Material of the course "Exploring the Frontier of Database Research: the Alliance with Foundation Models" held by Matteo Paganelli (Postdoc at the University of Modena and Reggio Emilia) within the ICT doctoral school program.

Program

Module 1: Introduction to Foundation Models (FMs) - 3h

• 1.1 The transformer architecture
• 1.2 Hands-on transformers with HuggingFace's Transformers library

Module 2: FMs for data preparation - 3h

• 2.1 The data integration pipeline
• 2.2 State-of-the-art methods for data preparation

Module 3: Tabular FMs - 2h

• 3.1 Introduction to tabular FMs
• 3.2 Hands-on tabular FMs

Module 4: In-context learning in data preparation tasks - 2h

• 4.1 Applications
• 4.2 Entity Matching with the OpenPrompt library (PromptEM)

Module 5: FMs limitations & next research directions - 2h

• 5.1 Hands-on Retrieval-Augmented Generation (RAG) with LangChain
• 5.2 FMs limitations
• 5.3 Next research directions

Exam (3 CFD)

There is no deadline, you can take the exam whenever you want.

• Option A: Select 3 papers for a given topic and create a report or presentation summarizing them and highlighting their similarities/differences
  • The list of papers is available at https://github.com/pagamatteo/db_llm_alliance/blob/main/exam/Papers.csv
• Option B: Experiment with the code of a SOTA approach from https://github.com/pagamatteo/db_llm_alliance/blob/main/exam/Paper_codes.csv by …
  • Applying the model on a new dataset (use the related works of the paper to find datasets that were not used in the paper's experimentation)
  • Testing a different transformer model and comparing the performance with the original ones
  • Wrapping the repository code in a Google colab notebook (this activity is more about refactoring and simplifying the code making it accessible in an interactive environment to show the main features of the original approach)
  • Select only one of the previous points
• Option C: Develop your own project related to a data preparation task (e.g., entity matching, schema matching, data imputation, error detection, simple data transformations, question answering, claim verification, table understanding, and other table-related tasks) and including some of the libraries discussed in the course (e.g., LangChain, OpenPrompt, Transformers). Here are some examples:
  • Create a chatbot with LangChain that interacts with PDF documents: https://python.langchain.com/v0.2/docs/how_to/document_loader_pdf/
  • Create a chain with LangChain that uses LLM Agents: https://python.langchain.com/v0.2/docs/tutorials/agents/
  • Create a chain with LangChain that interacts with a DB in SQL: https://python.langchain.com/v0.2/docs/tutorials/sql_qa/
  • ... (open to suggestions)