Project AI for Medical Imaging (8BE130)

Table of Contents

1. Course Overview
   • Course Files
2. Logistics & Teaching Methods
   • Workload
   • Flipped Classroom Model
   • Work in Groups
   • Conflict Resolution & Individual Grading
3. Course Outline & Schedule
   • Weeks 2-5 lecture structure
   • Week 6 lecture structure (The Pitches)
4. Deliverables & Submission
   • Deliverables for Assignments 1 & 2
   • Deliverables for the Open Project
   • Submission Mechanics
5. Grading
   • Assessment Rubric for Assignments 1 & 2
   • Assessment Rubric for the Open Project

---

1. Course Overview

The course consists of two types of assignments with distinct goals:

• Assignments 1 & 2 (Skill Building): You will learn foundational concepts including Transformers, GPT architectures, and Vision-Language models through structured tasks.
• Open Project (Application): You will apply this accumulated knowledge in an open setting to design experiments and answer a specific research question.

Course Files

There are four files that contain all instructions for the course:

File	Description
README.md	The file that you are reading now, containing course information, logistics, and guidelines
assignment_1_GPT.ipynb	Jupyter notebook for Assignment 1 focusing on GPT language models
assignment_2_VLM.ipynb	Jupyter notebook for Assignment 2 focusing on vision-language models
open_assignment.md	Instructions for the open-ended project component of the course

Detailed instructions, theory questions, and code skeletons are provided directly within the Jupyter notebooks. Additionally, there are links to videos, papers and other recommended reading materials that you will find helpful in your work.

We will also use Canvas for communication and assignment submission.

2. Logistics & Teaching Methods

Workload

The course is worth 5 ECTS. You are expected to spend 28 hours per ECTS (140 hours total) on preparation and assignment work.

Flipped Classroom Model

This is a project-based course. You are expected to work independently and proactively within your groups. We do not provide step-by-step instructions; you must navigate technical challenges and make methodological decisions under the guidance of TAs and instructors.

The course has one kick-off meeting and five lectures. The lectures are in a flipped classroom setting. In this course, "flipped classroom" means you read the papers and other materials, and watch the videos at home as a preparation for the lecture. During the lecture, you actively discuss questions and topics about the material with other groups. You do not sit and listen to a lecture.

Work in Groups

All assignments are performed in groups of 4 students via self-assignment in Canvas. Groups are divided into two sets (Odd-numbered groups and Even-numbered groups) with rotating roles:

• Clients: Responsible for identifying gaps in understanding. Prepare theory and coding/implementation questions to pose to Consultants during the flipped classroom.
• Consultants: Support the discussion. Review the suggested material to prepare to brainstorm, answer Client questions, and exchange resources during the flipped classroom.

Conflict Resolution & Individual Grading

Professional collaboration within the groups is expected. If significant imbalance occurs, approach the TA/instructor immediately. Upon request, a formal peer review will assess individual contributions, potentially resulting in individual grades for students differing from the group average.

3. Course Outline & Schedule

For specific deadline dates, please check Canvas.

Week	Activity	Deadlines	Odd-numbered Groups Role	Even-numbered Groups Role
Week 1	Kick-off meeting
Week 2	Assignment 1 flipped classroom		Clients	Consultants
Carnival	🎭 🎪 🍺
Week 3	Assignment 1 flipped classroom	Assignment 1 due	Consultants	Clients
Week 4	Assignment 2 flipped classroom		Consultants	Clients
Week 5	Assignment 2 flipped classroom	Assignment 2 due, Pitch slide due	Clients	Consultants
Week 6	Open assignment flipped classroom		Pitch open assignment	Pitch open assignment
Week 7	Office hours / no lecture
Week 8	Office hours / no lecture
Exam week 1
Exam week 2		Open assignment deadline

Weeks 2-5 lecture structure

Each flipped lecture follows a structured Peer Consultancy model designed to resolve specific blockers and deepen theoretical understanding.

Introduction (10 min.)

Part 1: The Consultancy (35 min)

• Pair Up: One Client group pairs with one Consultant group.
• Discussion:
  • Clients present their prepared questions or coding/implementation blockers.
  • Consultants use their prepared notes to brainstorm solutions, explain concepts, or point to specific resources.
• The "Unsolvables": If the paired groups cannot agree on an answer or solve a specific error, they must write this down as a "Priority Question."

Part 2: Plenary Synthesis (45 min)

• Escalation: The instructor collects the "Priority Questions" identified in Part 1.
• Expert Review: The instructor addresses these high-level blockers, clarifies common misconceptions, and provides deep-dive explanations on the most difficult theoretical concepts encountered during the week.

Week 6 lecture structure (The Pitches)

The pitch session follows a structured Formative Review model designed to validate research questions and methodological plans before deep work begins.

Introduction (10 min.)

Part 1: The Pitch Rounds (70 min)

• The Stage: Groups present sequentially using the slides compiled on the instructor's computer.
• Process:
  • Pitch (2 min): The group presents their single slide outlining the research question and proposed methodology.
  • Feedback (3-5 min): Instructors and peers act as "critical friends" to identify scope risks, suggest simplifications, or confirm feasibility.
• Strict Timing: A hard stop is enforced to ensure every group receives equal feedback time.

Part 2: Synthesis & Next Steps (10 min)

• Green Light: Instructors confirm which projects are approved to proceed and which require a revision.
• Common Patterns: A summary of recurring issues observed across the pitches to guide the final project phase.

4. Deliverables & Submission

For Assignments 1 and 2

1. Exercises: Complete answers to the practical exercises.
2. Flipped Classroom Log: A log of the preparation and participation in the flipped classroom.

ℹ️ Both of these can be filled at the end of the Jupyter notebook and you can submit that in Canvas.

For the Open Project

• Scientific Poster: A scientific poster summarizing research questions, experiments, and results.

Submission Mechanics

Submit one set of deliverables per group via Canvas. Punctual submissions are expected, though small delays for valid technical issues (e.g., code crashes) are not penalized. Inform the instructors immediately of significant delays.

5. Grading

The final course grade is based on three components:

• Assignment 1:
  • 20% answers to practical exercises
  • 5% flipped classroom log
• Assignment 2: 25%
  • 20% answers to practical exercises
  • 5% flipped classroom log
• Open Project: 50%

Assessment Rubric for Assignments 1 & 2

Component	Insufficient	Satisfactory	Excellent
Understanding (Theory)	Descriptions of methods are copied/pasted or incorrect; fails to convey understanding.	Descriptions are written in own words and show good understanding of the material.	Demonstrates mastery of the material; makes insightful connections between different methods or theories.
Code Quality	Code crashes or has errors; lack of documentation; unstructured "spaghetti code."	Code is self-contained and runs without errors; basic documentation provided; easy to reproduce results.	Code is user-friendly, modular, and well-structured; detailed documentation.
Results & Analysis	Plots/Results are missing, mislabeled, or confusing; no analysis provided.	Clear plots with captions; correctly interpreted results (e.g., loss curves).	Professional visualizations; insightful analysis of why the model behaved that way (e.g., overfitting signs).
Flipped Classroom Logs	Group was absent or logs are missing, incomplete, or trivial.	Logs are complete and clearly document the questions asked as Client and resources/advice shared as Consultant.	Logs demonstrate deep engagement; questions were insightful and Consultant contributions added significant value.

Assessment Rubric for the Open Project

Component	Insufficient	Satisfactory	Excellent
Research Design (Methodology)	Research question is unclear or trivial; methodological choices are poor or unjustified.	Clearly defined research question; methodological choices are correct and well-motivated.	Novel or challenging research question; use of state-of-the-art methodology; deep technical insight.
Experiments	Experimental setup is unstructured or minimal; fails to answer the research question.	Experiments are systematic and directly address the research question; results are reliable.	Comprehensive set of experiments that thoroughly validate the hypothesis; rigorous testing.
Code Quality	Code crashes or has errors; lack of documentation.	Code is self-contained and runs without errors; basic documentation provided.	Code is user-friendly, modular, well-structured, and version-controlled (e.g., GitHub).
Poster Content (Intro/Methods)	Introduction is missing or irrelevant; methods are unclear.	Clearly describes the research question, goal, and methods used.	Compelling "big picture" context; methods are explained clearly but concisely.
Poster Design & Language	Cluttered layout; poor flow; grammar errors impede understanding; inconsistent referencing.	Balanced layout; clear line of thought; professional language; consistent referencing.	Excellent flow and visual narrative; polished layout (TU/e style); highly professional and compelling.
Results & Discussion	Results are hard to interpret; analysis is missing or superficial.	Clear visualization of results; discussion of strengths/weaknesses is present.	Self-contained visualizations that highlight key findings; critical evaluation and discussion of future directions.