Deep learning is a subset of machine learning that uses neural networks with many layers (hence "deep") to learn hierarchical representations of data. Each layer transforms its input into a slightly more abstract form — raw pixels become edges, edges become textures, textures become parts, and parts become objects. This automatic feature learning replaced the manual feature engineering (SIFT, HOG, Haar) that dominated computer vision before 2012.

The field took off when AlexNet won ImageNet in 2012 using a GPU-trained convolutional neural network. Since then, architectures have evolved rapidly: ResNet (2015) introduced skip connections enabling 100+ layer networks, EfficientNet (2019) optimized the scaling of depth/width/resolution, and Vision Transformers (2020) brought attention-based architectures to vision. On the detection side, the YOLO family, Faster R-CNN, and DETR each represent different design philosophies for real-time localization.

Deep learning requires large labeled datasets and significant compute (GPUs or TPUs) for training, but inference can run on everything from cloud servers to edge devices and mobile phones. Transfer learning — taking a model pre-trained on a large dataset and fine-tuning it on your specific task — has made deep learning accessible even with limited data and hardware budgets.