Many open source libraries for conformal prediction, such as MAPIE and PUNCC, fall short when it comes to handling tabular data in classification tasks. Unfortunately, some of these libraries adopt a one-size-fits-all approach by repurposing models designed for computer vision,

Once academia hops on a hyped topic it often keeps churning papers for the sake of churning papers. Traffic prediction domain is one example, transformers and llm for time series is another, but there are many others.

For over 20 years #bayesianism suppressed conformal prediction from even been mentioned in scientific papers and books. Here is yet another example, a book written by lecturer from University of Warwick and published by Springer that purports to be general introduction to uncertainty

When it comes to uncertainty quantification, conformal prediction strongly dominates alternatives - it produces prediction intervals with guaranteed coverage, in line with confidence specified by the user (e.g. 90%) - conformal prediction also produces stable and informative

Conformal prediction for feature engineering and selection. The next BIG idea for ML research and applications 🔍 **Unlocking the Power of Conformal Prediction for Feature Engineering and Feature Selection** 🚀 Conformal prediction is proving to be a game-changer not just in

Credit: not me. #snakeoil

#bayesianism

#bayesianism

🚨 **Contrary to popular belief, the Bootstrap Method doesn’t give accurate estimates of uncertainty!** 🚨 In a recent study published in *Nature Computational Materials* (2022), researchers explore the use of bootstrap methods for uncertainty quantification and highlight

Bayesianism. Conformal prediction. #bayesianism #conformalprediction