AI417UPM/A4_4320154_FaresAldeeb

AI417 – Assignment 4 (Swiss Knife)

Student Information

• Name: Fares Aldeeb
• UPM ID: 4320154
• Course: AI417

---

Repository Contents

This repository contains two trained neural network models from Assignment 3:

• DummyNet
• VanillaNet

Both models are fully compatible with torch.hub.load().

Files included:

• dummy.py → Dummy model architecture
• vanilla.py → Vanilla model architecture
• dummy-weights.bin → Trained weights for DummyNet
• vanilla-weights.bin → Trained weights for VanillaNet
• hubconf.py → TorchHub configuration file

---

Model Description

1️⃣ DummyNet

• Fully connected feedforward network
• Architecture:
  • 784 → 512 → 256 → 128 → 10
• Xavier weight initialization
• No activation functions between layers

2️⃣ VanillaNet

• Fully connected feedforward network
• Architecture:
  • 784 → 512 → 256 → 128 → 10
• ReLU activation after each hidden layer
• Kaiming weight initialization

Both models are trained on the FashionMNIST dataset.

---

How to Load the Models

Example usage:

from huggingface_hub import snapshot_download
import torch as tr

repo_id = "AI417UPM/A4_4320154_FaresAldeeb"

repo_path = snapshot_download(repo_id)

# Load DummyNet
dummy = tr.hub.load(repo_path, "DummyNet", source="local", pretrained=True)
dummy.eval()

# Load VanillaNet
vanilla = tr.hub.load(repo_path, "VanillaNet", source="local", pretrained=True)
vanilla.eval()

# Example forward pass
output = dummy(tr.randn(1, 784))
print(output.shape)