Embedded Files
Project Overview
Project Overview
The objective of this project is to create a simple Artificial Neural Network on a RISC-V simulator. Using vector dot product, matrix-matrix multiplication, argmax and an activation function, these functions are combined to load a pretrained network and classify handwritten digits.
The objective of this project is to create a simple Artificial Neural Network on a RISC-V simulator. Using vector dot product, matrix-matrix multiplication, argmax and an activation function, these functions are combined to load a pretrained network and classify handwritten digits.
Language Used: RISC-V Assembly Code
Language Used: RISC-V Assembly Code
Since this was programmed in assembly code, it was important that our code followed the calling conventions.
Since this was programmed in assembly code, it was important that our code followed the calling conventions.
Neural Network Model
Neural Network Model
Handwritten digits from 0-9 are input as a "flattened" vector of size 784x1. This vector gets multiplied with a pre-trained m0 matrixed. Subsequently, a rectifier function is performed on the output to filter out negative correlations. Afterwards, the result is multiplied with a pre-trained m1 matrix, and argmax is used to determine the index with the highest correlation.
Page updated
Google Sites
Report abuse