Laboratory of Structural Methods of Data Analysis in Predictive
Modeling Moscow Institute of Physics and Technology

Лекция Modern Parametric Statistics

Spokoiny Vladimir
Должность: Руководитель

Описание курса

Final Exam
Place: IITP. room 404
Dates: 27-28 February 2014


Vladimir Spokoiny (PreMoLab MIPT)

Venue: Independent University of Moscow, Bolshoj Vasilievsky per.11., lecture room will be announced

Exam dates
26-28 February 2014

Course Materials:

  • Textbook:
  • Video:


Application areas: 
Data mining

Research topics: 
Statistical methods
Structural adaptive inference

Даты проведения и расписание:
Дата Расписание


Choice of the bandwidth in local parametric estimation using propagation approach.
- Sizer and Intersection-of-confidence-intervals idea and local model selection.
- Choice of tuning parameters by propagation.
- Propagation property and oracle risk bound.


Local parametric approach:
- Applications to regression, generalized regression, density models.
- Examples of local constant and local linear approximation.
- Local Fisher and Wilks results.


Penalized maximum likelihood and the problem of choosing the penalty:
- Fisher and Wilks for penalized maximum likelihood.
- Uniform confidence bands and concentration of the empirical risk.
- Choice of tuning parameters by propagation idea.


19-00 - 20-30
Nonparametric function estimation:
- White noise and regression models, generalized regression, density models.
- Sieve approximation, modeling bias, bias-variance decomposition.
- Model selection via unbiased risk estimation.


19-00 - 20-30
Few chapters of modern parametrics:
- Concentration and large deviations,
- Fisher and Wilks expansions and corollaries.


19-00 - 20-30
Basics of parametric statistics: maximum likelihood approach, exponential family, linear model.
Properties of maximum likelihood, concentration and confidence sets, Gauss-Markov, Cramer-Rao and van Trees results.

Дополнительные материалы