Computer scientist and research engineer, Quillbury
1 School of Computing, Quillbury University · 2 Visiting researcher, Fernmoor Systems Lab
Abstract. I study shortest paths, compressed graphs and concurrent data structures, and I care about two things at once: that an algorithm is provably correct, and that its implementation is fast on real inputs. Papers, open-source libraries, lectures and talks are collected here, with BibTeX for everything.
Keywords: shortest paths · succinct data structures · verified concurrency · teaching with visualisation
Distances d[v]
| vertex | s | a | b | c | d | e | t |
|---|---|---|---|---|---|---|---|
| d | 0 | 3 | 1 | 8 | 11 | 10 | 13 |
Q = ∅
Three statements I would put on a postcard. Each links to the paper with the full proof; the sketches below are honest but short.
Theorem 1 (Brede and Ruel, SAF 2026). Single-source shortest paths on a graph stored in k-compressed form can be computed in O(m log k) time, without decompressing it.
Best paper award. Implemented in kestrel; see [1].
Each compressed block behaves like a small sorted run. A bucket queue keyed by block keeps at most k live candidates, so every decrease-key costs O(log k) and each edge is touched once.
Lemma 2 (Felber and Brede, 2025). A history of length n is linearisable if every replay of depth 2 log n is.
The basis of our bounded checker; [4].
A violation shrinks to a minimal witness whose overlapping operations form a chain of logarithmic depth, which a bounded replay always reaches.
Theorem 3 (Brede, Kawabata and Felber, 2026). The tallow priority queue is wait-free and linearisable under release–acquire memory, and every operation completes in at most 3p + 2 of its own steps for p threads.
Machine-checked, about 9,400 lines of proof; preprint [2].
A helping array announces each pending operation. We prove an invariant over the array and the heap together, then show that any thread finishes after at most one full round of helping.
The four newest, two of them preprints. The full list has filters by year and topic, and a BibTeX button on every entry.
Ilse Brede and Tomasz Ruel.
Proceedings of the Symposium on Algorithmic Foundations (SAF 2026), pp. 112–131.
Ilse Brede, Mei Kawabata and Jonas Felber.
Preprint, submitted September 2026.
Adaeze Okonkwo-Hale and Ilse Brede.
Preprint, June 2026.
Jonas Felber and Ilse Brede.
Proceedings of the Workshop on Verified Systems (WVS 2025), pp. 41–55.
Every paper with a running time claim ships with code that you can benchmark yourself. Four libraries are maintained; the rest are archived with their papers.
A short version of the CV, in prose.
I am a Senior Lecturer in Computer Science at Quillbury University, where I lead a small group of four working on graph algorithms and proof engineering. Before that I wrote my PhD at Pellbrook on shortest-path trees that carry their own correctness certificates1, and spent two summers building routing software for a logistics team.
Most of my work starts with an algorithm that is fast in theory and annoying in practice. I try to close that gap from both ends: prove a tighter bound, then write the implementation carefully enough that the bound shows up in the benchmark. When the code is concurrent, I also check it mechanically2, because tests alone have fooled me before.
I teach second-year algorithms to about two hundred students each autumn, and I build visual tools such as the player in Figure 1 so that loop invariants stop being abstract. I supervise PhD students in algorithms and verification, and I review for two symposia each year.
Outside work I run the Quillbury Theory Seminar, sail badly on the reservoir, and keep a list of proofs that were wrong in print, including one of mine.
What is happening this term. Older talks, with slides, are on the teaching and talks page.
“Compressed graphs, uncompressed proofs”, Nordvik.
Presenting the best-paper result with Tomasz Ruel.
Proof and code released together, both open.
Verified graph algorithms. Details.
I take one or two PhD students a year and say yes to most talk invitations within a day’s travel. Tell me what you are working on.
Lemma 6 (Correspondence). A message that names a concrete question receives a reply within five working days.