afshinea/stanford-cs-230-deep-learning

Features

• Deep Learning Cheatsheets - The core deliverable of this repository: illustrated cheat sheets for Stanford CS 230 deep learning course.
• Convolutional Neural Network Architectures - Teaches configuring layer types, filter sizes, and activations for processing grid-structured data.
• Convolutional Neural Networks - Summarizes CNN architecture, layers, and operations in a single-page visual cheat sheet.
• Training Tips - Provides practical advice on hyperparameters, regularization, and debugging strategies for training.
• Dropout Regularization - Describes random neuron deactivation during training to force robust feature learning.
• CNN Architectures - Teaches building CNNs with convolution, pooling, and fully connected layers for image classification.
• Normalization Layers - Covers batch normalization and layer normalization techniques for stabilizing deep network training.
• Neural Network Architectures - Illustrates key deep learning architectures including CNNs, RNNs, and GANs with diagrams and notation.
• Pre-trained Model Transfer - Explains adapting pre-trained models by freezing or retraining layers for new tasks.
• Regularization Techniques - Documents dropout, weight penalties, and batch normalization to prevent overfitting in neural networks.
• Weight Initialization - Covers Xavier and other initialization methods that set initial weights to improve training stability.
• Weight Regularization - Explains L1, L2, and Elastic Net penalties that constrain weight magnitudes to reduce overfitting.
• Recurrent Neural Networks - Teaches processing sequential data with architectures that maintain hidden states across time steps.
• Cheatsheet Collections - The primary identity of this repository: a collection of cheat sheets for Stanford CS 230.
• Deep Learning Review - Reviews illustrated cheat sheets covering CNNs, RNNs, and training techniques for deep learning.
• Deep Learning - Provides single-page visual cheat sheets compressing Stanford CS 230 deep learning topics.
• ReLU Variants - Covers ReLU, Leaky ReLU, and ELU activation functions for introducing non-linearity.
• Input Sequence Attentions - Covers attention mechanisms that weight input positions to emphasize important information.
• Anchor Box Detections - Covers object detection with bounding boxes, anchor boxes, and non-max suppression.
• Object Detection - Covers object detection using region-based and single-shot detection architectures.
• Softmax Classifiers - Covers softmax layers that convert score vectors into probability distributions for classification.
• Gated Gradient Stabilizers - Covers GRU and LSTM gated architectures that prevent vanishing and exploding gradients.
• Word Embeddings - Learns dense vector representations that capture semantic similarity between words.
• Neural Style Transfers - Covers neural style transfer combining content and style images using pretrained networks.
• Vanishing Gradient Mitigations - Covers vanishing gradient mitigation as a key training challenge addressed in the cheat sheets.
• Semantic Similarity Calculation - Compares semantic similarity between words using cosine distance on vector embeddings.
• Sequence Evaluation Metrics - Describes BLEU score computation for evaluating generated sequences against reference texts.
• Text Translation Services - Covers encoder-decoder architectures with beam search for translating text between languages.
• Image Augmentations - Covers training data augmentation through flipping, rotation, cropping, and color shifts.
• Translation Quality Evaluation - Evaluates machine translation quality by comparing n-gram overlap with human reference translations.
• Xavier Initializations - Documents Xavier initialization that accounts for layer dimensions to improve convergence.
• Artistic Style Transfer - Covers artistic style transfer by optimizing combined content and style cost functions.
• Text Sequence Generators - Generates text by predicting the next token and sampling from the probability distribution.
• Concept-Diagram Pairings - Pairs schematic diagrams with mathematical notation to bridge intuition and formal understanding.
• Deep Learning Training Tips - Weaves practical training advice on hyperparameters, regularization, and debugging into theoretical explanations.
• Deep Neural Network Training Optimization - Provides practical tips for training deep models including hyperparameter tuning and regularization.
• Learning and Reference - Stanford CS 230 cheatsheets.
• AI Cheat Sheets - Essential reference materials for deep learning coursework.

Star history