Meetings: Tuesdays and Thursdays, 2:20-3:35pm, LH 023
Website: http://jfukuyama.github.io/teaching/stat681
| Instructor: Prof. Julia Fukuyama | jfukuyam at iu dot edu | |
| Office hours: Wednesdays 1-2pm, Thursdays 3:45-4:45pm | Swain East 225 | |
| Associate Instructor: Mr. Derek Brown | browndet at iu dot edu | |
| Office hours: Thursdays 9-11am | Swain East 200 |
Many scientifically useful generative models are straightforward to simulate from but have likelihoods that are unavailable or prohibitively expensive to evaluate. Simulation-based inference provides tools for estimating parameters, quantifying uncertainty, and evaluating model fit in these settings.
Students will construct and investigate generative models and use simulated data for estimation, uncertainty quantification, model checking, and hypothesis testing. Topics will include conditional simulation, Bayesian inference, approximate Bayesian computation, simulated method of moments, bootstrap methods, permutation tests, and modern approaches to simulation-based inference. Applications may include epidemic and ecological models, economic and agent-based models, and statistical issues in AI evaluation.
Class meetings will combine lecture, discussion, demonstrations, and guided computational work.
By the end of the course, students should be able to:
Grades will be based on:
The course competencies will be assessed primarily through a midterm examination on October 29. Students who have not yet demonstrated proficiency in a competency will have limited opportunities to reassess it using new problems. Later evidence of proficiency may replace weaker earlier evidence rather than being averaged with it.
I will give suggested homework problems + a practice midterm to help you better understand the material and so that you can prepare for the exam. These will not be graded, but you can come talk to us about them if you don’t feel like you understand something.
Labs are intended primarily to provide guided practice and feedback. The lab grade will be based principally on engaged, good-faith completion rather than on producing a completely correct solution during the class period.
The final project will require students to formulate or select a generative model, implement a simulation-based inferential method, evaluate the results, and discuss the assumptions and limitations of their analysis. Graduate projects will be expected to show greater independence, depth, and engagement with relevant literature; more structured project options will be available to undergraduate students. Students may also use a brief individual discussion of the project to provide additional evidence of proficiency in course competencies.
Regular attendance is important because the labs and discussions cannot be fully reproduced from written course materials. Attendance itself is not graded. Students who miss a lab or are unable to finish it during class may ordinarily complete it within one week. If illness, religious observance, a university-sponsored activity, or another serious circumstance affects a lab, examination, or scheduled reassessment, please contact me as early as reasonably possible so that alternative arrangements can be made.
Because the final project is due near the university’s grade-submission deadline, extensions may be limited except in cases involving an approved accommodation or serious extenuating circumstances.
Students requiring accommodations should contact IU Bloomington Accessible Educational Services at iubaes@iu.edu or visit the AES website. Please provide me with your accessibility memorandum as early as possible so that we can make appropriate arrangements.