50+ days of Machine Learning

 Hello :) Today is Day 150! A quick summary of today: read a few more papers about taxi demand and Origin-Destination matrix prediction using Graph Neural Networks added more audio clips to the Scottish (Glaswegian) dataset Obisdian makes this nice graphs of papers referencing each other. First paper is  Modeling Long- and Short-Term Temporal Patterns with Deep Neural Networks (some will be pictures because I am yet to find a better way to share the math notations) Introduction Multivariate time series forecasting can learn traffic jam patterns ahead of time. However, a common challenge is capturing dynamic dependencies across multiple variables. For example the traffic on a highway can experience two patterns - daily (morning vs evening traffic) and weekly (weekday vs weekend). If a model is not able to handle complex data, then time-series forecasting is not feasible. To this end, the authors propose the usage of deep neural networks (RNN and CNN), and in particular - Long- and Short