Group assignment 2

Portfolio Exercise 2: Transformer Models

Note: M3 - Group Assignment 2 Deadline: Wednesday, February 14th at 10:00 AM

Introduction

This exercise is designed to deepen your understanding and skills in modern deep learning techniques. We have two main tasks for you. The first is focused on using SBERT for semantic search, and the second involves hands-on exercises with gradient descent and the attention mechanism.

Part 1: SBERT and Semantic Search

Task Description

Create something innovative using SBERT and semantic search, or even more! The guidelines are intentionally broad to encourage creativity. Here are some ideas to get you started:

• Implement a GIF search engine or YouTube search function using images and CLIP.
• (Optional) Use SetFit for supervised tasks with SBERT models.
• Consider building a search engine using a Gradio or Streamlit app.

Part 2: Gradient Descent and Attention Mechanism Exercises

Task Description

1. Gradient Descent Exercise: Execute the process of updating weights for two examples using Stochastic Gradient Descent (SGD). Document each step, including input calculation, prediction, loss assessment, weight adjustments, and updates.

2. Attention Mechanism Exercise: Implement the attention mechanism on two distinct sentences. Choose sentences with polysmous words to demonstrate its functionality effectively.

• Workshop 2 - Transformer and Attention examples

Data

• You may utilize datasets from 🤗 Hugging Face, Kaggle, or create your own.
• For inspiration, refer to the GIF search engine and YouTube search projects.

Delivery

• Create a dedicated GitHub repository for this assignment.
• Store all relevant materials, including the Colab notebook, in the repository.
• Provide a README.md file with a concise description of the assignment and its components.
• You may work individually or in groups of up to three members.
• Submit your work by emailing a link to the repository to Hamid (hamidb@business.aau.dk).