CSCE 5218 – Deep Learning

Spring 2026    

---

About Paper Review

---

Detailed Instructions About Paper Reviews:

1. Only the papers listed under "Paper review list" are qualified to be reviewed.


2. The review is required to be submitted through Canvas on the indicated due date.


3. Please name a review rigorously in the format of LastnameFirstname-XX-Review-LastNameOfFirstAuthor.pdf (only pdf file will be accepted), where XX indicates the paper number on the list. For example, OluwadareOluwatosin-01-Review-Krizhevsky.pdf would be the review I submitted for the paper 01 (on the list) authored by Krizhevsky et al.


4. Follow the required paper review format below. Reviews that do not have at least the component listed will lose lots of points.


5. Paper summaries will be graded electronically. I will update the score through Canvas. For each paper, the best score is 100.


6. Format: Please use the single-column CVPR-style template for all paper reviews. Download the .tex review template. See the example review file provided.

---

Paper Review Format (Required Sections)

Your paper review must include the following four sections.

1. Summary – 10 points

   Describe the paper’s goal, method, and results in your own words.
   
   Guiding Questions:
   • What problem does the paper address?
   • What method or architecture does it introduce?
   • What were the major results?

2. Three Key Things You Learned – 35 points

   List and explain at least three important concepts, techniques, or lessons you gained from the paper.
   
   Examples:
   • “I learned how convolution filters extract features…”
   • “I learned that residual connections help with gradient flow…”
   • “I learned why pretraining on large datasets improves downstream tasks…”
   • “Explanation of why they were new or important…”

3. New Knowledge – 20 points

   Identify ideas, terms, or methods that were new to you and describe how they expanded your understanding.
   
   Guiding Questions:
   • What concepts or techniques were unfamiliar before reading?
   • What new tools, datasets, or architectures did you discover?
   • What results or analysis surprised you?

4. Questions or Areas for Improvement – 25 points

   Discuss parts of the paper that were unclear, confusing, or could have been explained better from a student perspective.
   
   Examples:
   • “I found the mathematical notation unclear.”
   • “The dataset description was too brief.”
   • “I didn’t understand why they chose this baseline.”

---

Additional Grading Criteria (Applied Across the Review)

• Writing Quality & Following Instructions – 10 points

  Clear, well-organized writing; proper grammar; correct PDF naming and formatting; and meeting required length (minimum 400 words, maximum 2 pages). Minor typos will not affect your grade, but writing that is unclear enough to impede understanding will result in a score reduction.

---

Additional Guidelines (Read Before Submitting)

1. To get a score of 100 (best), the review should be well written. Roughly speaking, a review could be scored 100 if it read like it was carefully reviewed.


2. The minimum length of the review is 400 words and 2 Pages Maximum.


3. In general, the purpose is to show that you have really read and thought carefully about the paper.


4. General comments or umbrella comments are not recommended, infact they should be avoided. Be specific in your questions, critiques, and summary drafts.


5. In general, minor writing problems (typos or grammatical issues) won't affect the grading. But if the writing is bad enough to affect my understanding, the score will be tuned down accordingly.


6. Use of Generative AI: Our syllabus contains instructions about the limit to the use of Generative AI. Adhere strictly to it.

---