From the fundamentals of neural networks through reinforcement learning, object detection and recognition, and generative models, explained through both implementation and math.
This is an editorially selected reading order. Begin at step 1 or go directly to the topic you need.
Machine learning is not just model selection: define the task, data split, loss, evaluation, and deployment change.
Neural networks stack affine transforms and nonlinear activations, then reduce loss by backpropagation and gradient-based optimization.
Object Detection draws boxes around things; Semantic Segmentation colors in every pixel by meaning. From the difference between the two, through Instance and Panoptic Segmentation, the lineage from R-CNN to YOLO, DETR and DINO, all the way to their 3D point-cloud counterparts — a systematic tour of the fundamentals of scene understanding.
Break down Ultralytics YOLO11 using official documentation and primary sources: C3k2, C2PSA, anchor-free detection, DFL/CIoU losses, and measured benchmarks for n/s/m/l/x.
This article breaks down SegFormer, introduced by NVIDIA, The University of Hong Kong, and collaborators at NeurIPS 2021, based on the original paper. We examine the hierarchical MiT encoder without positional encoding, the parameter-efficient All-MLP decoder, the effective receptive field, and the measured mIoU, parameter counts, and compute from B0 to B5 with equations.
Pose estimation infers human or object keypoints and 6D pose from images. This primer covers coordinates, losses, occlusion, evaluation, and safety.
Generative models represent data distributions and create conditional outputs. Learn likelihood, VAE, GAN, diffusion, evaluation, provenance, and safety.