Program Structure

Module 1: Deep Learning Foundations

• Tensors, activations, and network building blocks.
• Forward/backward pass and gradient descent.
• Loss functions and metrics (classification vs regression).
• Overfitting, regularization, and validation strategy.

Module 2: Training Neural Networks (Practical Skills)

• Data loaders, batching, augmentation (overview).
• Optimizers: SGD, Adam; learning rate schedules.
• Initialization, normalization, dropout.
• Hyperparameter tuning and error analysis.

Module 3: CNNs for Computer Vision

• Convolutions, pooling, and feature maps.
• Image classification and transfer learning.
• Model evaluation and confusion matrix.
• Intro to detection/segmentation concepts.

Module 4: Sequence Models and NLP

• Word embeddings and text preprocessing.
• RNN/LSTM concepts (overview) and limits.
• Transformers and attention (high-level).
• Fine-tuning pretrained models (intro workflow).

Module 5: Generative Deep Learning (Overview + Workflow)

• Autoencoders and latent space intuition.
• GAN concepts and training instability (overview).
• Diffusion basics: denoising and sampling (high-level).
• LLM workflows: prompting, adapters, RAG basics (overview).

Module 6: Model Debugging and Reliability

• Common failures: leakage, imbalance, label noise.
• Calibration and threshold tuning concepts.
• Bias, fairness, and safe evaluation basics.
• Interpretability basics: saliency/feature importance (intro).

Module 7: Deployment and MLOps Basics

• Saving/loading models and reproducible environments.
• Serving: simple API and batch inference.
• Monitoring: drift, latency, and errors.
• Cost control: model size, quantization concepts (intro).

Module 8: Capstone Build Sprint

• Project planning: dataset, baseline, KPIs.
• Training + evaluation + iteration.
• Final report and presentation.
• Portfolio packaging: GitHub + demo + slides (optional).

Final Project

• Choose one track: vision, NLP, or generative AI.
• Deliverables: trained model + evaluation + inference demo + report.
• Optional: deploy as a simple app/API.