corresponding to a dividend of the activation function by reading, from a memory, a value of the selected first lookup table based on input data of the activation function;
- determining divisor data corresponding to a divisor of the activation function by accumulating the dividend data using an accumulator;
- based on the precision of the activation function, selecting a second lookup table from a second set of second lookup tables;
- and
- determining output data corresponding to an output of the activation function by reading, from the memory, a quotient value from the selected second lookup table based on a column-index value determined by the dividend data and a row-index value determined by a determined number of most significant bits of the accumulator, wherein the second lookup table comprises a column-index corresponding to dividends, a row-index corresponding to divisors, and quotient values, including the quotient value, indexed by the column-index and the row-index.
A processor-implemented data processing method, the method comprising:
- normalizing input data of an activation function, the activation function comprising a dividend and a divisor;
- based on a precision of the activation function, selecting a first lookup table from a first set of first lookup tables;
- determining dividend data by reading, from the selected first lookup table of one or more memories, a value of a column-index corresponding to a value of the normalized input data;
- determining divisor data by accumulating the dividend data using an accumulator;
- based on the precision of the activation function, selecting a second lookup table from a second set of second lookup tables;
- and
- determining output data of the activation function by reading, from the selected second lookup table of the one or more memories, a quotient value of a column-index value corresponding to the value of the normalized input data and of a row-index value corresponding to a value of a determined number of most significant bits of the accumulator, wherein the second lookup table comprises a column-index corresponding to dividends, a row-index corresponding to divisors, and quotient values, including the quotient value, indexed by the column-index and the row-index.
Dependencies are inferred from the claim text.