| Day | Topic | Goal |
|---|---|---|
| 1 | What is Deep Learning? | Introduction, history, real-world examples, first neuron intuition, forward propagation of single neuron |
| 2 | Neurons, Weights, Bias & Activations | Deep dive into neuron structure, weights, bias, activation functions, visualizations |
| 3 | Forward Propagation | Manual calculations, numpy implementation, ReLU, Sigmoid, Tanh |
| 4 | Tiny Neural Network | Build 2-layer network from scratch in numpy |
| 5 | Loss Functions | MSE, Cross-Entropy, small examples |
| 6 | Optimizers | Gradient Descent intuition, learning rate, update rules, simple code |
| 7 | Mini Exercise | Manual neuron calculations, small network experiments |
| 8 | Phase Summary | wrap-up, code + visuals |
| Day | Topic | Goal |
|---|---|---|
| 9 | Gradient & Derivative Intuition | Calculus review, small examples |
| 10 | Backpropagation Basics | Manual chain rule calculations |
| 11 | Backpropagation in Numpy | Tiny network training step-by-step |
| 12 | Full Forward + Backprop Example | Numpy mini training loop |
| 13 | PyTorch Introduction | Tensors, basic operations, GPU usage |
| 14 | PyTorch Neural Network | Build first simple model |
| 15 | Training loop in PyTorch | Forward + loss + backward + optimizer step |
| 16 | Overfitting vs Underfitting | Visualize loss curves, concept explanation |
| 17 | Validation & Test split | Data handling in PyTorch, metrics |
| 18 | Hyperparameters | Learning rate, epochs, batch size, grid search intuition |
| 19 | Phase Summary | Notebook wrap-up, code + visuals |
| Day | Topic | Goal |
|---|---|---|
| 20 | CNN introduction | Filters, stride, padding, convolution example |
| 21 | CNN Layers | Pooling, flatten, fully connected layers |
| 22 | CNN in PyTorch | Build simple CNN for MNIST |
| 23 | CNN training | Forward + backward + optimizer, visualize filters |
| 24 | RNN introduction | Sequence data, hidden state, unrolling |
| 25 | LSTM & GRU | Why LSTM > RNN, gates explanation |
| 26 | Simple RNN in PyTorch | Manual sequence prediction |
| 27 | LSTM example | Text sequence prediction |
| 28 | NLP preprocessing | Tokenization, embedding, padding |
| 29 | RNN mini project | Predict sentiment on small dataset |
| 30 | CNN + RNN comparison | When to use which, pros/cons |
| 31 | Regularization in CNN/RNN | Dropout, batch norm, visualization |
| 32 | Hyperparameters for CNN/RNN | Learning rate, optimizer tuning |
| 33 | Phase Summary | Notebook wrap-up with multiple small examples |
| Day | Focus | Goal |
|---|---|---|
| 34 | CNN Regularization | Dropout placement, BatchNorm behavior (train vs eval), overfitting control |
| 35 | CNN Weight Initialization | Xavier vs He, dead ReLU problem, empirical comparisons |
| 36 | CNN Optimizers | SGD vs Adam vs AdamW vs RMSProp, convergence tradeoffs |
| 37 | CNN Learning Rate Scheduling | StepLR, ReduceLROnPlateau, CosineAnnealing, LR intuition |
| 38 | CNN Data Augmentation | Geometric & color transforms, over/under-augmentation risks |
| 39 | CNN Multi-Class Classification | Softmax, CrossEntropy, class imbalance, top-k accuracy |
| 40 | CNN Multi-Label Classification | BCEWithLogits, threshold tuning, PR tradeoffs |
| 41 | CNN Evaluation & Debugging | Confusion matrix, ROC/PR curves, error analysis |
| 42 | CNN Early Stopping & Checkpointing | Validation-driven stopping, best-model saving |
| 43 | CNN Hyperparameter Tuning | Batch size–LR coupling, weight decay, controlled experiments |
| 44 | CNN Advanced Mini Project | Apply all CNN techniques end-to-end (no shortcuts) |
| Day | Topic | Goal |
|---|---|---|
| 45 | Intro to Model Deployment | Understand research vs production, what “serving a model” means, and full pipeline (train → save → serve → predict) |
| 46 | Saving & Loading Models | Use torch.save / torch.load, save best CNN weights, load in separate script, verify correctness |
| 47 | Inference Pipeline | Build strict function: image → preprocessing → model → prediction (no training code allowed) |
| 48 | FastAPI Basics | Learn routes, request/response cycle, use /docs UI for testing endpoints |
| 49 | CNN + FastAPI Integration | Wrap inference pipeline into /predict endpoint, test with real image upload |
| 50 | Pydantic & Error Handling | Validate inputs (image only), handle bad inputs cleanly, return proper JSON responses |
| 51 | Docker Basics | Learn containers, write Dockerfile, build image, run container locally |
| 52 | Production Containerization | Add FastAPI + model into Docker, run with Uvicorn (production-style server) |
| 53 | Stress Testing & Performance | Test multiple inputs, edge cases, batch inference, CPU vs GPU performance |
| 54 | Project Finalization | Clean code, structure repo, write README, add usage guide & demo outputs |
| Day | Topic | Goal |
|---|---|---|
| 55 | Detection Fundamentals | Understand classification vs detection, bounding boxes, IoU, NMS, mAP metric |
| 56 | Dataset Selection & Preparation | Choose dataset (COCO subset / Pascal VOC / custom), organize images & annotations, train/test split |
| 57 | CNN Backbone Review | Review your CNN knowledge, understand feature maps, anchors, and why CNNs are backbone of detectors |
| 58 | Pretrained Detection Models | Explore YOLOv5 / YOLOv8 inference on sample images, visualize predictions, understand confidence thresholds |
| 59 | Fine-tuning Detection Model | Load pretrained model, fine-tune on small custom dataset, adjust anchors and learning rate |
| 60 | Advanced Training Tricks | Use augmentation (flips, scale, color), early stopping, learning rate scheduling for detection |
| 61 | Evaluation & Metrics | Compute mAP, IoU; analyze errors, visualize false positives/negatives |
| 62 | Multi-class / Multi-object Scenarios | Handle multiple objects per image, class imbalance, threshold tuning |
| 63 | Optimization & Inference | Batch inference, GPU utilization, speed-memory tradeoffs, confidence calibration |
| 64 | Detection Project | Build a mini project: custom dataset, trained model, evaluation, visualizations |
| 65 | Deployment | Wrap detection model in a simple FastAPI endpoint locally, allow image upload → detection output |
| 66 | Portfolio Polish | Prepare notebook for portfolio: clean code, explanations, results, plots |