Day #145 - ML Paper Learning Notes

Article - Link - Prediction, Interaction, and User Behaviour

Summary of Learning's

• Classifying user activity in Smartphone using Machine Learning Model (Random Forest)
• Generating text using CharRNN
• RNN and LSTM with Keras

Article - Continuing Creative Sequences
Code - Link
Paper #1 - Time-series Extreme Event Forecasting with Neural Networks at Uber

Key Learning

• Time-series modeling based on Long Short Term Memory (LSTM)
• Objective is to Estimate peak electricity demand, traffic jam severity and surge pricing for ride sharing
• Derived new feature variables - Weather information (e.g., precipitation, wind speed, temperature)
• Derived new feature variables - City level information (e.g., current trips, current users, local holidays)
• Features vectors are then aggregated via an ensemble technique
• Three criteria for picking a neural network model for time-series: (a) number of timeseries (b) length of time-series and (c) correlation among the time-series. If (a), (b) and (c) are high then the neural network might be the right choice

Paper #2 - AcctionNet: A Dataset Of Human Activity Recognition Using On-phone Motion Sensors
Network Architecture - The FCNN architecture

convolution2d_input_1: InputLayer
convolution2d_1: Convolution2D
activation_1: Activation
convolution2d_2: Convolution2D
activation_2: Activation
maxpooling2d_1: MaxPooling2D
dropout_1: Dropout
flatten_1: Flatten
features: Dense
activation_3: Activation
dropout_2: Dropout
dense_no_soft: Dense
activation_4: Activation

Classes, Biking, Downstairs, Stationary Biking, Jumping, Lunging, Running, Squatting, Standing, Upstairs, Walking

Human Activity Recognition Using Smartphones Data Set 

Happy Learning!!!