Quick tech ticks and comparisons
Topics in the Deep Learning category.
Learn how a neural network decides whether an email is spam by turning real-world information into numbers, multiplying by weights, adding a bias, and applying an activation function. Uses a single, concrete example throughout.
The direct continuation of Part 1. Part 1 built the spam model with handcoded weights. Part 2 answers the natural next question — how does a model actually learn those weights? Covers labels, loss, backpropagation, and the complete training loop using the same spam example.
Learn tensors and tensor operations using the same spam-detection example. Covers feature vectors, batches of emails, weights, matrix multiplication, broadcasting, and reshaping. Builds directly on the training loop from Part 2.
Learn how to load, batch, shuffle, and iterate over spam-detection data in PyTorch using torch.utils.data.Dataset and DataLoader. Uses the same email example throughout.
Learn how PyTorch creates, tracks, and updates learnable parameters for a spam-detection model. Understand weights, biases, requires_grad, and how to inspect and manage model parameters.
Learn exactly what happens during forward and backward propagation, using a two-layer spam-detection model. Worked numerical example with the chain rule included.
Learn how deep learning mirrors ideas from the brain, using the spam-detection example throughout. Covers pattern recognition, memory, attention, generalisation, and learning from mistakes.
A one-page reference for every deep learning formula and concept from the tutorial series, all using the spam-detection example. Each formula has a memory trick.
A zero-background introduction to Convolutional Neural Networks. Starts with a simple cat-vs-dog photo example and builds up, one idea at a time, through convolution, activation, feature maps, padding, and pooling — then goes further into the exact output-size and parameter-count formulas, finishing with a full layer-by-layer walkthrough of the real VGG16 architecture.
A simple next step after the beginner CNN guide. Explains advanced CNN ideas in plain English, with diagrams throughout — deeper networks and the degradation problem, overfitting, dropout, batch normalization, residual connections, global average pooling, 1x1 bottleneck convolutions, CNN backpropagation through filters, feature maps, ReLU, and pooling, transfer learning, data augmentation, and modern CNN architecture patterns.
Covers the full CNN model design pipeline — from data collection and preprocessing through training — then explains when to train from scratch versus use transfer learning or fine-tuning, and finally shows how to visualize what a CNN has actually learned using Grad-CAM and activation maps.
A complete hands-on walkthrough of building, training, and evaluating a CNN on the CIFAR-10 dataset using PyTorch. Covers loading and preprocessing the dataset, defining a CNN architecture, writing the training loop with validation, and evaluating with accuracy, confusion matrix, and classification report.
An overview of the main real-world applications of CNNs — image classification, object detection, image segmentation, facial recognition and analysis, and optical character recognition (OCR). Explains what each task is, how CNNs are used, key architectures, and real-world examples.
A beginner-friendly introduction to Recurrent Neural Networks. Learn why normal neural networks cannot handle sequences, how an RNN reads data step by step, and how to build one in PyTorch using simple everyday examples.
A beginner-friendly guide to LSTM and GRU — the improved versions of RNN that can remember things from much longer ago. Learn why basic RNNs forget, how LSTM gates solve this, and how to use both in PyTorch.
A beginner-friendly guide to practical RNN applications. Learn how to use LSTM for time series forecasting, sequence classification, and sequence labelling with clear step-by-step Python examples.