Santiago and the ML Models

Personal newsletter where I share what I'm learning, writing, and up to in the field of ML/AI

The hottest Substack posts of Santiago and the ML Models

And their main takeaways

Temporal degradation framework and other ideas

19 implied HN points • 05 Jun 23

🔬 Science Data science Machine Learning Model Evaluation

The author is working on a Temporal Model Degradation Framework for AI models.
They have implemented an experiment with early results showing model performance degradation over time.
The author plans to conduct a Continuous Retraining Experiment to test if continuous retraining can prevent model degradation.

Monitoring Workflow for Machine Learning Systems

19 implied HN points • 06 Mar 23

🕹 Technology Machine Learning Monitoring

Machine learning models naturally degrade over time due to changing environments and dynamics.
Traditional ML monitoring methods focus on data drift and realized model performance, which can be limited.
A new ML monitoring workflow emphasizes estimating model performance in real-time and using drift detection for root cause analysis, reducing false alerts.

Writing my master's thesis in public

4 HN points • 17 Mar 23

🚌 Education Data science Machine Learning Research Methods

Writing a master's thesis can be overwhelming, but sharing the process publicly can make it less lonely.
Choosing the right thesis topic and finding a supportive supervisor are crucial steps.
For a thesis project, replicating testing frameworks, analyzing datasets, and implementing new methods are significant tasks.

Continuous retraining and formalizing the model aging framework

0 implied HN points • 19 Jun 23

🕹 Technology Data science Machine Learning

Santiago Víquez is formalizing a model aging framework in his Master's thesis
Improvements in the temporal degradation framework include using cross-validation for model training and validating models based on test set error
Continuous retraining experiments show that model performance may be best before any updates are implemented