Department seminar by Ashkan Ertefaie, University of RochesterExport this event to calendar

Thursday, May 9, 2019 — 4:00 PM EDT

 Robust Q-learning


The main goal of precision medicine is to use patient characteristics to inform a personalized treatment plan as a sequence of decision rules that leads to the best possible health outcome for each patient. Q-learning is a reinforcement learning algorithm that is widely used to estimate an optimal dynamic treatment regime using both multi-stage randomized clinical trials and observational studies. Starting with the final study stage, Q-learning finds the treatment option that optimizes the desired expected outcome. Fixing the optimally-chosen treatment at the last stage, Q-learning moves backward to the immediately preceding stage and searches for a treatment option assuming that future treatments will be optimized. The process continues until the first stage is reached. Q-learning requires specifying a sequence of regression models and the validity of the concluding results relies on assuming that the models are correctly specified. Specifically, due to the nature of backward induction, the subsequent models are likely to be a complex function of covariates which may result in non-ignorable residual confounding under model misspecification. We propose a robust Q-learning method that leverages flexible machine learning techniques to reduce the chance of model misspecification, thereby while maintaining the efficiency of Q-learning, mitigating the main drawback of this method. We derive the asymptotic properties of our method and show that, under certain conditions, it will result in asymptotically linear estimators with certain influence functions.

Location 
M3 - Mathematics 3
Room: 3127
200 University Avenue West

Waterloo, ON N2L 3G1
Canada

S M T W T F S
28
29
30
31
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
  1. 2019 (58)
    1. November (4)
    2. October (7)
    3. September (4)
    4. August (2)
    5. July (2)
    6. June (2)
    7. May (7)
    8. April (7)
    9. March (6)
    10. February (4)
    11. January (13)
  2. 2018 (44)
    1. November (6)
    2. October (6)
    3. September (4)
    4. August (3)
    5. July (2)
    6. June (1)
    7. May (4)
    8. April (2)
    9. March (4)
    10. February (2)
    11. January (10)
  3. 2017 (55)
  4. 2016 (44)
  5. 2015 (38)
  6. 2014 (44)
  7. 2013 (46)
  8. 2012 (44)