Deep Learning
Deep learning is a subset of machine learning that uses artificial neural networks with multiple layers to learn complex patterns from vast amounts of data.
Layers: Input, Hidden (multiple), Output. Optimization: Loss functions, Optimizers (Adam, RMSprop). Hardware: GPUs (NVIDIA A100/H100), TPUs. Frameworks: PyTorch, TensorFlow, Keras, JAX.
graph LR
Center["Deep Learning"]:::main
classDef main fill:#7c3aed,stroke:#8b5cf6,stroke-width:2px,color:white,font-weight:bold,rx:5,ry:5;
classDef pre fill:#0f172a,stroke:#3b82f6,color:#94a3b8,rx:5,ry:5;
classDef child fill:#0f172a,stroke:#10b981,color:#94a3b8,rx:5,ry:5;
classDef related fill:#0f172a,stroke:#8b5cf6,stroke-dasharray: 5 5,color:#94a3b8,rx:5,ry:5;
linkStyle default stroke:#4b5563,stroke-width:2px;
🧒 Explain Like I'm 5
Imagine you want to teach a robot to tell the difference between a cat and a dog. In the old days, you had to tell it: 'Look for pointy ears' or 'Look for a wagging tail'. With [Deep Learning](/en/terms/deep-learning), you just show it 1,000,000 pictures of cats and dogs and say 'This is a cat, that is a dog'. The robot looks at the pictures over and over until it learns the secret patterns that make a cat a cat. It learns like a human baby does—by seeing examples.
🤓 Expert Deep Dive
Technically, Deep Learning is characterized by the use of 'Deep Architectures' (networks with many hidden layers). The fundamental training mechanism is 'Backpropagation' combined with 'Stochastic Gradient Descent' (SGD). By calculating the 'Gradient' of the loss function with respect to the weights, the network can adjust its parameters to reduce error. Crucial components include 'Activation Functions' (like ReLU) which introduce non-linearity, and 'Regularization' techniques (like Dropout) to prevent overfitting. Modern breakthroughs are driven by specific architectures: 'Convolutional Neural Networks' (CNNs) for vision, 'Recurrent Neural Networks' (RNNs) for sequences, and 'Transformers' for the attention-based language modeling used in Large Language Models (LLMs).