Interpretable Machine Learning with Serg Masis

Serg Masis is the author of best-selling book 'Interpretable Machine Learning with Python' and senior Data Scientist at Sygenta. He has mentored many data scientists around the world. 

Timestamps:

00:00 intro

08:30  Old 4.77 MH  z Computer, Late 80s and Programming

11:51 Fairness, Accountability and Transparency in Machine Learning, Startup and Harvard

16:33  Fairness vs Preciseness, Bias and Variance Tradeoff, Are Engineers to blame?

21:43 Mask-Detection Problem in Coded-Bias, Biased Samples,  Surveillance using CV

32:38 Fixing Biased Datasets, Augmenting Data and Limitations 

37:39 Algorithmic Optimisation and Explainability

40:51 Eric Schmidt on Behavioral Prediction, SHAP values, Tree and DeepExplainers

44:50 Challenges of using SHAP and LIME & Big Data

49:37 GPT3, Large Models and ROI on Explainability

01:00:00  TCAS, Collision Risks and Interpretability, Ransom Attacks

01:08:09 Guitar, Bass, and Led Zepplin

01:09:31 Birth Order and IQ, Science vs Folk Wisdom

01:13:30  Reverse Discrimination & Men, Bias in Child Custody, Prison Sentences, and Incarceration

01:23:11 Receidivism to Criminal Behaviour, Ethnic over-representation & Systematic Racism

01:24:44  Human Judges vs AI,  Absolute Fairness, Food and Parole

01:30:20  Face Detection in China, Privacy vs Convenience, Feature Engineering and Model Parsimony 

01:35:51 Sparsity, Interaction Effects, and Multicollinearity

01:38:23  Four levels of Global and Local Predictive Explainability

01:43:17  Recursive and Sequential Feature Selection

01:47:42  Ensemble, Blended and Stacked Models and Interpretability

01:53:45  In-Processing and Post-Processing Bias Mitigation

01:57:00  Future of Interpretable AI