Tag
Research
-
What EWC buys you that flat L2 does not
A registered kill-gate compares EWC against a flat L2 control on Permuted MNIST, and the same headline retention number hides two very different costs.
-
How wrong is a model's confidence?
A ResNet-34 on CIFAR-100 claimed 99% confidence while landing correctly 94% of the time; temperature scaling cut the gap by nearly two-thirds.
-
Catastrophic forgetting, measured
A small network lost 12.3 points of accuracy learning a second task, real but short of my registered threshold for severe. EWC recovered 98% of it.
-
Blackwell quantisation on my own card
Q4_K_M more than doubled throughput and cut VRAM by two thirds on my RTX 5070 Ti, staying 98% similar to full precision. The interesting part was where that 2% lived.
-
Detecting PII before the model sees it
Two registered experiments, one frozen dataset, two detectors that each miss a different category. The entity-level numbers say which, and by how much.
-
Why I pre-register a kill-gate for every experiment
A kill-gate is the number, written down in advance, at which I abandon my own hypothesis. How three real gates fired, passed and refused to answer.