Dulhan Jayalath

University of Oxford · United Kingdom

Unclaimed · AI-structured profileOpenAlex

Non-invasive speech decoding and BCI researcher at Oxford — from benchmark competition to evaluation metrics

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-16) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.

Research profile

Output—

The output axis is not yet scored: publication-volume data beyond the small arXiv-based record is insufficient in this snapshot. It would measure cumulative and per-year research output across indexed venues once a fuller, peer-reviewed publication record accumulates.

Impact—

The impact axis is not yet scored: citation data is insufficient at this career stage — h-index 1 and i10-index 0 are too early to characterise influence. It would measure citation uptake and field-normalised impact as the publication record matures.

Focus0/100

The focus dimension scores 0/100, with the field distribution split almost evenly — one third each — across three closely related labels: computational neuroscience / neural speech decoding, brain-computer interfaces (machine learning / neuroscience), and computational neuroscience (non-invasive speech decoding / BCI). Topic counts nonetheless cluster tightly: speech recognition and synthesis (10) and EEG and brain-computer interfaces (10) dominate, with speech and audio processing (5) behind, suggesting one coherent research programme spread across overlapping labels rather than genuine scatter.

Collaboration61/100

Collaboration scores 61/100. The recent three-year record spans 18 works with 20 distinct co-authors, and all three 2026 catalog papers are multi-author efforts in which Jayalath appears as first or middle author. Recurring collaborators include Oiwi Parker Jones and Gilad D. Landau, alongside a wider rotating author pool — a pattern typical of lab-scale, benchmark-driven BCI research groups.

Activity86/100

Activity scores 86/100, the strongest axis in this profile. Recorded works per year climb from 2 (2023) and 2 (2024) to 6 (2025) and 10 (2026), giving 18 works in the most recent three-year window. All three Evolve-catalogued papers appeared in 2026, indicating an accelerating, arXiv-first publishing rhythm typical of an intensive early-career phase.

Reproducibility85/100

Reproducibility scores 85/100, averaged over the 3 papers Evolve has assessed so far. Two catalog entries — the OVMI communication measure and the 2026 PNPL competition paper — carry high reproducibility readiness, while the semantic-bottleneck decoding paper is rated medium, reflecting the group's reliance on shared benchmarks and evaluation infrastructure.

Research summary

Dulhan Jayalath is an early-career researcher at the University of Oxford working at the intersection of machine learning and non-invasive brain-computer interfaces (BCIs), centred on decoding speech from brain activity. The OpenAlex-based bibliometric snapshot indicates an early-stage record — h-index 1 and i10-index 0 — but a steeply rising output curve: recorded works grow from 2 in 2023 and 2 in 2024 to 6 in 2025 and 10 in 2026, totalling 18 works over the last three years with 20 distinct co-authors. Topic assignments concentrate on speech recognition and synthesis (10), EEG and brain-computer interfaces (10), speech and audio processing (5), topic modeling (4), and gaze tracking and assistive technology (3). Three 2026 arXiv preprints appear in the Evolve catalog: a first-author paper proposing OVMI, a coverage-weighted mutual-information measure giving speech BCIs a common, open-vocabulary communication metric; the 2026 PNPL competition paper on word classification and efficient cross-subject generalisation in LibriBrain100; and a middle-author 'semantic bottleneck' approach to MEG-based speech decoding using semantic embeddings and VICReg-style contrastive learning. Evolve's reproducibility assessment across these 3 papers averages 85/100, with two rated high and one medium readiness.

Collaboration pattern

Collaboration scores 61/100. Jayalath's recent three-year record spans 18 works with 20 distinct co-authors, and all three 2026 catalog papers are multi-author projects in which he appears as first or middle author. Recurring collaborators include Oiwi Parker Jones, Gilad D. Landau and Benjamin Ballyk, plus a broader rotating pool (Francesco Mantegna, Gereon Elvers, Tasha Kim and others), consistent with a lab-scale, benchmark-driven BCI group at Oxford.

Research keywords

Non-invasive speech decodingBrain-computer interfacesMagnetoencephalography (MEG)Semantic embeddingsContrastive learningMutual information metricsInformation transfer rateCross-subject generalisationWord classificationBenchmark competitionsEEG signal analysisEvaluation metrics

Highlights

First-author metric giving speech BCIs a common communication measure: OVMI
Proposes OVMI, a coverage-weighted mutual-information measure that gives speech BCIs a common, open-vocabulary communication metric. Evolve judges the theory sound and the comparisons revealing, while noting the scalar estimates rest on strong channel assumptions. Reproducibility readiness: high.
Co-authoring the 2026 PNPL competition on LibriBrain100
Contributes to the competition paper on word classification and efficient cross-subject generalisation in LibriBrain100 MEG data. Evolve highlights strong benchmark design and infrastructure, while noting only early reference baselines exist. Reproducibility readiness: high.
Semantic-bottleneck approach to non-invasive speech decoding
Middle-author work decoding text from MEG via semantic embeddings with VICReg-style contrastive learning. Evolve notes it outperforms BrainECHO on sentence metrics, but the gains are small and partly self-referential, on a single subject. Reproducibility readiness: medium.

Timeline

2026First-author preprint 'A Common Measure of Communication for Speech Brain-Computer Interfaces' introduces OVMI, a coverage-weighted mutual-information metric for evaluating speech BCI communication.
2026Co-authors the 2026 PNPL competition paper on word classification and efficient cross-subject generalisation in LibriBrain100.
2026Contributes to 'The Semantic Bottleneck' on MEG-based non-invasive speech decoding with semantic representations; OpenAlex counts-by-year shows 10 recorded works in 2026.

Papers in the Evolve catalog

Frequent co-authors

Oiwi Parker Jones×3 Gilad D. Landau×2 Benjamin Ballyk×2 Francesco Mantegna×1 Gereon Elvers×1 Tasha Kim×1 Miran Özdogan×1 Luisa Kurth×1 Teyun Kwon×1 SungJun Cho×1 Alex Fung×1 Anna Greer×1

About this profile

This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.

Bibliometric figures (h-index 1, i10-index 0, counts by year) come from an OpenAlex snapshot only; no ORCID was provided, so the profile is name-resolved from public bibliographic data.
All three catalog papers are 2026 arXiv preprints; peer-reviewed venue status is not reflected in this snapshot.
The output and impact dimensions are unscored because citation-based data is insufficient at this career stage (h-index 1, i10-index 0).
The reproducibility score averages Evolve assessments of only 3 papers (mean 85/100) and may not generalise to the full record.