Deep Learning Lab – CS3807

This repository contains all lab experiments for the Deep Learning Laboratory (CS3807), Shiv Nadar University Chennai, AY 2026–27.

Each experiment is maintained in its own subfolder, with a dedicated README, source code, dataset information, dependency list and execution instructions.

Experiments

#	Experiment	Description	Link
1	Single Layer Perceptron	Binary classification on the Banknote Authentication dataset using a perceptron implemented from scratch	Lab-1-perceptron
2	Multi-Layer Perceptron	Multi-class image classification on the Fashion-MNIST dataset using an MLP with automated hyperparameter optimization	Lab-2-MLP
3	CNN	understand the working principle of Convolutional Neural Networks by implementing convolution, pooling, feature map visualization, and image classification using TensorFlow/Keras.	Lab-3-MLP
4	Transfer Learning	The objective of this experiment are Study the evolution of deep CNN architectures, Compare LeNet-5, AlexNet, VGG16, GoogleNet and ResNet,Understand transfer learning,Fine tune pretrained CNN models,Compare classification performance of different architectures	Lab-4-Transfer-Learning
5	study-on-CNN-Training	This experiment provides a comprehensive study of Convolutional Neural Network training and optimization using the Oxford-IIIT Pet dataset with images resized to 224×224×3. Students investigate weight initialization, regularization, Batch Normalization, Dropout, optimization algorithms, CNN hyperparameters, transfer learning, and fine-tuning using the MobileNetV2 architecture. Finally, 5-fold cross-validation is used to select the best-performing configuration, followed by evaluation on an independent test set.	Lab-5-study-on-CNN-Training
6	RNN,LSTM-and-GRU	This experiment provides an end-to-end study of sequence learning using RNN, LSTM, and GRU architectures. Students preprocess and represent sequential data, implement the three recurrent models, analyze their training behavior, and evaluate their performance using appropriate classification metrics and visualizations. The experiment also introduces the use of CNN-extracted features with recurrent networks for video understanding and demonstrates sequence-to-sequence learning. Students compare the models based on predictive performance, model complexity, and computational cost, and draw conclusions from the experimental results.	Lab-6-RNN,LSTM-and-GRU
7	Autoencoder,Denoising-Autoencoder,VAE	This experiment provides an end-to-end study of autoencoders for image representation, reconstruction, denoising, and generation. Students implement and evaluate fully connected autoencoders, convolutional autoencoders, denoising autoencoders, and variational autoencoders using image data. The experiment focuses on latent representations, reconstruction quality, denoising performance, and the generative capability of VAEs through quantitative evaluation and visual analysis.	Lab-7-Autoencoder,Denoising-Autoencoder,VAE
8	Effect of Attention Type and Attention Position in a Pretrained ResNet50	To understand how different attention mechanisms modify a pretrained convolutional network, and how the position at which attention is inserted influences image classification performance, using ResNet50 pretrained on ImageNet as the single backbone throughout. Specifically, to implement channel attention, spatial attention, sequential channel–spatial attention, a lightweight self-attention module, and four established attention mechanisms (SE, ECA, CBAM, Coordinate Attention); to insert them at different candidate positions (P2, P3, P4) within the backbone; and to compare every resulting model under an identical training/evaluation protocol, both quantitatively (accuracy, precision, recall, macro F1, parameters, inference time) and qualitatively (Grad-CAM, internal attention maps).	Lab-8-Effect-of-Attention-Type-and-Attention-Position-in-a-Pretrained-ResNet50

More experiments will be added here as the semester progresses.

Repository Structure

deep-learning-lab/
├── README.md
├── experiment-1-perceptron/
│   ├── README.md
│   ├── requirements.txt
│   ├── Lab1_perceptron.ipynb
│   └── data_banknote_authentication.txt
├── experiment-2-mlp/
│   ├── README.md
│   ├── requirements.txt
│   └── Lab_2_MLP.ipynb
├── experiment-3-CNN/
│   ├── README.md
│   ├── requirements.txt
│   └── Lab3.ipynb
├── experiment-4-Transfer-Learning/
│   ├── README.md
│   ├── requirements.txt
│   └── Lab_4.ipynb
├── experiment-5-study-on-CNN-Training/
│   ├── README.md
│   ├── requirements.txt
│   └── Lab_5.ipynb
├── experiment-6-RNN,LSTM-and-GRU/
│   ├── README.md
│   ├── requirements.txt
│   └── Lab_6.ipynb
├── experiment-7-Autoencoder,Denoising-Autoencoder,VAE/
│   ├── README.md
│   ├── requirements.txt
│   └── Lab_7.ipynb
├── experiment-8-Effect-of-Attention-Type-and-Attention-Position-in-a-Pretrained-ResNet50/
│   ├── README.md
│   ├── requirements.txt
│   └── Lab_8.ipynb

General Notes

• Each experiment subfolder is self-contained: it can be cloned, its dependencies installed, and its notebook run independently of the others.
• Refer to the README inside each experiment's folder for objective, methodology, results and execution instructions specific to that experiment.