Published Oct 23, 2019

Reproducing Deep Learning Models

Kyle Polich hosts Rajiv Shah as he critiques and attempts to reproduce a deep learning model for earthquake prediction, revealing challenges like data leakage and underscoring the need for baseline models, transparency, and robust validation techniques in model building.

Episode Highlights

Topics covered

Deep Learning Critique
01
Model Reproducibility
02
Model Evaluation Techniques
03

Popular Clips

Reproducibility in Machine Learning
Play Clip
Model Building Wisdom
Play Clip
Earthquake Aftershock Prediction
Play Clip
Machine Learning Best Practices
Play Clip
Avoiding Data Leakage
Play Clip
Deep Learning in Earthquakes
Play Clip
Model Performance Analysis
Play Clip
Data Science Bootcamp
Play Clip

Reproducing Deep Learning Models

Topics covered

Popular Clips

Episode Highlights

Deep Learning CritiqueRajiv Shah critiques a deep learning model for earthquake prediction, highlighting the lack of a baseline model and the trade-off between interpretability and accuracy. His analysis underscores the importance of foundational steps and transparency in model building.

Deep Learning Critique

Model ReproducibilityRajiv Shah's attempt to reproduce an earthquake-predicting deep learning model revealed significant challenges and insights. His work underscores the importance of open science and collaboration in improving model reliability and academic discourse.

Model Reproducibility

Model Evaluation TechniquesRajiv Shah uncovers critical issues in reproducing a deep learning model for earthquake prediction, focusing on data leakage and performance metrics. His analysis reveals the importance of robust validation techniques to ensure model reliability.

Model Evaluation Techniques