[Tutorial] Explainable AI using Knowledge Graphs

Table of Contents (powered by https://videoken.com)
0:00:00 Explainable AI using Knowledge Graphs
0:02:52 Semantics of the Black Box
0:05:01 Tutorial Thesis
0:06:50 Evolution of Black Box
0:06:57 Deep Learning: Deeper you go, Darker it gets.
0:08:53 Nature of Deep Learning Models (limitations)
0:11:22 Importance of Semantics
0:13:06 Non- Trivial Case of Question Answering with Context:
0:14:37 Multi Hop Reasoning
0:15:32 Adding knowledge to data-intensive learning for performance and interpretability
0:15:56 Deep Learning: Computational Wonders & Supreme Predictors us Simple forms of knowledge
0:18:26 AlphaFold
0:20:50 Learning the Long Tail !!!
0:23:02 Predictions vs Symbolic Knowledge
0:25:08 Ability to Reason: Ticking the CheckList
0:26:59 Statistical Learning - Knowledge Representation
0:28:51 Explaining the clusters
0:30:43 More Context for Precise Response
0:32:18 Biased Blackboxes: Data has Inherent Bias that DL won't find - own--How to discover and mitigate?
0:37:21 Why K-IL? - Autonomous Driving K-IL: Shallow Infusion: Knowledge Graph Embeddings for Autonomous Driving
0:39:27 Opportunities Ahead
0:39:44 Just to Summarize..
0:42:15 Knowledge Graphs and Knowledge-infused Learning
0:42:35 K-IL: Probably Approximately Correct Learning
0:44:51 PAC Learning
0:45:41 Definition: Knowledge Infusion
0:47:48 Definition of Knowledge Graph
0:49:41 Challenges in Deep Learning : Why K-IL
0:51:48 Knowledge-infused Learning (K-IL)
0:55:27 K-iL : Shallow Infusion
0:57:55 K-IL: Shallow Infusion (shallow KR, shallow merging tech
0:59:48 Shallow Infusion: Contextualized Representation - Explanations
1:00:56 K-IL: Shallow Infusion: Knowledge Graph Embeddings for Autonomous Driving
1:01:05 K-IL: Semi-Deep Infusion
1:03:34 K-IL: Semi-Deep Infusion : Matching Reddit Conversation to DSM-5
1:03:39 Reddit - DSM-5
1:03:44 Interpretable and Explainable Learning
1:07:09 Knowledge in Recommender System : IMDb
1:07:21 K-IL : Objective Functions (OF) and Evaluation (EV)
1:07:24 Resources
1:07:33 Dataset 2b: Rank Tweets with Emoji and Mental Health Condition
1:07:36 Explainability and Interpretability using Domain Knowledge
1:08:48 Explainability is the ability to generate human-comprehensible explanations around the decision-making process
1:09:21 Interpretability is the ability to discern the internal mechanisms of any module
1:09:51 Explainability vs Interpretability
1:10:28 Empirically, an explainable system would comprise of collectively exhaustive interpretable subsystems and orchestration among them.
1:10:52 Goals of Explainability, and role of Domain knowledge
1:11:13 Role of Domain Knowledge in Explainability
1:12:12 Methods of Explainability
1:16:53 Post-hoc Explainability
1:17:12 Local Explanations
1:18:12 Explanation by Examples
1:18:30 Integrated Gradients
1:19:38 Integrated Hessians
1:19:42 Visualizations
1:19:45 Explain-then-Predict - An example
1:19:47 Domain Knowledge Fusion Opportunities for Better Explainability
1:20:54 Explanations and Interpretations for Education Domain Use Case
1:21:32 Prerequisite aware - Deep Knowledge Tracing (PDKT-C)
1:22:05 Explainable AI in Education @ Embibe
1:22:49 High Stake and Social Good Applications
1:23:13 Embibe AI Platform with KG as the Backbone
1:23:18 Knowledge Graph
1:24:54 [Demo]
1:27:49 Thanks! Open to questions?