Model Selection for Optimal Prediction in Statistical Learning - Part 2 / 2

Time: Wednesday 27-May-2020 16:00 (This is a past event.)

Speaker:
Discussion Moderator:

Artifacts

Motivation / Abstract
Professor Ernest will walk us through a statistical framework for model selection. His emphasis on investigating underlying probabilistic phenomenon is crucial to a methodical understanding of how the data behaves. This consistency will be shown through every step of the modelling journey; choosing the most appropriate metric for model accuracy and likelihood function, aggregation techniques, and how to evaluate model performance from a probabilistic and statistical perspective.
Questions Discussed
- Choosing appropriate metrics for model accuracy
- Choosing appropriate likelihood functions
- Aggregation techniques
- Model performance evaluation
... But from a probabilistic perspective
Key Takeaways
- Professor Ernest walked us through how to choose our function space by teaching us how to go about choosing our estimator functions. 
- He then showed us how to make decisions regarding refining our function space by showing us how to cross-validate and aggregate our models.
- Finally, he walked us through how to determine the success of our function space by demonstrating how tools like confidence intervals and VC dimensions can be leveraged to measure accuracy.  
Subjects:
Stream Categories:
 SpotlightAuthor SpeakingML Interpretability