About
Hi! I’m an ML researcher / engineer. Most recently I worked on a project on Bayesian networks at CHAI in UC Berkeley; in 2024 I built LLM agents & infrastructure at Finster AI in London. Before this I enjoyed a 2 year datascience/AI master’s at ETH Zurich where I focused on NLP and mathematical models for social networks. I’m broadly interested in language, cognition, interpretable computational methods and neurosymbolic AI.
Prior all of that, I did a math bachelors at the University of Cambridge, and grew up in London except a 1 year stint 2008-9 in Beijing!
Blog
I aspirationally want to write / self-reflect more. I also document some fun projects, expound / research ideas. Posts are of mixed quality and content/topic!
Publications
GNN-Guided Block Selection in Gibbs MCMC
Bayesian networks (BNs) are great and interpretable, however exact/approximate inference can be intractably slow for large networks. We propose an automated block detection algorithm method to amortise inference time by training a Graph Neural Network (GNN) to propose Blocks (for blocked Gibbs MCMC) directly from a BNs structure.
| NeurIPS 2025 SPIGM workshop pdf | Github repo |
Balanced Bidirectional Breadth-First Search on Scale-Free Networks
Arxiv Preprint: theory and experiments around benchmarking bidirectional BFS runtime in Chung-Lu graphs and GIRGs
Master’s Thesis - expressivity of geometric inhomogeneous random graphs (GIRGs) - metric and non-metric
Paper at CompleNet 2024 on investigating the GIRG model’s empirical fit to real life social network graphs.Github repo
Algorithms for acyclic weighted finite-state automata with failure arcs
Paper at EMNLP 2023 on graph algorithms for efficiently computing the pathsum in failure-arc augmented Weighted Finite State Automata.
Unity Games
Cyber City VR: VR FPS game project with students in UCL, designed for the Meta Quest 2. I lead a develooper team of a handful of other students.
Big Bang Blaster: Top down 2D shooter game project with students in Berkeley. Much appreciation to the teammates and our mentor Kevin.
Tutoring
I have tutored some math and coding (mostly python!). Feel free to email me benjamin.l.dayan@gmail.com if interested.