Integrated circuit acceleration method and system based on self-adaptive piecewise linear approximation curve

Abstract

The invention belongs to the technical field of artificial intelligence, in particular to an integrated circuit acceleration method and system based on a self-adaptive piecewise linear approximation curve. The integrated circuit acceleration method comprises the steps: determining a linear equation correlation coefficient corresponding to input curve activation data through a table lookup method,forming an approximation straight line corresponding to the curve activation data through a linear equation correlation coefficient; and forming a piecewise linear approximation curve of a curve activation function through fitting of an approximation straight line. The integrated circuit acceleration method solves the problems that in the prior art, an existing artificial intelligence algorithm islow in curve activation function calculation speed and low in calculation result precision, and has the beneficial technical effects of being high in curve calculation speed, high in calculation result precision, small in occupied hardware resource, good in operation real-time performance and suitable for any curves.

Description

technical field

[0001] The invention belongs to the technical field of artificial intelligence, and in particular relates to an integrated circuit acceleration method and system based on an adaptive segmented linear approximation curve. Background technique

[0002] In the research field of artificial neural networks, the linear activation function only linearly combines the input and then outputs, and cannot approximate complex functions. Multi-layer neural networks are no different from single-layer. Curved activation functions play a crucial role in the function approximation capabilities of deep neural networks. Artificial intelligence algorithms often use curve activation functions, such as sigmoid function or tanh function, because integrated circuits are not easy to implement curve calculations such as exponential operations or trigonometric function operations, and there is no unified formula for calculating different curve activation functions. The traditional proc...

Examples

Embodiment 1

[0120] Embodiment one: if figure 1 As shown, an integrated circuit acceleration method based on an adaptive piecewise linear approximation curve, including:

[0121] Determine the linear equation correlation coefficient S101 corresponding to the input curve activation data S204 by the look-up table method, form the corresponding approximation straight line S102 of the curve activation data through the linear equation correlation coefficient, and form the piecewise linear approximation curve of the curve activation function by fitting the approximation straight line S103.

[0122]Due to the use of the table look-up method to determine the corresponding linear equation correlation coefficient of the input curve activation data, the corresponding approximation straight line of the curve activation data is formed by the linear equation correlation coefficient, and the piecewise linear approximation curve of the curve activation function is formed by the fitting of the approximatio...

Embodiment 2

[0240] Such as Figure 8 , 9 , 10, 11 The following uses the calculation of the Sigmoid curve activation function as an example to illustrate an integrated circuit acceleration algorithm and system based on an adaptive piecewise linear approximation curve, as follows:

[0241] Step 1: Establish a curve approximation database: In the initialization stage, divide all addresses into 8 intervals. The system selects two adjacent points A0 and A1 of the curve activation function, constructs a straight line passing through the two points, calculates the slope coefficient and bias coefficient of the corresponding straight line equation, and writes it into the curve approximation database. The system selects two adjacent points A1 and A2 of the curve activation function, constructs a straight line passing through the two points, calculates the slope coefficient and bias coefficient of the corresponding straight line equation, and writes it into the curve approximation database. Selec...

technical field

[0001] The invention belongs to the technical field of artificial intelligence, and in particular relates to an integrated circuit acceleration method and system based on an adaptive segmented linear approximation curve. Background technique

[0002] In the research field of artificial neural networks, the linear activation function only linearly combines the input and then outputs, and cannot approximate complex functions. Multi-layer neural networks are no different from single-layer. Curved activation functions play a crucial role in the function approximation capabilities of deep neural networks. Artificial intelligence algorithms often use curve activation functions, such as sigmoid function or tanh function, because integrated circuits are not easy to implement curve calculations such as exponential operations or trigonometric function operations, and there is no unified formula for calculating different curve activation functions. The traditional proc...

Examples

Embodiment 1

[0120] Embodiment one: if figure 1 As shown, an integrated circuit acceleration method based on an adaptive piecewise linear approximation curve, including:

[0121] Determine the linear equation correlation coefficient S101 corresponding to the input curve activation data S204 by the look-up table method, form the corresponding approximation straight line S102 of the curve activation data through the linear equation correlation coefficient, and form the piecewise linear approximation curve of the curve activation function by fitting the approximation straight line S103.

[0122]Due to the use of the table look-up method to determine the corresponding linear equation correlation coefficient of the input curve activation data, the corresponding approximation straight line of the curve activation data is formed by the linear equation correlation coefficient, and the piecewise linear approximation curve of the curve activation function is formed by the fitting of the approximatio...

Embodiment 2

[0240] Such as Figure 8 , 9 , 10, 11 The following uses the calculation of the Sigmoid curve activation function as an example to illustrate an integrated circuit acceleration algorithm and system based on an adaptive piecewise linear approximation curve, as follows:

[0241] Step 1: Establish a curve approximation database: In the initialization stage, divide all addresses into 8 intervals. The system selects two adjacent points A0 and A1 of the curve activation function, constructs a straight line passing through the two points, calculates the slope coefficient and bias coefficient of the corresponding straight line equation, and writes it into the curve approximation database. The system selects two adjacent points A1 and A2 of the curve activation function, constructs a straight line passing through the two points, calculates the slope coefficient and bias coefficient of the corresponding straight line equation, and writes it into the curve approximation database. Selec...

Images