PhD Seminar • Machine Learning • Out-of-distribution Detection with Flow Generative ModelsExport this event to calendar

Friday, June 24, 2022 — 1:00 PM to 2:00 PM EDT

Please note: This PhD seminar will be given online.

Dihong Jiang, PhD candidate
David R. Cheriton School of Computer Science

Supervisor: Professor Yaoliang Yu

Out-of-distribution (OOD) data come from a distribution that is different from training data. Detecting OOD data contributes to secure deployment of machine learning models. Currently, deep generative models have been widely used as an unsupervised approach for OOD detection.

In this work, we aim to re-examine the potential of generative flow models in OOD detection. We first propose a simple combination of univariate one-sample statistical test (e.g., Kolmogorov-Smirnov) and random projections in the latent space of flow models to perform OOD detection. Then, we propose a two-sample version of our test to account for imperfect flow models. Quite distinctly, our method does not pose parametric assumptions on OOD data and is capable of exploiting any flow model. Experimentally, firstly we confirm the efficacy of our method against state-of-the-art baselines through extensive experiments on several image datasets; secondly we investigate the relationship between model accuracy (e.g., the generation quality) and the OOD detection performance, and found surprisingly that they are not always positively correlated; and thirdly we show that detection in the latent space of flow models generally outperforms detection in the sample space across various OOD datasets, hence highlighting the benefits of training a flow model.

Download the full paper.


To join this PhD seminar on Zoom, please go to https://uwaterloo.zoom.us/j/93898859913.

Location 
Online PhD seminar
200 University Avenue West

Waterloo, ON N2L 3G1
Canada
Event tags 

S M T W T F S
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
1
2
  1. 2022 (126)
    1. July (8)
    2. June (17)
    3. May (20)
    4. April (24)
    5. March (22)
    6. February (16)
    7. January (19)
  2. 2021 (210)
    1. December (21)
    2. November (13)
    3. October (12)
    4. September (21)
    5. August (20)
    6. July (17)
    7. June (11)
    8. May (16)
    9. April (27)
    10. March (20)
    11. February (13)
    12. January (19)
  3. 2020 (217)
  4. 2019 (255)
  5. 2018 (217)
  6. 2017 (36)
  7. 2016 (21)
  8. 2015 (36)
  9. 2014 (33)
  10. 2013 (23)
  11. 2012 (4)
  12. 2011 (1)
  13. 2010 (1)
  14. 2009 (1)
  15. 2008 (1)