Table of Contents (powered by [ Ссылка ])
0:00:00 Causality, part 2 - Bernhard Scholkopf and Stefan Bauer
0:02:13 MLSS 2020 Causal Inference II
0:03:04 Additional Material
0:06:08 Causal Models as Posets of Distributions
0:07:47 Very brief orientation
0:10:22 Key problem - Many SCMs generate same distribution
0:10:53 Assumptions that enable Causal Discovery
0:12:11 Causal Structure Learning
0:15:28 Identifiability of linear non-Gaussian models
0:16:29 Independent Component Analysis
0:18:41 LINGAM: Linear non-Gaussian acyclic models causal discovery
0:20:36 Structure Learning: Time Series
0:20:58 Time series and Granger causality
0:22:01 Confounded Granger
0:23:25 Intervention Invariance
0:24:55 SCMs for ODEs & SDEs
0:26:16 Classic Approach and Causal Approach
0:28:07 How to measure invariance of an ODE?
0:30:08 Application to Signalling Pathway
0:31:43 Causal vs. Predictive -insample
0:32:18 Causal vs. Predictive - Out-of-Sample
0:33:09 Variable Selection - Rank individual variables on how often they appear in top ranked model.
0:33:46 Stabilized Regression
0:35:34 Summary I
0:39:39 Open Dynamic Robot Initiative
0:40:13 Follow-up: Transferable Dynamics Learning
0:41:03 A causal perspective on deep representation learning
0:45:57 Causal representation learning
0:46:37 Representation Learning: A Review and Perspectives
0:47:39 Causal Framework
0:49:13 Representation learning
0:49:54 Disentangled representations
0:50:16 What is disentanglement?
0:51:23 Unsupervised Learning of Disentangled Representation
0:51:43 Why Disentanglement?
0:51:59 Disentanglement methods: VAE + Regularizer
0:53:02 Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
0:54:14 Learning disentangled representations is challenging
0:55:30 Disentanglement Challenge
0:56:35 Summary and Open Questions
0:58:33 Weakly-Supervised Disentanglement
0:59:16 Fairness
0:59:32 Causality and fairness
1:00:06 Removing proxy discrimination
1:01:57 Are structured representations helpful for fairn stefanBauer
1:04:14 Implications of Correlations
1:04:24 Towards disentangled representations in rea environments
1:04:51 Disentangling correlated factors is nontrivial
1:05:58 Disentanglement metrics not affected by correlated
1:06:36 Disentangling correlated factors gets difficult for we correlation
1:07:02 What happens for the example model?
1:08:37 Structure by Architecture
1:08:46 Encoding Causal Structure
1:09:45 Structural Causal Autoencoders
1:11:37 Quantitative Results
1:11:51 Disentanglement by Architecture
1:12:32 Key Insights
1:13:38 Outlook: Towards Causal World Models
1:14:37 Learning Independent mechanisms
1:15:11 Method
1:16:18 Recurrent Independent Mechanisms
1:17:55 Upcoming - Using Robotic Systems as Benchmark
1:18:34 Some Scepticism of Simulation Environments
1:19:12 Hardware Design
1:19:44 What we have so far:
1:20:16 Summary
1:23:21 Advertisement - Upcoming ICML Workshop Inductive Biases, Invariances and Generalization in RL(BIG)
1:23:57 Advertisement - Open Internship Positions
1:24:25 Thank you

свернуть