/pTSAfall2021

Probabilistic time series

Primary LanguageJupyter Notebook

pTSAfall2021

timeseries2021

DS-GA 3001.001 Probabilistic Time Series Analysis

Lecture

Mondays from 2:00pm-3:40pm SILV_520 (Silver Center for Arts & Science, 100 Washington Square East, room 520)

Lab (required for all students)

DS-GA 1018.002 Lab (cap = 40) Wednesday from 3:45pm-4:35pm SILV_520 (Silver Center for Arts & Science, 100 Washington Square East, room 520)

DS-GA 1018.003 Lab (cap = 40) Wednesday from 9am-9:50am GCASL_475

Instructor

Cristina Savin, csavin@nyu.edu Office hours: TBD

TAs

Section 002 - Haresh Rengaraj Rajamohan (hrr288@nyu.edu)

Section 003 - Ying Wang (yw3076@nyu.edu)

Office hours: TBD

Overview

This graduate level course presents fundamental tools for characterizing data with statistical dependencies over time, and using this knowledge for predicting future outcomes. These methods have broad applications from econometrics to neuroscience.The course emphasizes generative models for time series, and inference and learning in such models. We will cover range of approaches including Kalman Filter, HMMs, AR(I)MA, Gaussian Processes, and their application to several kinds of data.

Note: information presented is tentative, syllabus may be subject to change as course progresses. Brightspace version is always up-to-date and to be used as main reference.

Grading

problem sets (25%) + midterm exam (20%) + final project (25%) + lab(20%) + participation(10%)

Participation: piazza, engagement during lectures, labs, and office hours

Piazza

We will use Piazza as the main platform for communication, for announcements, and discussions about the course. Interactions on Piazza, particularly good answers to other students' questions, will count toward the participation grade.

Projects

Work in groups of 2-3 students.* Topics are flexible, including applying know algorithms to an interesting dataset, reviewing and implementing a state of the art solution, to improving an existing algorithm. Project proposals due Oct 22th.

*Check with CS if you are considering working individually or in a larger group.

Online recordings

Lecture videos will be posted to NYU Classes and we will be providing zoom access when students are unable to come to class (due to quarantine, issues with travel, etc). Class attendance is generally required.

Schedule and detailed syllabus

Date Lecture Assignments
Sept. 13 Lecture 1: Logistics. Introduction to time series. Graphical models
Sept. 15 [Recitation]
Sept. 20 Lecture 2: Basic statistics of time series. AR: inference and learning
Sept. 22 Lab 1: AR
Sept. 27 Lecture 3: ARIMA models
Sept. 29 Lab 2: ARIMA
Oct. 4 Lecture 4: LDS, Kalman filtering
Oct. 6 Lab 3: Inference in LDS
Oct. 12 Lecture 5: EM Kalman
Oct. 13 Lab 4: LDS parameter learning
Oct. 18 Lecture 6: Particle filtering Project proposal due
Oct. 20 Lab 5: Particle filtering
Oct.25 Lecture 7: Hidden Markov Models
Oct.27 Lab 6: HMMs
Nov.1 Lecture 8: Links between models, generalizations.
Nov.3 Midterm recap (no lab)
Nov.8 Mid-term exam
Nov.10 No lab. Projects Q&A
Nov.15 Lecture 8: Intro to GPs
Nov.17 Lab 7: GP regression
Nov. 22 Lecture 10. Deep learning for time series
Nov. 24 Lab 8: RNNs
Nov.29 Lecture 11. Spectral methods
Dec.1 Lab 9: Spectral methods
Dec.6 12. Guest lecture: Text generation, transformers (He He)
Dec. 8 No lab
Dec. 13 Final projects presentation Project reports due Dec.20th

Bibliography

There is no required textbook. Assigned readings will come from freely-available online material.

Core materials

  • Time series analysis and its applications, by Shumway and Stoffer, 4th edition
  • Pattern recognition and machine learning, Bishop
  • Gaussian processes Rassmussen & Williams

Useful extras

Academic honesty

We expect you to try solving each problem set on your own. However, if stuck you should discuss things with other students in the class, subject to the following rules:

  • Brainstorming and verbally discussing the problem with other colleagues ok, going together through possible solutions, but should not involve one student telling another a complete solution.
  • Once you solve the homework, you must write up your solutions on your own.
  • You must write down the names of any person with whom you discussed it. This will not affect your grade.
  • Do not consult other people's solutions from similar courses.
  • Credit should be explicitly given for any code you use that you did not write yourself.
  • Violations result in a zero score on that assignment, and a notice to the DGS.

Late submission

Penalties: 20% points off assignment for each extra day of delay.