Time Series and Machine Learning Reading Group
Oct 2022- Feb 2023, University of Southampton
In this reading group we will read a series of papers on time series, with focuses on nonparametric/semiparametric modeling, learning theory, and forecasting with deep neural networks .
This reading group is hybrid — we meet weekly on Monday 13:00-14:30, both at 5001/B54 and via MS Teams. Feel free to choose your preferred method to join in.
Please check this website regularly for the most up-to-date arrangement.
Materials
-
Week 1. Rates of convergence for empirical processes of stationary mixing sequences
-
Week 2. Invariance principles for absolutely regular empirical processes
-
Week 3. Estimating semiparametric ARCH(∞) models by kernel smoothing methods (Part I)
-
Week 4. On Hoeffding’s inequality for dependent random variables
-
Week 5. Rademacher complexity bounds for non-i.i.d. processes
-
Week 6. Simultaneous nonparametric inference of time series
-
Week 7. Sequence to sequence learning with neural networks
-
Week 8. Local linear fitting under Near Epoch Dependence
-
Week 9. DeepAR: probabilistic forecasting with autoregressive recurrent networks
-
Week 10. On nonparametric estimation in nonlinear AR(1)-models
-
Week 11. Enhancing the locality and breaking the memory bottleneck of transformer on time series forecasting
-
Week 12. Stability bounds for stationary φ-mixing and β-mixing processes
Group Members
Supplementary References
- Empirical Process Techniques for Dependent Data. Herold Dehling, Thomas Mikosch, Michael Sørensen, Springer, 2002.
- Theory and Algorithms for Forecasting Non-Stationary Time Series.
Vitaly Kuznetsov and Mehryar Mohri, NIPS Tutorial, 2016
- Mixing: Properties and Examples. Paul Doukhan, Springer-Verlag, 1994.
- Minimax time series prediction. Wouter M. Koolen, Alan Malek, Peter L. Bartlett, Yasin Abbasi Yadkori, NIPS, 2015
Roles of Presenter and Discussant
If it is your first time attend a reading group, you might find the reading group tips by Lester Mackey and Percy Liang helpful.
Every time we will have one people (presenter) present the main contents and another people (discussant) raise questions and lead the discussion.
-
As a presenter: you should have an in-depth reading and develop a solid understanding of all the details in the assigned topic. You should prepare well, and make sure you deliver a logically clear and technically accessible presentation. In short words, it is your job to have everyone in the meeting understand the main ideas of the reading. Beside above lecture notes, you may also find the following supplementary references helpful for preparing your presentation.
-
As a discussant: you should be more familiar with the content than if you were simply in the group. You don’t need to know everything. You can pause the presentation, ask questions (to the presenter or to the audience), and facilitate discussions. It is your job to help the presenter to have everyone (yourself included!) in the meeting understand the main ideas of the reading and having learned something.
Before each session, although not compulsory I would recommend following amount of time spent on reading:
- Presenter: > 10 hours;
- Discussant: 5 hours;
- General audience: 2 hours.
If you encounter any problem during your reading, feel free to discuss with me or other staff members.
Past Reading Groups:
Webpage maintained by Chao Zheng. Last updated on 18/11/2022
Time Series and Machine Learning Reading Group
Oct 2022- Feb 2023, University of Southampton
In this reading group we will read a series of papers on time series, with focuses on nonparametric/semiparametric modeling, learning theory, and forecasting with deep neural networks .
This reading group is hybrid — we meet weekly on Monday 13:00-14:30, both at 5001/B54 and via MS Teams. Feel free to choose your preferred method to join in.
Timetable (provisional)
Please check this website regularly for the most up-to-date arrangement.
Materials
Week 1. Rates of convergence for empirical processes of stationary mixing sequences
Week 2. Invariance principles for absolutely regular empirical processes
Week 3. Estimating semiparametric ARCH(∞) models by kernel smoothing methods (Part I)
Week 4. On Hoeffding’s inequality for dependent random variables
Week 5. Rademacher complexity bounds for non-i.i.d. processes
Week 6. Simultaneous nonparametric inference of time series
Week 7. Sequence to sequence learning with neural networks
Week 8. Local linear fitting under Near Epoch Dependence
Week 9. DeepAR: probabilistic forecasting with autoregressive recurrent networks
Week 10. On nonparametric estimation in nonlinear AR(1)-models
Week 11. Enhancing the locality and breaking the memory bottleneck of transformer on time series forecasting
Week 12. Stability bounds for stationary φ-mixing and β-mixing processes
Group Members
Supplementary References
Vitaly Kuznetsov and Mehryar Mohri, NIPS Tutorial, 2016
Roles of Presenter and Discussant
If it is your first time attend a reading group, you might find the reading group tips by Lester Mackey and Percy Liang helpful.
Every time we will have one people (presenter) present the main contents and another people (discussant) raise questions and lead the discussion.
As a presenter: you should have an in-depth reading and develop a solid understanding of all the details in the assigned topic. You should prepare well, and make sure you deliver a logically clear and technically accessible presentation. In short words, it is your job to have everyone in the meeting understand the main ideas of the reading. Beside above lecture notes, you may also find the following supplementary references helpful for preparing your presentation.
As a discussant: you should be more familiar with the content than if you were simply in the group. You don’t need to know everything. You can pause the presentation, ask questions (to the presenter or to the audience), and facilitate discussions. It is your job to help the presenter to have everyone (yourself included!) in the meeting understand the main ideas of the reading and having learned something.
Before each session, although not compulsory I would recommend following amount of time spent on reading:
Past Reading Groups:
Webpage maintained by Chao Zheng. Last updated on 18/11/2022