FULL FORCE FRIDAY:🆕Mar 13 ENCORE!🎧

ANCST - Neural Collapse EP 🇩🇪⚛️

4 Track EP from Berlin, German Blackened Metal/Cyber Crust outfit⚛️

BC➡️angstnoise.bandcamp.com/album/neural...⚛️

#Ancst #NeuralCollapse #BlackenedCyberCrust #GermanMetal #FFFMar13 #KMäN

Today at #AISTATS2026

An Illusion of Unlearning? Assessing #MachineUnlearning Through Internal Representations.

MU algorithms seem to achieve perfect unlearning of classes through misalignment of weights & features rather than truly unlearning. We analyze this through #NeuralCollapse

Research shows neural collapse continues despite low‑rank bias, indicating the effect endures across varied model constraints. Read more: https://getnews.me/study-finds-neural-collapse-persists-despite-low-rank-bias/ #neuralcollapse #machinelearning

Neural Collapse theory applied to multi‑label federated learning improves accuracy on rare labels in benchmarks. Read more: https://getnews.me/neural-collapse-multi-label-federated-learning-reduces-label-skew/ #neuralcollapse #federatedlearning

Controlling Neural Collapse boosts OOD detection and transfer‑learning, the study shows. Presented at ICML 2025, it was tested on CIFAR‑10‑C, SVHN and ImageNet‑O. Read more: https://getnews.me/controlling-neural-collapse-boosts-ood-detection-and-transfer-learning/ #neuralcollapse #ooddetection

A new 'avoid-shortcut' framework aligns training with Neural Collapse without extra steps, and the paper's second preprint version was posted in September 2025. https://getnews.me/debiased-learning-via-neural-collapse-avoiding-shortcut-models/ #neuralcollapse #debiasing

Study uses Neural Collapse to align federated client features, boosting accuracy and label‑wise F1 on four benchmark datasets across eight multi‑label settings. Read more: https://getnews.me/neural-collapse-for-multi-label-federated-learning-with-skewed-labels/ #federatedlearning #neuralcollapse