1 of 12

The Role of Math-Notation Complexity in ML Conference Publication Success

Project Pitch�Group 11

Linus A Schneider (6989196)

Leon Lemke (7003493)

Jaisidh Singh (6960379)

Anupam Sourav Patra (7077885)

2 of 12

Motivation (author’s POV)

��

Patrick Kidger (2182 citations), 11. September 2024, on x.com

3 of 12

Motivation (reviewer’s POV)

​

​

​

4 of 12

Hypothesis

“More complicated math notation in ML papers leads to a better chance of acceptance at conferences.”

5 of 12

Dataset: Collection

  • What do we use:
    • ML Papers
    • Review Data�
  • Where do we get them from:
    • ICLR 2023
    • NeurIPS 2023�
  • How do we get the data:
    • Use scraped data from OpenReview
    • Search TeX source for papers using arXiv-API

6 of 12

Dataset: Pre-Processing

︙

⃕

⃕

Extract equations

2

4

1

2

︙

Get complexity feature vector

LaTeX source

7 of 12

Analysis: Features

Independent

Dependent

Number of unique symbols across paper

Acceptance/Rejection

Maximum of representational complexity across equations

Review scores

Mean number of new symbols per equation

Novelty Score

SD representational complexity across equations

​

8 of 12

Analysis: Modelling

  1. Regression Models:
    1. Statistical analysis:�Correlations of features with Review Scores
    2. Linear Regression: �Math Notation Complexity vs Recommendation Scores
    3. Logistic Regression: �Math Notation Complexity vs Accept/Reject
  • Stratified Analysis:
    • By Conference
    • By Acceptance Type
    • By Score Categories

9 of 12

Analysis: Verification

10 of 12

Questions?

11 of 12

Dataset: Pre-Processing

︙

⃕

⃕

2

4

1

2

max

︙

4

12 of 12

Analysis: Features

  • Math Notation Complexity

​

  • Acceptance/Rejection

​

  • Recommendation Scores

​

  • Category Scores (Novelty, Correctness, …)

​