Write a function
Matrix operation
Working with data structures
Knowledge with general programming
Working with Jupyter Notebook
Running experiment and document in Jupyter Notebook
Project 1
Reading and processing dataset
Project 2
Implementation of Neural Network
Back-prop, Gradient descent.
Implementation of NN layer, feed forward.
Project 3
Project 4
Mutual-class classification
Python and Numpy tutorial: https://cs231n.github.io/python-numpy-tutorial/#jupyter-and-colab-notebooks CNN: https://cs231n.github.io/convolutional-networks/ BackProp: https://cs231n.github.io/optimization-2/#backprop Efficificient back-prop by Prof.Yann Lecun: http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf
Classify hand-written digit: http://neuralnetworksanddeeplearning.com/chap1.html
Cool visualization: https://cs231n.github.io/assets/nn3/opt2.gif https://cs231n.github.io/assets/nn3/opt1.gif
Goal: (Probability, JupNotebook, running experiments with python) Estimated Time: 2-3 hour
From python’s built-in ramdomInt (sampling a number from [0,1]) simulate and estimate the value PI
Draw a square with side length 1, sample a point, count how many points lies within the circle Visualize sampling process in Jupitor Notebook See how simulation converge to the real value of PI
Goal: (Image processing, matrix operation, convolution kernel on image, convoluting, pooling) Estimated Time: 3-4 hour
Given a image apply different matrix operation Turn image gray-scale Blurring the image [1,1,1] [1,1,1] [1,1,1] Sharpen the image
Goal: (ML, NN, data process, Gradient Descent,) Estimated Time: 5 hour
Implement NN layer, pooling, activation func.
TBD: need a good project? If no GPU machine, just run Feed forward layer without training? Simple training network. Maybe linear classifier or Perceptron?
Goal: (CNN, back prop, multi-class classification) Estimated Time: 1 week training time
Build a vanilla CNN model to perform multi-class classification) TBD: Dataset Dogs/Cat Umich’s action-actor Machine requirement Perhaps google cloud machine? Free $200 with student email. Need to check UR GPU machine?