Basic Information

• WiFi Credentials :

  SSID: IIIT-Guest

  WPA/WPA2 Key: I%GR#*S@! (case sensitive)

• Google Groups :

CVIT Summer School 2023 - Google Groups

Day 0 : Preparatory

Homework(Optional)

Problem Statement:

The objective of this experiment is to train a neural network on the CIFAR-100 dataset. The network will be trained by exploring the impact of different parameters, such as learning rate, activation functions, and number of layers, on its performance.

Specifically, the problem entails the following steps:

• Design and implement a neural network architecture suitable for the CIFAR-100 dataset, allowing flexibility in choosing the number of layers and activation functions.
• Experiment with various learning rate values (e.g., 0.001, 0.01, 0.1) to observe their impact on the network's training and validation accuracy.
• Evaluate the performance of the network by progressively increasing the number of layers (e.g., 2, 4, 6 layers), studying how it affects accuracy and potential overfitting.
• Explore different activation functions, such as ReLU, sigmoid, and tanh, to observe their influence on the network's learning capabilities and overall performance.

The primary goal of this experiment is to analyze the effects of these parameter variations on the network's ability to learn and classify the CIFAR-100 dataset accurately. By training and evaluating the network with different combinations of activation functions, learning rates, and layer depths, we aim to identify the optimal configuration that yields the highest accuracy and generalisation performance on this challenging image classification task.

Day 1 : Fundamentals