Prerequisites Basic computer literacy and a willingness to document experiments.
Learning objectives Explain the stage-specific model and its limits Apply the listed concepts using a controlled exercise Record results, failure cases and a reproducible evidence trail
Topics to learn Vectors, matrices, tensors and broadcasting Computational graphs and gradients Loss functions and chain rule Numerical precision and deterministic seeds Neural Networks specialization: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Tools NumPy PyTorch tensors Jupyter
Hands-on exercise Implement a tiny differentiable computation and compare gradients with finite differences. Context for this route: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Suggested project A notebook explaining tensor shapes, gradients and numerical checks. The outcome must directly demonstrate Neural Networks, not merely list tools.
Learning resources Completion criteria Tensor dimensions and gradient flow are correct and independently verified. Evidence must be reproducible and tied to Neural Networks.
Next: Neural network building blocks
✓ Mark this node as complete
Prerequisites Completion of the previous stage and a working local practice environment.
Learning objectives Explain the stage-specific model and its limits Apply the listed concepts using a controlled exercise Record results, failure cases and a reproducible evidence trail
Topics to learn Linear layers, activations and initialization Loss functions for classification and regression Backpropagation and gradient descent Train/validation curves Neural Networks specialization: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Tools PyTorch Matplotlib
Hands-on exercise Train an MLP on a public toy dataset and diagnose underfitting and overfitting. Context for this route: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Suggested project A baseline MLP with experiment log and learning-curve interpretation. The outcome must directly demonstrate Neural Networks, not merely list tools.
Learning resources Completion criteria The model trains reproducibly and failure modes are explained. Evidence must be reproducible and tied to Neural Networks.
Next: Optimization and training stability
✓ Mark this node as complete
Prerequisites Completion of the previous stage and a working local practice environment.
Learning objectives Explain the stage-specific model and its limits Apply the listed concepts using a controlled exercise Record results, failure cases and a reproducible evidence trail
Topics to learn SGD, momentum and adaptive optimizers Learning-rate schedules and warmup Normalization, dropout and weight decay Gradient clipping and exploding/vanishing gradients Neural Networks specialization: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Tools PyTorch optimizers TensorBoard
Hands-on exercise Run a controlled optimizer ablation while changing one variable at a time. Context for this route: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Suggested project A comparison report with fixed seeds and training configurations. The outcome must directly demonstrate Neural Networks, not merely list tools.
Learning resources Completion criteria Conclusions are based on comparable runs and held-out metrics. Evidence must be reproducible and tied to Neural Networks.
Next: Convolutional and sequence architectures
✓ Mark this node as complete
Prerequisites Completion of the previous stage and a working local practice environment.
Learning objectives Explain the stage-specific model and its limits Apply the listed concepts using a controlled exercise Record results, failure cases and a reproducible evidence trail
Topics to learn CNN blocks and receptive fields RNN/LSTM concepts and sequence batching Attention and Transformer blocks Inductive bias and compute budget Neural Networks specialization: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Tools PyTorch torchvision Transformers
Hands-on exercise Implement or adapt one CNN and one sequence model for distinct small tasks. Context for this route: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Suggested project Two task-specific models with justified architecture choices. The outcome must directly demonstrate Neural Networks, not merely list tools.
Learning resources Completion criteria Architecture selection is linked to the data structure and results. Evidence must be reproducible and tied to Neural Networks.
Next: Data pipelines and scalable experiments
✓ Mark this node as complete
Prerequisites Completion of the previous stage and a working local practice environment.
Learning objectives Explain the stage-specific model and its limits Apply the listed concepts using a controlled exercise Record results, failure cases and a reproducible evidence trail
Topics to learn Dataset versioning and leakage controls DataLoader, caching and augmentation Mixed precision and GPU memory Checkpointing and experiment metadata Neural Networks specialization: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Tools PyTorch DataLoader DVC concepts MLflow
Hands-on exercise Profile a training pipeline and identify the input or compute bottleneck. Context for this route: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Suggested project A reproducible training harness with checkpoint recovery and run tracking. The outcome must directly demonstrate Neural Networks, not merely list tools.
Learning resources Completion criteria A run can be resumed and its dataset/configuration are traceable. Evidence must be reproducible and tied to Neural Networks.
Next: Advanced architectures and transfer learning
✓ Mark this node as complete
Prerequisites Completion of the previous stage and a working local practice environment.
Learning objectives Explain the stage-specific model and its limits Apply the listed concepts using a controlled exercise Record results, failure cases and a reproducible evidence trail
Topics to learn Transfer learning and fine-tuning strategies Self-supervised and representation learning concepts Parameter-efficient adaptation Ensembling and uncertainty basics Neural Networks specialization: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Tools PyTorch Hugging Face ONNX
Hands-on exercise Fine-tune a pretrained model and compare full vs partial parameter updates. Context for this route: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Suggested project A model adaptation experiment with a compute/performance trade-off analysis. The outcome must directly demonstrate Neural Networks, not merely list tools.
Learning resources Completion criteria Evaluation is held out and resource costs are reported. Evidence must be reproducible and tied to Neural Networks.
Next: Evaluation, interpretability and deployment
✓ Mark this node as complete
Prerequisites Completion of the previous stage and a working local practice environment.
Learning objectives Explain the stage-specific model and its limits Apply the listed concepts using a controlled exercise Record results, failure cases and a reproducible evidence trail
Topics to learn Calibration and subgroup analysis Explainability and attribution limits Model export and inference optimization Drift and feedback monitoring Neural Networks specialization: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Tools Captum ONNX Runtime FastAPI
Hands-on exercise Package a model and build automated regression checks for its predictions. Context for this route: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Suggested project A tested inference service with monitoring and a model card. The outcome must directly demonstrate Neural Networks, not merely list tools.
Learning resources Completion criteria Deployment is reproducible and operational limitations are visible. Evidence must be reproducible and tied to Neural Networks.
Next: Capstone: deep-learning solution
✓ Mark this node as complete
Prerequisites Completion of the previous stage and a working local practice environment.
Learning objectives Explain the stage-specific model and its limits Apply the listed concepts using a controlled exercise Record results, failure cases and a reproducible evidence trail
Topics to learn Problem framing and baseline Experiment design and model selection Testing, performance and risk assessment Portfolio communication Neural Networks specialization: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Tools PyTorch Git Experiment tracker
Hands-on exercise Run a project from raw data to evaluated model and perform a peer review. Context for this route: Track-specific focus: Neural Networks. Use the tools and standards named in this path; avoid treating this route as a generic overview of Artificial Intelligence.
Suggested project A complete repository with data protocol, training code, report, model and demo. The outcome must directly demonstrate Neural Networks, not merely list tools.
Learning resources Completion criteria Results can be reproduced and limitations are stated without overstating accuracy. Evidence must be reproducible and tied to Neural Networks.
End of roadmap: Complete the final project and build your portfolio.
✓ Mark this node as complete