Future AI PulseTag: Machine learning
Explore the archive. Publication dates show when each story was posted.
-
NVIDIA Kumo Relational: define the prediction before connecting the tables
NVIDIA’s new relational foundation model uses historical examples across connected tables. A practical pilot still needs clear targets and time boundaries.
-
Permutation importance: find which inputs your model relies on
Use permutation importance to inspect a trained model, interpret score drops, and avoid misleading conclusions when inputs overlap.
-
Precision and recall: why 97% accuracy can still miss important cases
A worked example explains precision, recall and accuracy, with a prompt for checking denominators and counts before trusting a model report.
-
Preprocessing leakage: why splitting after data cleaning can spoil an ML test
A scaler or feature selector can leak test information before model fitting. Learn how to inspect preprocessing boundaries and ask AI for a dependency audit.