Machine-Learning-tensorflow

Course work/course design in CMPS242, Autumn 2017, UCSC

Cooperated with Yuan Yang, De Huo

1. Ham or Spam

• Text Classification problem
• Used NLTK(Natural Language Toolkit) to extract text feature
• Implemented Batch/Stochastic Gradient Descent logistic regression
• Implemented EG+-(Exponentiated Gradient +-). Reference: Exponentiated Gradient +-
• Compared among different algorithms
• Achieved 99% Correctness

2. Trump VS Clinton

• Complex Text Classification problem
• Implemented LSTM(Long Short-term memory) RNN(Recurrent Neural Network) with Feed-Forward Neural Network
• Cross Entropy Soft Max Loss Function
• Achieved 90% Correctness

3. Image Captioning

• Used Flickr-8k dataset
• Used Keras-tensorflow neural network
• Image feature extraction using Inception V3
• Word feature extraction using LSTM RNN
• Image caption model using bidirectional LSTM RNN
• Achieved 60% Correctness (due to training time limitation)