Deep Learning
Laboratory Manual
A comprehensive reference manual containing 4 structured hands-on experiments for Deep Learning. Each experiment details theoretical background, implementation code, standard observation parameters, and viva questions.
Course Experiments Catalog
SEC-01 / LABSLearning the XOR Function using Deep Feedforward Neural Networks
Design and implement a Multi-Layer Perceptron (MLP) in Keras to learn the non-linear XOR function using gradient-based learning and hidden units.
Regularization Techniques for Deep Generalization
Implement and study L1/L2 parameter norm penalties, dataset augmentation, dropout, and noise injection on the MNIST dataset using Keras.
Optimization Algorithms Comparison on Toy Dataset
Implement and compare the convergence performance of Stochastic Gradient Descent (SGD), Momentum, and Adam optimizers on a non-linear toy dataset.
Convolutional Neural Networks (CNN) on MNIST
Design and train a Convolutional Neural Network on the MNIST handwritten digit dataset to study convolution filters, max pooling, and dense classification layers.
Learning Outcomes
CO/PO- Understand multi-layer perceptron architectures and implement gradient descent optimization from scratch.
- Analyze and apply parameter norm regularization (L1/L2) to prevent overfitting.
- Design and train convolutional (CNN) and recurrent (RNN/LSTM) networks for vision and sequential tasks.
Lab Environment Setup
ENV- TensorFlow 2.x / Keras
- Google Colab
- Python 3.10+
- NumPy / Pandas