Saman Halgamuge

The University of Melbourne · Australia

Unclaimed · AI-structured profileORCID OpenAlex

University of Melbourne AI researcher: three decades from neural networks and fuzzy logic to LLM-era machine learning

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—

No output score is available yet. The raw record holds per-year work counts, but Evolve has not computed this composite dimension. It would measure overall publication volume and productivity relative to field norms — here, a record running from 2 works (1993) to 18 (2026).

Impact—

No impact score is available yet. Citation aggregates such as h-index (49) and i10-index (164) exist in the raw data, but the composite impact dimension has not been computed. It would measure citation influence and how far this work has propagated across the field.

Focus0/100

Scoring 0/100 — the lowest possible — this dimension reflects exceptional breadth rather than concentration: the recent field distribution splits evenly (0.33 each) across LLM multi-agent systems, graph machine learning, and renewable-energy forecasting, while the long-run topic record spans neural networks (57 works), fuzzy logic and control (54), genomics and phylogenetics (37), metaheuristics (30), and data mining (20).

Collaboration44/100

Scoring 44/100, recent collaboration is real but compact: 46 works over the last three years with 8 distinct co-authors recorded. The recurring partner group — Ken Chen, Wei Wang, Sachith Seneviratne, Hansani Weeratunge, Asela Hevapathige, Ahad N. Zehmakan, Asela Wijesinghe and Maneesha Perera — suggests a stable Melbourne-centred cluster; on all three cataloged 2026 papers Halgamuge appears as last author.

Activity100/100

A perfect 100/100 activity score matches the raw record: 46 works in the last three years, with 19 works in 2024, 9 in 2025 and 18 already in 2026, against a by-year record running continuously from 1993 (2 works) to the present — more than three decades without a documented gap.

Reproducibility55/100

The 55/100 score is the mean readiness across Evolve's 3 reviewed papers, each rated medium. Reviewers consistently credit solid proposed methods — reverse-Bayes anchoring, sky-image fusion, the ECI diagnostic — while flagging single-run experiments, backbone-relative gains, and no statistical significance testing, leaving empirical claims plausible but not yet independently hardened.

Research summary

Saman Halgamuge is a researcher at The University of Melbourne whose OpenAlex-resolved record (ORCID 0000-0002-2536-4930) spans more than three decades: the by-year record begins in 1993 with 2 works and reaches 18 works in 2026, with an h-index of 49 and an i10-index of 164. Activity is exceptional — 46 works in the last three years, scoring 100/100 — while topical data traces the career arc: Neural Networks and Applications (57 works) and Fuzzy Logic and Control Systems (54) lead, followed by Genomics and Phylogenetic Studies (37), Metaheuristic Optimization (30) and Data Mining (20). Evolve's catalog captures the current phase through three 2026 arXiv preprints, in each of which Halgamuge is the last author: a Bayesian backward-reasoning anchor for LLM multi-agent decision-making, a bidirectional sky-image/time-series fusion on a frozen LLM for solar forecasting, and an ECI diagnostic for the echo-chamber effect in graph neural networks. Across these papers the field distribution splits evenly (one third each) among LLM multi-agent systems, graph machine learning, and renewable-energy forecasting. Reproducibility averages 55/100 over the three reviewed papers, all rated medium, with reviewers noting single-run experiments and missing significance tests.

Collaboration pattern

Recent collaboration is anchored at The University of Melbourne: 46 works in three years with 8 distinct co-authors, scoring 44/100 on Evolve's collaboration axis. Recurring partners include Ken Chen, Wei Wang, Sachith Seneviratne, Hansani Weeratunge, Asela Hevapathige, Ahad N. Zehmakan, Asela Wijesinghe and Maneesha Perera. Across all three 2026 catalog papers Halgamuge takes the last author position without corresponding authorship, consistent with a senior supervisory role rather than hands-on first-author experimentation.

Research keywords

LLM multi-agent systemscollective decision-makingBayesian backward reasoninggraph neural networksmultimodal fusionsolar and renewable energy forecastingtime-series forecastingneural networks and applicationsfuzzy logic and control systemsmetaheuristic optimizationdata mininggenomics and phylogenetics

Highlights

Label-free anchoring for disagreeing LLM agents
A 2026 arXiv preprint proposes Bayesian backward reasoning as a label-free anchor, with Jensen-Shannon ranking (MinJS/FwdJS/LogLin), for multi-agent collective voting — showing consistent gains on DDXPlus, though evaluation rests on one synthetic dataset with no statistical tests.
Sky images meet frozen LLMs for solar forecasting
SolCloudLLM adds bidirectional multimodal fusion of sky images and time series to a frozen LLM forecaster, delivering solid MSE and few-shot gains on SIRTA/SKIPP'D — though results come from single runs without multimodal baselines.
Detecting the echo chamber effect in GNNs
This 2026 preprint proposes the ECI diagnostic and a CASP plugin for community-level collapse in graph neural networks, arguing the problem is not just oversmoothing; the theory is clean and honest, but gains are backbone-relative with no significance testing.

Timeline

1993Earliest year in the OpenAlex by-year record, with 2 works — the start of a publication record now spanning more than three decades.
2002An early-career peak in the record: 20 works, close to the 19 recorded in both 2005 and 2007.
2014The highest single-year count in the by-year record: 21 works.
2024Late-career output remains strong: 19 works recorded, followed by 9 in 2025.
2026Three arXiv preprints enter the Evolve catalog — LLM multi-agent decision-making, sky-image solar forecasting, and GNN echo-chamber detection — with 18 works recorded in the by-year data.

Papers in the Evolve catalog

Frequent co-authors

Ken Chen×2 Wei Wang×2 Sachith Seneviratne×2 Hansani Weeratunge×2 Asela Hevapathige×1 Ahad N. Zehmakan×1 Asiri Wijesinghe×1 Maneesha Perera×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 aggregates (h-index 49, i10-index 164, per-year counts) come from an OpenAlex identity snapshot only; no citation-level validation was performed.
Output and impact dimension scores are unavailable (dimMissing) and are not estimated anywhere in this profile.
Evolve has reviewed only three 2026 arXiv preprints; the long back-catalogue is not individually assessed.
Identity is name-resolved from public bibliographic data (ORCID 0000-0002-2536-4930); no biographical facts beyond this record are asserted.