ML Paper Summary

Learning objectives

By completing this assignment, you will:

• Read and understand a highly influential machine-learning paper in detail.
• Summarise complex research clearly and concisely.
• Practise writing and formatting a short conference-style paper.
• Evaluate a paper’s methods, evidence, contributions, and limitations.
• Explain and critically discuss the complete paper in an individual conversation.

Papers, instructors, and booking links

Choose exactly one paper. Each paper is assigned to one instructor, who will conduct your individual paper discussion.

Book an appointment only with the instructor assigned to your paper. Do not reserve appointments with multiple instructors. If no suitable appointments remain for a paper, choose another paper with available appointments.

Paper	Why important?	Instructor	Book a discussion
Attention Is All You Need (2017)	Introduced the Transformer—the basis of modern LLMs.	Dmytro Fishman	Book with Dima
U-Net (2015)	Established a foundational architecture for image segmentation.	[Instructor]	[Calendly link]
AlexNet (2012)	Triggered the modern deep-learning revolution in computer vision.	[Instructor]	[Calendly link]
XGBoost (2016)	Made boosted trees practical, scalable, and highly effective.	[Instructor]	[Calendly link]
Generative Adversarial Nets (2014)	Introduced adversarial learning for generative models.	[Instructor]	[Calendly link]

Do not use article length as the main selection criterion :) Choose a paper whose problem, method, or historical importance genuinely interests you.

Important deadlines

1. Select a paper and book the corresponding discussion - by 1 October 2026
2. Submit the paper summary - at least 48 hours before the booked appointment
3. Attend the individual paper discussion - at the booked Calendly appointment

1. Select a paper and book a discussion

By 1 October 2026, you must:

• Choose one paper from the table above.
• Book one 20-minute appointment with the assigned instructor via the Calendly link provided in the table.
• Use your university email address when booking.
• Answer the booking-form question about why you selected the paper.

You do not need to submit separate proof of booking. When you book a meeting with an instructor, the system sends out a confirmation that we will use as proof.

If you need to reschedule, use the rescheduling link in the Calendly confirmation email. Do not create a second booking. Contact the course staff before changing your paper or instructor after the selection deadline.

2. Read and understand the complete paper

Read the complete original paper carefully. Your written summary will help you prepare for the discussion, but it is not the only material that will be discussed.

You should understand:

• The problem addressed by the paper.
• The paper’s main contribution.
• The proposed method and why it should work.
• The experimental design and principal results.
• The evidence supporting the authors’ claims.
• Important assumptions and limitations.
• Connections to concepts covered in the course.

3. Write the paper summary

Complete the written assignment individually. The summary must be a concise, self-contained account of the selected paper written in your own words.

Length and format

• Submit a PDF file only to the Homeworks page.
• The main text must be no longer than two pages.
• References, acknowledgements, and appendices do not count toward the two-page limit.
• Appendices must not be used to circumvent the page limit.
• Use a professional two-column conference-paper format.
• We recommend working in Overleaf and using an ICML-style template.

Structure

Your summary should resemble a short conference paper and normally include:

• A title making clear that the document is a summary.
• Your name and university affiliation.
• An abstract.
• An introduction.
• A logically organized account of the background, method, and results.
• Discussion of the paper’s importance and limitations.
• A conclusion.
• A reference list.

You do not need a separate section for every item. For example, background and related work may be incorporated into the introduction. Do not use vague section headings such as “Main” or “Main Body.”

Include at least 3–5 relevant references, including the original paper and appropriate background or related work.

Figures, tables, equations, and references

• Number and caption all figures, tables, equations, and algorithms where appropriate.
• Refer to every included figure or table in the main text.
• Cite any graphical or tabular material taken or adapted from another source.
• Do not use screenshots of equations, tables, algorithms, or captions. Typeset them properly.
• Make citations part of complete sentences.
• Ensure that every in-text citation has a corresponding reference-list entry.
• Use conventional academic language and formatting.
• Clearly distinguish your summary from the original authors’ work. For example: “The authors propose…” or “Vaswani et al. show…”
• Do not present the original authors’ contributions as your own.

Originality and use of generative AI

The summary must be a new text written in your own words. Do not copy unattributed sentences or passages from the original paper, other papers, summaries or websites.

You may use generative-AI tools as writing or study aids, but you remain responsible for:

• Verifying every factual and technical claim.
• Checking equations, algorithms, terminology, and references.
• Removing fabricated or irrelevant material.
• Ensuring that the final document is coherent and original.
• Understanding and being able to explain everything in the submitted summary.

A polished AI-generated summary will not compensate for an inability to discuss the paper. Academic-integrity violations will be handled under the university's academic-integrity rules.

4. Submit the written summary

Upload the summary PDF through Homeworks at least 48 hours before your booked appointment.

Use the following file-name format:

FamilyName_GivenName_PaperShortTitle.pdf

The appointment does not replace the written submission. The instructor needs the summary in advance to prepare for the discussion.

5. Attend the individual paper discussion

The individual discussion will take approximately 20 minutes.

Your submitted summary will provide a starting point, but the complete original paper is the subject of the discussion. Be prepared to answer questions about material that does not appear in your two-page summary.

The discussion may include questions such as:

• What problem does the paper solve, and why is it important?
• What is the paper’s central idea?
• How does the proposed method work?
• Why did the authors make particular architectural or methodological choices?
• Which results are most important, and do they support the conclusions?
• What assumptions does the method make?
• What are the paper’s weaknesses or limitations?
• How does the work relate to concepts discussed in the course?
• What might happen if an important component or assumption were changed?
• How did this paper affect subsequent machine-learning research?

You may be asked to clarify statements, figures, equations, or claims from your summary and the original paper. The aim is to demonstrate genuine understanding, not to recite memorized text.

Grading

The assignment is worth 15 points.

Paper selection and appointment booking are worth 1 point. Written summary will give you 3 points, depending on its accuracy, relevance, clarity, structure and so on. The individual paper discussion is worth the remaining 11 points.

The written summary is important preparation, but the individual discussion is the principal assessed component. A well-formatted summary cannot compensate for a lack of understanding of the original paper. Conversely, small imperfections in the written summary need not prevent a strong overall grade when the student demonstrates excellent understanding during the discussion.

Final checklist

Before considering the assignment complete, confirm that you have:

• Selected one paper.
• Booked exactly one appointment with the corresponding instructor by 1 October 2026.
• Read the complete original paper.
• Submitted the two-page summary PDF at least 48 hours before the appointment.
• Attended the booked individual discussion about the contents of the paper and the summary.