Publications & Research

Thesis Research

Benchmarking LLM Energy Efficiency

Masters of Data Science for Public Policy Thesis, 2025 · Hertie School of Governance

Analysis of transformer model energy consumption across different deployment configurations. Demonstrates substantial variability (up to 500×) in inference-time energy efficiency under fixed FLOPs constraints, highlighting the limitations of theoretical proxies for real-world energy costs.

Advisor: Prof. Lynn Kaack · Award: Data Science Thesis Award 2025

Web Article · Submitted Thesis · Experimental Data & Analysis Scripts · Measurement Tool

Governance of Data Institutions

Masters of Public Policy Thesis, 2024 · Hertie School of Governance

Examines Data Institutions as a democratic approach to data-centric AI governance. Proposes that commons-based DIs can insert a decentralised, community-accountable governance layer into ML development, addressing limitations of current ‘open AI’ debates.

Advisor: Prof. Joanna Bryson

Web Article · Submitted Thesis · Poster

Research Projects

Energy Efficiency Benchmarking

Weizenbaum Institute & Open Data Institute

Comprehensive benchmarking of computational efficiency across different AI systems and architectures. Focuses on sustainable computing practices and green AI principles.

Focus Areas:

Model efficiency metrics
Deployment optimization
Carbon footprint assessment

ML-Strom: Machine Learning for Electricity Grid Analysis

With: Prof. Lion Hirth (Hertie School), Prof. Lynn Kaack (Hertie School)

Applied machine learning to electricity grid forecasting and optimization. Developed models for renewable energy integration and grid stability analysis.

Status: Ongoing research collaboration

UK Climate Attitudes: Bayesian hierarchical IRT model for latent public sentiment from proprietary longitudinal polling data

With: Looking for Growth proprietary polling data (N=3,000 nationally representative UK survey, 2025)

Bayesian hierarchical latent trait model measuring three dimensions of British climate attitudes: economic optimism, environmentalism, and support for radical reform. Identifies party affiliation as the strongest predictor, reveals counterintuitive patterns (older cohorts support reform; material insecurity correlates with pro-environment views), and provides actionable audience segmentation for climate communication.

Web Article · GitHub: Analysis & Model Code

Difference-in-Differences Analysis: Policy Impact Evaluation

With: Prof. Miriam Kayser (Hertie School)

Causal inference methods for evaluating policy interventions. Applied DiD methodology to assess impact of policy changes in economic and social domains.

Methods: Econometrics, causal inference, Python/R

Conference Presentations & Talks

“Sustainable Computing in the Age of AI” — Berlin Data Science Meetup, 2024
“Governance Challenges in Data Institutions” — Hertie School Symposium, 2024
“Energy-Efficient Machine Learning” — Weizenbaum Institute Seminar, 2024

Skills & Methodologies

Research Methods:

Causal inference & econometrics
Machine learning & deep learning
Bayesian hierarchical modeling
Survey design & analysis
Qualitative policy analysis

Technical Skills:

Python, R, Stan, SQL
PyTorch, TensorFlow, JAX
Data visualization (Matplotlib, ggplot2)
Statistical analysis & hypothesis testing

Research Interests

Sustainable Computing: Energy efficiency, environmental impact of ML/AI
Data Governance: Institutional design, data trusts, collective data management
Policy & Technology: Algorithmic decision-making, technology governance
Computational Social Science: Forecasting, geospatial analysis, causal inference, agent-based modeling

Ways to Access My Work

GitHub: github.com/henrycgbaker — code and data analysis
Research Affiliations:
- Weizenbaum Institute profile
- Open Data Institute profile
Direct Contact: henry.c.g.baker@gmail.com

Last updated: December 2025. More publications coming soon as research projects complete.

Henry C G Baker