22-part comprehensive master curriculum: Perceptrons, Backpropagation, Optimizers, CNNs, Sequence Models, Attention, Transformers, Autoencoders & GANs.
Comprehensive foundations of artificial neurons, Rosenblatt perceptron, multi-layer architectures, activation functions, and biological vs artificial neural computation.
Layer-by-layer forward execution, matrix multiplications, linear transformations, logits calculation, probability mapping, and computational graphs.
Comprehensive mathematical guide to regression losses (MSE, MAE, Huber) and classification losses (BCE, Categorical Cross-Entropy, Focal Loss).
Analytical and computational derivation of backpropagation: partial derivatives, Jacobian matrices, multivariate chain rule, and gradient flows.
Mechanisms of Batch Gradient Descent, Stochastic Gradient Descent (SGD), Mini-Batch GD, learning rate schedules, and loss surface navigation.
In-depth guide to modern deep learning optimizers: SGD with Momentum, Nesterov, AdaGrad, RMSProp, Adam, and decoupled weight decay AdamW.
Preventing vanishing and exploding gradients through principled weight initialization: Xavier/Glorot, He/Kaiming, and variance preservation analysis.
Techniques for model generalization: L1 Lasso, L2 Ridge weight decay, inverted Dropout, early stopping, and data augmentation strategies.
Stabilizing internal covariate shift: Batch Normalization, Layer Normalization in Transformers, Instance Normalization, Group Normalization, and RMSNorm.
Deep Feedforward Networks (MLP): universal approximation theorem, width vs depth tradeoffs, residual connections, and production network design.
Foundations of spatial computer vision: 2D convolutions, kernels, stride, padding, receptive fields, pooling layers, and full CNN pipelines.
Evolution of computer vision backbones: LeNet-5, AlexNet, VGG-16/19, Inception modules, ResNet residual skip-connections, and modern ConvNeXt.
End-to-end computer vision: classification, object detection (YOLO, Faster R-CNN), semantic & instance segmentation (U-Net, Mask R-CNN), and mAP/IoU evaluation.
Processing temporal and sequential data: Recurrent Neural Network (RNN) hidden state dynamics, Backpropagation Through Time (BPTT), and vanishing gradients.
Overcoming long-term dependencies: Cell state highway, forget gate, input gate, output candidate generation, and bidirectional LSTM architectures.
Streamlined gated sequence processing: Reset and update gate mechanisms, hidden state interpolation, computational efficiency, and GRU vs LSTM comparison.
The breakthrough in neural representations: Additive (Bahdanau) attention, multiplicative (Luong) attention, and Query-Key-Value mathematical formulation.
Attention Is All You Need: Scaled dot-product self-attention, Multi-Head Attention, sinusoidal and rotary positional encodings, feed-forward sublayers, and layer normalization.
Taxonomy of transformer models: Autoencoding (BERT Masked LM), Autoregressive (GPT causal decoder), Sequence-to-Sequence (T5), and modern open LLMs.
Mapping discrete entities to dense continuous vector spaces: One-hot vs dense vectors, Word2Vec (Skip-gram & CBOW), subword tokenization, and vector similarity metrics.
Unsupervised representation learning: Undercomplete, sparse, and denoising autoencoders, Variational Autoencoder (VAE) probabilistic latent spaces, and KL divergence.
Adversarial zero-sum game dynamics: Generator vs Discriminator architectures, minimax loss functions, mode collapse mitigation, and Wasserstein GAN with Gradient Penalty.