PhD Seminar • Artificial Intelligence — Deep Homogeneous Mixture Models: Representation, Separation, and ApproximationExport this event to calendar

Tuesday, June 12, 2018 — 3:00 PM EDT

Priyank Jaini, PhD candidate
David R. Cheriton School of Computer Science

At their core, many unsupervised learning models provide a compact representation of homogeneous density mixtures, but their similarities and differences are not always clearly understood. In this work, we formally establish the relationships among latent tree graphical models (including special cases such as hidden Markov models and tensorial mixture models), hierarchical tensor formats and sum-product networks. Based on this connection, we then give a unified treatment of exponential separation in exact representation size between deep mixture architectures and shallow ones. In contrast, for approximate representation, we show that the conditional gradient algorithm can approximate any homogeneous mixture within epsilon accuracy by combining O(1/ epsilon^2) "shallow" architectures, where the hidden constant may decrease (exponentially) with respect to the depth. Our experiments on both synthetic and real datasets confirm the benefits of depth in density estimation.

This is a joint work with Pascal Poupart and Yaoliang Yu.

Location 
DC - William G. Davis Computer Research Centre
2306C
200 University Avenue West

Waterloo, ON N2L 3G1
Canada

S M T W T F S
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
3
4
5
  1. 2019 (192)
    1. September (19)
    2. August (18)
    3. July (12)
    4. June (23)
    5. May (23)
    6. April (32)
    7. March (25)
    8. February (16)
    9. January (24)
  2. 2018 (220)
    1. December (16)
    2. November (19)
    3. October (26)
    4. September (22)
    5. August (17)
    6. July (20)
    7. June (13)
    8. May (25)
    9. April (34)
    10. March (24)
    11. February (3)
    12. January (1)
  3. 2017 (36)
  4. 2016 (21)
  5. 2015 (36)
  6. 2014 (33)
  7. 2013 (23)
  8. 2012 (4)
  9. 2011 (1)
  10. 2010 (1)
  11. 2009 (1)
  12. 2008 (1)