#statstab #357 Uncertainty Estimation with Conformal Prediction

Thoughts: Haven't parsed this properly but maybe be an interesting discussion point. How best to quantify uncertainty?

#conformalprediction #bayesian #confidenceintervals #uncertainty

https://m-clark.github.io/posts/2025-06-01-conformal/

#statstab #223 Conformal predictions w/ {marginaleffects}

Thoughts: Sometimes you need a range of likely future values. To get an assumption-free Prediction Interval, use conformal methods.

#r #stats #marginaleffects #prediction #conformalprediction

https://marginaleffects.com/bonus/conformal.html

Some thoughts on #conformalprediction
for #timeseries with Xiaoqian Wang https://robjhyndman.com/publications/cpts.html

In the last couple of weeks I've been learning about #ConformalPrediction, a family of algorithms to measure the uncertainty of predictions made by #MachineLearning models.

Here are a few links to get you started:
- CP course by @ChristophMolnar https://mindfulmodeler.substack.com/p/week-1-getting-started-with-conformal
- Multi-class notebook (in Spanish) https://nbviewer.org/github/MMdeCastro/Uncertainty_Quantification_XAI/blob/main/UQ_multiclass.ipynb
- MAPIE library: https://mapie.readthedocs.io/en/latest/index.html
- TorchCP library: https://github.com/ml-stat-Sustech/TorchCP

Nos vemos *hoy* en nuestra reunión de marzo: Analítica acelerada con Shapelets y conformal prediction, este mes en The Bridge

https://www.meetup.com/pydata-madrid/events/299749589/

¡Te esperamos a las 19:00! Y después, networking

The distinction between marginal and conditional coverage finally clicked for me. #conformalprediction provides the former but not the latter, and for many (most?) real-world use cases in ML one wants the latter.

If it sounds too good to be true...

#AWS Fortuna is skyrocketing! Just a few days, and so many GitHub stars and forks! ️

Fortuna supports #ConformalPrediction, #BayesianInference and other methods for #UncertaintyQuantification in #DeepLearning.

Try it out and let us know!
https://github.com/awslabs/fortuna

In collaboration with @cedapprox, @andrewgwils and team.

Today, we open sourced Fortuna (https://github.com/awslabs/fortuna) a library for uncertainty quantification.
Deep neural networks are often overconfident and do not know what they don’t know. Quantifying the uncertainty in the predictions they make will help deploy deep learning more responsibly and more safely.
#responsibleAI #ConformalPrediction #BayesianInference #UncertaintyQuantification #deeplearning #opensource