1.[Paper: On Learning Invariant Representations for Domain Adaptation by Han Zhao]07:50
2.Background08:43
3.Motivation12:40
4.A Simple Example14:00
5.An Information-Theoretic Lower Bound18:20
6.Experiments21:07
7.Summary22:00
8.Q/A23:57
9.[Paper: Lexicographic and Depth-Sensitive Margins in Homogeneous and Non-Homogen...27:39
10.Setting28:30
11.Inductive Bias Sources29:11
12.Previous Results30:07
13.homogeneous and non-homogeneous models.30:44
14.Main Contributions - Homogeneous Models31:06
15.Thank You!32:09
16.[Paper: Adversarial Generalization of Time Frequency Feature with Application in...32:20
17.Time to time-frequency32:42
18.Time-frequency to time33:36
19.Is it consistent?34:09
20.Applied to GANs35:04
21.Evaluation35:19
22.[Paper: On the Universality of Invariant Networks by Haggai Maron]36:42
23.Invariant tasks36:59
24.Goal of this paper37:26
25.Formal definition of group action37:39
26.Invariant and equivariant functions38:30
27.G-invariant networks38:52
28.Main question:
How expressive are
G-invariant networks?39:37
29.How expressive are G-invariant networks?39:59
30.Theoretical results40:23
31.Universality of high-order networks40:26
32.Lower bound on network order40:38
33.The End41:08
34.[Paper: Fine-Grained Analysis of Optimization and generalization for Overparamet...41:28
35."Rethinking generalization" Experiment [Zhang et al '17]41:57
36.Setting: Overparam Two-Layer ReLU Neural Nets43:28
37.Training Speed44:14
38.Explaining Generalization despite vast overparametrization45:48
39.[Paper: Gauge Equivariant Convolutional Networks and Icosahedral CNNs by Maurice...46:59
40.Convolutions on Riemannian manifolds47:29
41.Convolution kernels on manifolds47:46
42.Weight sharing on manifolds - parallel transport48:04
43.Weight sharing on manifolds - Gauge Equivariant CNNs48:51
44.Gauges and gauge transformations49:28
45.Local gauges on manifolds50:58
46.Gauge transformations of feature fields51:14
47.Feature fields - Examples53:14
48.Gauge equivariant convolution54:58
49.Equivariance under active isometrics56:23
50.The Icosahedron56:49
51.Curvature of the Icosahedron57:15
52.The Gauge58:58
53.Gauge equivariant convolution59:29
54.Experiments - MNIST1:00:43
55.Q/A1:03:24
56.[Paper: Feature-Critic networks for Heterogeneous Domain Generalisation by Timot...1:07:40
57.Motivation1:08:02
58.Heterogeneous DG is a Common Workflow1:08:42
59.Methodology: Key Idea1:09:29
60.Algorithm1:10:10
61.Results1:11:45
62.Thanks for Listening!1:12:39
63.[Paper: Learning to Convolve- A Generalized Weight-Typing Approach by Daniel E W...1:12:55
64.Symmetry1:13:10
65.Equivariance & Convolution1:13:55
66.Group Convolutions1:14:48
67.Unitary Group Convolutions1:15:23
68.Learning Convolutions1:15:46
69.Experiments: MLP - CNN1:16:28
70.Experiments: Filters1:17:00
71.Transformation robustness1:17:22
72.Thanks1:17:54
73.[Paper: On Dropout and Nuclear Norm Regularization by Poorya Mianjy]1:18:07
74.Problem Setup 1:19:04
75.Empirical Observation1:20:33
76.Main Results1:22:02
77.Thanks for your attention!1:23:15
78.[Paper: Gradient Descent Finds Global Minima of Deep Neural Networks by Simon Du...1:23:33
79.Empirical Observations on Empirical Risk1:24:01
80.Setup1:24:52
81.Trajectory-based Analysis1:25:21
82.Proof Sketch1:25:54
83.Main Results1:27:11