In this homework, I need to work on five thousand reviews and train a proper model using those data. With my supervised learning experiment, using one thousand most frequent unigrams and bigrams can effectively reduce the MRE score to 0.289548. Besides that, with Term Frequency – Inverse Document Frequency (tf-idf) features, I can further reduce
Continue Reading “Recommendation System and Web Mining Homework 4”
There are two tasks need to be fulfilled in this homework. Firstly, helpfulness prediction is a supervised learning task, where target value is a boolean value. One can tell that the more predictors are used, the better the model performance is. Secondly, rating prediction is another supervised learning task with ratings ranging from one star
Continue Reading “Recommendation System and Web Mining Homework 3”
There are three main tasks in this homework. Firstly, I need to train multiple models with different regularization coefficients and report their results. Secondly, results for precision and recall with different thresholds are required. At the end, explained variance with various number of components from Principle Component Analysis is also required. Experiment results and codes
Continue Reading “Recommendation System and Web Mining Homework 2”
In this homework, I am asked to perform learning tasks with different sets of features and report their corresponding coefficients and Mean Squared Error (MSE). Experiment results and codes are attached in the report.
In this project, we are interested in one of the hottest topics related to deep learning recently: conditional Generative Adversarial Network (cGAN), and try to add some stochasticity of it. Since it is a very recent work, we spend much time in collecting related information and trying to fully understand those papers. Later, we take
Continue Reading “Neural Networks and Pattern Recognition Final Report”
In this final homework, we are going to use recurrent network to learn and create a song from scratch. There are some hyper parameters that we have turned for a while, including the temperature of the deep learning model, number of neurons in hidden layer, dropout rate, and different optimizers. For model training strategy, one
Continue Reading “Neural Networks and Pattern Recognition Homework 4”
In this homework, we are given some pre-trained deep learning models, such as VGG16, to work on other image classification tasks, including Caltech256 and Urban Tribes. To prevent train the whole hidden neurons and links in VGG16 again, we have a surgery on VGG16: replace the last neuron layer with brand new neurons and train
Continue Reading “Neural Networks and Pattern Recognition Homework 3”
For the individual written part, I am asked to derive the back-propagation update formulas for multiple single variables and as a matrix. For the team programming part, we are required to implement a neural network in Python from scratch and work on a learning problem to verify our derived update formulas. Experiment results and codes
Continue Reading “Neural Networks and Pattern Recognition Homework 2”
In the written part, we are asked to solve some problems in our textbook, Bishop. Later on, we need to derive the updates for logistic regression and Softmax regression and run experiments on them to verify our derivations. Experiments results and codes are included in report.
In the written part, I am asked to calculate the closet point on a line to origin. In the programming part, I need to find out the fundamental matrix with following steps: feature detection, feature matching, outlier rejection, linear estimation, and non-linear estimation. The former three parts are very similar to what I have done
Recent Comments