Will Troiani

Mathematician (logic & geometry) • semantics of computation • singular learning theory • AI safety
Melbourne / remote william.a.troiani@gmail.com GitHub
Portrait of Will Troiani in a suit.
Will Troiani
Research Scientist
Performance Artist

Research summary

I work at the boundary between logic, algebraic geometry, and computation. A recurring theme is that computation has geometry: normalization and cut-elimination can be represented as algebraic operations (ideals, Gröbner bases, schemes) and—via singular learning theory—as local geometry near critical points.

I’m particularly interested in how this perspective can inform AI safety: interpretability as structural inference, model selection for singular models, and understanding when two systems that compute the same function nonetheless have meaningfully different internal structure.

A romantic image comes to mind: as one performs deductive reasoning, a kind of geometric hologram appears overhead, structures blooming in harmonic alignment with the logic, or else the contrary.

Diagram labeled X1, X2, ... illustrating a graphical presentation of the Falling Roofs algorithm.
A graphical presentation of the Falling Roofs algorithm, from the paper “Elimination and cut-elimination in multiplicative linear logic”.

Core interests

  • Linear logic, proof nets, cut-elimination
  • Algebraic geometry (Hilbert schemes; schemes and singularities)
  • Type theory (programs-as-proofs viewpoints)
  • Categorical logic and internal languages
  • Computability and denotational semantics
  • Singular learning theory (SLT)
  • Quantum error correction (via logical/semantic viewpoints)

Adjacent interests & practice

  • Mathematical musicology; mathematical aesthetics
  • Perspective/projective geometry
  • Foundations of mathematics; philosophy of mathematics; philosophy of science
  • Quantum computing
  • Formal verification
  • Mathematical communication; storytelling as a research tool
  • Generative/algorithmic art; audio‑visual performance

Experience

Research Scientist (Theory) Timaeus
Jul 2025 – Aug 2026 (fixed-term; remote)
  • Research focus: singular learning theory, geometry of computation, and implications for AI safety (model selection, structural inference, interpretability).
  • Employment administered via EOR arrangements (Deel Australia Services Pty Ltd; later Justworks Global Ventures Australia Pty Ltd), making the affiliation with Timaeus explicit.
Postgraduate Researcher (ARC-funded), University of Melbourne
2024 – Jul 2025
Supervisors: Daniel Murfet, David Ridout.
Research Assistant (DST-funded), University of Melbourne
2017 – 2018
Supervisor: Daniel Murfet.

Education

PhD in Mathematics, University of Melbourne (with study abroad at Université Sorbonne Paris Nord)
2020 – 2024
  • Thesis: Algebraic Geometry and Linear LogicPDF
  • Supervisors: Daniel Murfet; Thomas Seiller (USPN)
Master of Science (Mathematics & Statistics), University of Melbourne
2017 – 2019
  • Thesis: Simplicial Sets are AlgorithmsPDF
  • Supervisor: Daniel Murfet
Bachelor of Science (Mathematics & Statistics), University of Melbourne
2013 – 2016

Publications & research writing

Peer‑reviewed / accepted

The Internal Logic and Finite Colimits Published
Solo-authored • Logica UniversalisPDF
Internal languages of topoi; constructive descriptions of finite colimits; contrasts with ZF‑style set‑theoretic intuition.
With Daniel Murfet • Mathematical Structures in Computer Science • arXiv:2207.10871
Associates ideals to proof structures; links normalization to Gröbner bases via a Buchberger‑style “Falling Roofs” algorithm.
With Daniel Murfet • arXiv:2502.08911 (2025)
A geometric semantics for shallow MELL proofs via Hilbert schemes; invariance under cut‑elimination via explicit isomorphisms.
With Daniel Murfet • arXiv:2008.10131 (2020)
Relates simply‑typed λ‑calculus and intuitionistic sequent calculus; local vs global perspectives on the same computational structure.

Conference

Programs as Singularities Accepted (ODYSSEY 2025)
With Daniel Murfet • arXiv:2504.08075 (Apr 2025)
Develops a correspondence between Turing machine structure and singularities of real analytic functions by embedding discrete codes into a smooth “noisy” parameter space. Relates Taylor coefficients of a likelihood potential to error syndromes, connecting linear‑logic derivatives with singular learning theory and Bayesian inference.

Preprints (submitted to MSCS)

With Daniel Murfet • arXiv:2405.19051 (2024)
Interprets multiplicative proof‑net reduction through stabilizer codes; cut‑elimination as a quantum cooling / correction process.

Other research writing

HAL preprint
Contrasts categorical and syntactic perspectives on linear logic; revisits Girard’s normal functors model.
University of Melbourne (2019)
A constructive viewpoint on finite simplicial sets and the computation of geometric realisations.

Selected talks & lecture series

  • 18 Feb 2025 — ART Seminar, University of Melbourne: Patterns of Thought: A Philosophical Journey Through Mathematics and Music (seminar page).
  • 30 Mar 2023 — CHoCoLa (ENS Lyon): Computation in logic as the splitting of idempotents in algebraic geometry; two models of multiplicative linear logic (slides).
  • Nov 2023Derivatives in Logic and Learning (bonus seminar on Differential Linear Logic).
Seminar links (full index)

Metauni: Anything at all seminar

Metauni: Foundations (videos)

  • First order logic — video
  • Models of first order theories — video
  • Gödel (Part 1) — video
  • Gödel (Part 2) — video
  • Gödel (Part 3) — video
  • Gödel (Part 4) — video
  • Gödel (Part 5) — video
  • Gödel (Part 6) — video
  • Ax–Grothendieck (Part 1) — video
  • Ax–Grothendieck (Part 2) — video
  • Ax–Grothendieck (Part 3) — video
  • Ax–Grothendieck (Part 4) — video
  • Ax–Grothendieck (Part 5) — video

Other seminars

Expository writing & essays

Scholarships & funding

Scholarships

  • 2016 — AMSI Summer Research Scholarship (Vacation Research Scholarship), supervisor: Daniel Murfet.
  • 2016 — AMSI Travel Grant.
  • 2015 — University of Melbourne Vacation Research Scholarship, supervisor: Peter Forrester.

Grants / funding context

  • ARC-funded postgraduate research position (University of Melbourne), 2024 – Jul 2025.
  • ARIA (UK)Mathematics for Safe AI: grant held by Timaeus; I was hired under this award as Research Scientist (Theory).