Based on the slides of the talk: https://lovvge.github.io/Forecasting-Tutorial-WWW-2020/

Applications

• Financial / Medical / Traffic / Weather

Problem 1

• Goal: given a signal (eg., #packets over time)
• Find: patterns, periodicities, and/or compress

Problem 2

• Forecast

Problem 3

• Given: A set of correlated time sequences
• Forecast one of them.

Important Observations:

• Problems are related;

• Feature extraction =

  = (lossy) compression = Dimensionality reduction = Embedding = Latent/hidden variable detection = AutoEncoding

  …

Question: Find quickly stocks like ‘C’ (or customers like ‘smith’)

Answer: Summarize seq. to a few numbers/features (eg., avg, stdv, Fourier coeff.)