Neural Networks A Classroom Approach By Satish Kumar.pdf 'link' -

The policy network was trained using a dataset of human-played games, while the value network was trained using a combination of human-played games and self-play games generated by AlphaGo.

| Week | Topics | Practical Activity (Code) | |------|--------|----------------------------| | 1 | Neuron model, activation functions | Implement a single neuron in Python | | 2 | Perceptron learning | Code AND/OR gate training | | 3 | MLP architecture & backprop (derivation) | Hand-compute one epoch of XOR | | 4 | Backprop coding | Write a 2-layer net from scratch | | 5 | Momentum, learning rate tuning | Visualize error surfaces | | 6 | Hopfield networks | Store/recall patterns (digits) | | 7 | Self-organizing maps | Cluster colors in an image | | 8 | RBF networks | Function approximation | | 9 | Review & exam-style problems | Build a small classifier (e.g., iris) | | 10 | Final project from book’s appendix | Document and present results | Neural Networks A Classroom Approach By Satish Kumar.pdf

While many texts focus predominantly on supervised learning, Kumar gives substantial weight to unsupervised learning paradigms. The chapters on are particularly noteworthy. The explanation of competitive learning and the formation of topological maps is handled with clear examples, offering students insight into how networks can learn patterns without labeled data. The policy network was trained using a dataset