Articles, learning paths and news on AI, machine learning and data science.
Tutorial series: every part of a topic, in order.
by MLHub Labs · 12 parts · about 281 min · intermediate
How machines see: convolutions, the classic architectures, transfer learning, detection and segmentation.
A complete, runnable walkthrough of loading a pretrained ResNet18, replacing its head, fine-tuning it in two phases on an image-folder …
MLHub Deep Learning Organization
1 day, 12 hours ago • 29 min read
A practical guide to opening up convolutional networks with saliency maps and Grad-CAM, so you can see which pixels actually …
MLHub Deep Learning Organization
3 days, 13 hours ago • 24 min read
How vision models learn useful representations from unlabeled images alone, through contrastive learning (SimCLR, MoCo, BYOL) and masked image modeling …
MLHub Deep Learning Organization
5 days, 21 hours ago • 22 min read
How the transformer architecture, originally built for text, was adapted to see images by treating them as sequences of patches, …
MLHub Deep Learning Organization
6 days, 21 hours ago • 24 min read
How neural networks learn to label every pixel in an image, and the difference between painting a scene by category …
MLHub Deep Learning Organization
1 week ago • 24 min read
How detectors go from a grid of pixels to a list of labelled boxes: the idea of anchors, how YOLO …
MLHub Deep Learning Organization
1 week, 1 day ago • 27 min read
Why training a CNN from scratch is rarely the right move, and how to reuse a pretrained network's learned features …
MLHub Deep Learning Organization
1 week, 5 days ago • 21 min read
How making label-preserving copies of your training images—flips, crops, color jitter, Mixup, CutMix—helps a CNN generalize instead of memorize, and …
MLHub Deep Learning Organization
2 weeks ago • 25 min read
A guided tour through the classic CNN architectures — LeNet, AlexNet, VGG, Inception and ResNet — showing which specific ideas …
MLHub Deep Learning Organization
2 weeks, 3 days ago • 26 min read
A step-by-step tour of how convolution, activation, pooling and dense layers combine into a working image classifier, with worked shapes, …
MLHub Deep Learning Organization
2 weeks, 4 days ago • 19 min read
A hands-on, number-by-number look at how convolutional layers scan images, how padding and stride control output size, and why pooling …
MLHub Deep Learning Organization
2 weeks, 6 days ago • 27 min read
Before a convolutional network can look at anything, an image has to become a grid of numbers. This article builds …
MLHub Deep Learning Organization
3 weeks, 2 days ago • 13 min read
• 6 min read
• 5 min read
• 6 min read
• 6 min read
3 weeks, 4 days ago • 17 min read
0 likes • 117 views • 1 comment
3 days, 20 hours ago • 6 min read
0 likes • 49 views • 0 comments
4 days, 21 hours ago • 15 min read
0 likes • 56 views • 0 comments
Get the latest AI insights and tutorials delivered to your inbox.
Your feedback has been received and we'll review it soon.
