AI417UPM/A4_4310823_Sema

FashionMNIST Model Repository - Assignment 4

This repository contains two deep learning models trained on the FashionMNIST dataset as part of the AI417 Deep Learning course at UPM (Spring 2026).

Models Included

1. Dummy Model (dummy)

• Architecture: A simple feed-forward network with layers of sizes 784 -> 512 -> 784 -> 10.
• Initialization: Xavier Uniform weights and zero biases.
• Optimizer: Stochastic Gradient Descent (SGD) with momentum 0.9 and weight decay 5e-4.

2. Vanilla Model (vanilla)

• Architecture: A deeper feed-forward network with layers 784 -> 1024 -> 512 -> 10 using ReLU activations.
• Initialization: Kaiming Normal initialization, which is optimized for ReLU layers.
• Optimizer: AdamW optimizer for faster and more stable convergence.

How to use with PyTorch Hub

You can load these models directly into your Python environment using torch.hub.

Load the Vanilla Model (Pretrained)

import torch

repo = 'AI417UPM/A4_4310823_Sema'
model = torch.hub.load(repo, 'vanilla', pretrained=True)
model.eval()