BTech Semester 4 · Department of Computer Science

Deep Learning
Lab Practicals

10 hands-on notebooks covering neural networks, CNNs, RNNs, and modern DL applications

10 Practicals

Google Colab Ready

Python · TensorFlow · NumPy

CO1 – CO5 Mapped

10

Lab Notebooks

4

Course Parts

80+

Viva Questions

5

Course Outcomes

15+

Datasets Used

💡

How to open these notebooks: Download any .ipynb file → go to colab.research.google.com → click File → Upload notebook → select the file. All notebooks include a built-in dataset fallback so they run immediately — no Kaggle account needed. For real datasets, follow the Option A instructions inside each notebook.

Part 01

Foundations & Optimization

The Math of Neural Networks

Implement Backpropagation from scratch using only NumPy. Solve the XOR problem and visualize the learned decision boundary.

NumPy Only Backpropagation XOR Dataset CO1

Download Notebook

Deep MLP for Customer Churn Prediction

Build a 4-layer MLP for binary classification. Master one-hot encoding, feature scaling, and industry-standard evaluation metrics.

MLP Precision / Recall / F1 Bank Churn Dataset CO1CO5

Download Notebook

The Optimizer Showdown

Compare SGD, Momentum+Nesterov, and Adam on Fashion MNIST. Plot convergence curves and understand loss landscape geometry.

SGD · Momentum · Adam Convergence Analysis Fashion MNIST CO2

Download Notebook

Part 02

Unsupervised Learning & Regularization

PCA vs. Autoencoders

Compare linear (PCA) vs neural (Autoencoder) dimensionality reduction on MNIST. Visualize and compare 2D latent space cluster separability.

PCA Undercomplete Autoencoder MNIST CO2CO3

Download Notebook

Denoising Autoencoders

Train a convolutional DAE to restore images corrupted by Gaussian noise. Evaluate with MSE and PSNR across multiple noise levels.

Conv Autoencoder Gaussian Noise · PSNR MNIST + Synthetic Noise CO3

Download Notebook

Taming Overfitting

Apply Dropout, L2 regularization, Batch Normalization, and Early Stopping iteratively on CIFAR-10. Watch the overfit gap shrink at each step.

Dropout · L2 · BatchNorm Bias–Variance Tradeoff CIFAR-10 (small subset) CO3

Download Notebook

Part 03

Deep Learning Models

LeNet-5 and AlexNet Architectures

Implement LeNet-5 (1998) and AlexNet (2012) layer by layer. Visualize feature maps, architecture diagrams, and parameter count comparison.

CNN Architectures Feature Maps · Conv Filters MNIST · CIFAR-10 CO4

Download Notebook

Transfer Learning with VGG16 / ResNet

Load pre-trained ImageNet models. Freeze backbone, fine-tune top layers, and classify medical X-ray images. Compare scratch vs transfer learning.

VGG16 · ResNet50 Feature Extraction · Fine-Tuning Chest X-Ray (Pneumonia) CO4CO5

Download Notebook

Time Series with RNN vs LSTM

Predict stock prices using the Windowing method. Visualize LSTM gates, vanishing gradient problem, and compare RNN vs LSTM with MSE and RMSE.

LSTM · GRU · BPTT Windowing · Forecasting Google Stock Price CO4

Download Notebook

Part 04

Applications & Attention

Image Captioning with Attention

Build a full Encoder-Decoder pipeline. InceptionV3 extracts image features; a GRU decoder with Bahdanau Attention generates captions word by word.

Encoder–Decoder Bahdanau Attention Teacher Forcing · Beam Search Flickr8k CO3CO5

Download Notebook