In-depth articles on deep learning foundations, optimization, and computer vision — built for practitioners who want the intuition and the math.
The non-linear ingredient that makes deep learning possible. ReLU, Sigmoid, Tanh, Softmax — what they do and when to use each.
Feature scaling, batch norm, momentum, RMSProp, Adam, and LR decay — from first principles to production practice.
Stop your network memorizing noise. L1, L2, Dropout, and Early Stopping — how each works and when to combine them.
A frozen EfficientNetB0 backbone at 87–89% accuracy in under 3 minutes. Feature caching and classifier-head design.
Expand your dataset with flips, crops, and color jitter — plus modern approaches like MixUp and CutMix.
A deep dive into Krizhevsky et al. (2012) — the architecture that won ImageNet and kicked off the deep learning era.
Occasional posts on deep learning and computer vision. No spam.