Emergency To do:

1. Modify all summation related part in homework ops.mathematics check all of this

matmul
summation
transpose

1. Check summation gradient make hw1's solution to hw2's summation

Make all the axes to be tuples, such like in summation, remove the check if isinstance(self.axes, int):, because according to the notebook, the axes should be tuples.

Important file

autograd.py

Automatic Differentiation and gpu accelaration

Deep Learning System

This repository contains a series of homework assignments for the CMU Deep Learning System course. There are 4 homework assignments in total, from hw0 to hw4.

Homework Assignments Overview

hw0

I still haven't finish question 6, the c++ part.

Hw0 includes functions to read and parse the MNIST dataset, compute softmax loss, and perform Stochastic Gradient Descent (SGD) for both softmax regression and a simple two-layer neural network, providing a foundational framework for the MNIST digit classification problem.

Pay attention to

• src/simple.ml.py

hw1

I still haven't understand matmul backward pass, summation backward pass, broadcastto backward pass, reshape backward pass, negate backward pass, transpose backward pass.

HW1 expands on HW0 by implementing forward and backward computations for various operators within a computational graph framework, enabling automatic differentiation. And also implement a topological sort for reverse-mode backpropagation, adapt the softmax loss function for tensors, and train a two-layer neural network using stochastic gradient descent (SGD).

Pay attention to

• python/needle/ops/ops_mathematic.py