Some algorithms I have implemented recently. Most of the vision notebooks are based off ideas from the book "Computer Vision - Algorithms and Applications" by Springer. The machine learning models are a bit of everything. Some basic, some more advanced stuff. You need to show a spectrum of knowledge.
Locally adaptive histogram
Qlearning
I wrote a basic game to test q-learning.
Anti-aliasing
Gan
based off https://arxiv.org/abs/1610.09585
Dropout
based off http://jmlr.org/papers/volume15/srivastava14a/srivastava14a.pdf model without dropout
model with dropout
Feature checker
Based off the idea from https://github.com/evilsocket/ergo (relevance.py). What feature is contributing what to the models results? If you are going to use this in production, set some random data points to zero during trainings as well.
Filters fixing noise
Gradient descent
Finding a local minima
A*
Search algorithm.
K-means clustering
LeCun CNN
Implemented LeChun CNN model, based off http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf
Markov Random Field
Hough Transform
Basic implementation of hough Transform.
Auto encoder
Gan in numpy
Segmentation graph
MNIST generalization test
got the idea from https://arxiv.org/pdf/1611.03530.pdf , how much noise can a simple model handle and still do good evaluation on a noise free dataset?
the accuracy over time is based on the traing data.
Generalization
https://arxiv.org/pdf/1611.03530.pdf
Poisson image editing
Mostly a fork off this implementation, I fixed support for python3 and made it work with all the channels. Removed opencv depency as well.
Feature detection
Harris corner detector to find special features in each image. Using MSE to connect the special features.
Counterfactual regret minimization
Python implementation of the rock, paper, scissor section of http://modelai.gettysburg.edu/2013/cfr/cfr.pdf
Lucas–Kanade method, optical flow
Segnet
Had memoryerrors so this model was trained on only a subset of the training set (please hire me so I can build a computer for machine learning). I used the CamVid dataset.
Transfer learning
https://en.wikipedia.org/wiki/Transfer_learning
Autoencoder fixing image noise
Numpy rnn
Loosely based on iamtrask post. Maybe I wouldn't have coded this if he had used a linked list.