Who Leads Machine Learning Research? A Bibliometric Portrait of 1,000 Top-Cited Researchers on Google Scholar

Technical Summary: Scholar-ML-1000

Problem Statement
Despite widespread commentary on the structural transformation of machine learning (ML) from a computer science sub-field to a cross-disciplinary methodology, and the concurrent shift of research activity from universities to industrial laboratories, the global talent landscape remains poorly documented. Existing literature lacks a comprehensive, citation-sorted snapshot of the world's most-cited ML researchers. Prior datasets, such as the industry-focused MAANG-AI-450, are limited in scope, excluding academics and failing to capture the full disciplinary breadth of the field. This paper addresses the need to answer fundamental questions: Who are the most-cited ML researchers globally? What is the balance between academia and industry? Which institutions concentrate the most talent, and how diverse are the research interests within the top tier?

Methodology
The authors constructed Scholar-ML-1000, a dataset comprising the 1,000 most-cited researchers on Google Scholar who self-identify with the label "machine learning."

• Data Retrieval: Using a custom browser automation script, the authors queried Google Scholar's public author-search endpoint (label:machine_learning), sorting results by total citation count. They retrieved 100 consecutive pages (1,000 entries) over a 72-hour window in June 2026 to minimize temporal inconsistencies.
• Fields Collected: For each researcher, the dataset includes rank, name, affiliation, self-reported interests (keywords), and total cumulative citations.
• Data Cleaning: Manual inspection of the top 20 entries identified four anomalous accounts (citation-inflated or off-topic profiles, including a particle physicist whose citations stem from large detector collaborations rather than ML focus). These four were flagged and excluded from quantitative analysis, resulting in a clean dataset of n = 996.
• Classification: Affiliations were classified into "Academia," "Industry," "Both," or "Unknown" using keyword matching (e.g., "university," "professor" vs. specific company names like "Google," "Meta"). Geographic distribution was determined via a priority keyword matching system that prioritized explicit city/institution keywords over company headquarters locations to avoid misclassifying international researchers (e.g., distinguishing UK-based DeepMind researchers from US-based Google researchers).

Key Contributions

1. Dataset Release: The paper introduces and releases Scholar-ML-1000, the first publicly available, citation-sorted snapshot of the global ML researcher population covering the full top-1,000.
2. Citation Distribution Analysis: The study characterizes the global citation distribution, finding a Gini coefficient of 0.37. This is significantly lower than the 0.57 observed in the industry-only MAANG-AI-450 dataset, suggesting that extreme impact concentration is more characteristic of industry labs than the global ML ecosystem.
3. Industry–Academia Quantification: The paper quantifies the sector split, challenging the narrative that industry has completely captured ML research. It finds that academia remains the dominant sector in both researcher count and citation share.
4. Disciplinary Breadth: The dataset reveals substantial communities from non-traditional ML fields, specifically particle physics, computational biology, and astrophysics, which are often invisible in industry-focused datasets.

Key Results

• Citation Statistics: The clean dataset (n=996) spans a citation range of 40,809 to 1,118,561 (median: 69,320; mean: 101,379). The top 10% of researchers account for 31.6% of total citations, while the top 58.8% account for 80%.
• Sector Split: 52.3% of researchers hold primarily academic affiliations, accounting for 52.3% of total citations. Pure-industry researchers constitute 22.6% of the count but account for 26.6% of citations, indicating a modest citation productivity advantage for industry researchers.
• Institutional Concentration:
  • Google and Google DeepMind collectively list 209 researchers (21.0% of the dataset), a figure more than double the combined total of Stanford, MIT, and UC Berkeley (74 researchers).
  • Among universities, the top five are North American: Stanford (31), MIT (23), UC Berkeley (20), Carnegie Mellon (16), and University of Toronto (13).
• Geographic Distribution: US-based researchers constitute approximately 32.4% of the dataset (combining university and company HQ classifications). The United Kingdom ranks third with 13.1% (130 researchers), largely due to Google DeepMind's London headquarters. A significant portion (34.5%) remains "Unknown" due to affiliations lacking explicit location keywords.
• Disciplinary Diversity: Beyond core ML (Computer Vision, AI, Deep Learning), the top 1,000 includes distinct communities in Particle Physics (49), Computational Biology (50), and Astrophysics (15).

Significance and Claims
The paper claims to provide the first systematic bibliometric portrait of the global ML talent ecosystem. Its primary significance lies in correcting the perception of a total "brain drain" to industry; while migration is real, the global stock of highly-cited academic researchers remains substantial enough to maintain academia's majority share of the top tier.

The study highlights that Google and DeepMind represent an unprecedented concentration of high-citation talent within a single corporate entity, surpassing the combined output of the world's most prestigious CS departments. Furthermore, the inclusion of physicists and biologists underscores that ML has become a primary tool across diverse scientific domains, a nuance often missed in venue-based or industry-centric analyses. The authors position the dataset as a foundational resource for future "science of science" research, enabling longitudinal tracking of researcher mobility and the evolution of disciplinary boundaries.