Deakin University ML researcher: geometry-aware Bayesian fine-tuning meets personalized federated learning
3
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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 snapshot lacks a reliable total publication count, venue-quality weighting, and career-long trajectory; only 8 recent three-year works are visible. This axis would measure the volume and consistency of scholarly production over time.
Impact—
Impact cannot be scored yet: the snapshot shows h-index 0 and i10-index 0 with no citation counts, and the record is too young for meaningful reception. This axis would measure citations, uptake, and influence of the work over time.
Focus0/100
The focus axis scores 0, reflecting a deliberately spread portfolio: the field distribution splits exactly 50/50 between federated learning and Bayesian deep learning / parameter-efficient fine-tuning, and five topic clusters — led by Bayesian modeling and causal inference and by domain adaptation and few-shot learning (3 works each) — share the output, while a clinically applied oversampling paper widens the scope further. This reads as breadth across methods rather than drift, but concentration is currently minimal.
Collaboration62/100
Scored 62, collaboration reflects a strongly team-based pattern: 8 works in the recent three-year window involved 21 distinct coauthors. A recurring core — Sunil Gupta, Kien Do, Dang Nguyen, Dat Phan-Trong, Julian Berk, Debopriyo Bal — accompanies the Bayesian/PEFT and federated strands, while Jill Newby, Helen Christensen, Taylor A. Braund, and Leonard Hoon appear on the applied mental-health paper. Every cataloged paper places Duong in a middle position, never as corresponding author.
Activity64/100
Scored 64, activity is recent and accelerating: the year counts show one indexed work in 2024 rising to seven in 2026, with recent3yWorks totaling 8. Visible output is arXiv-heavy — three 2026 preprints sit in the Evolve catalog — signalling a high-tempo, submission-oriented phase. Because the record is short, this axis captures a current burst of productivity rather than a long sustained cadence, and no 2025 works appear in the provided year counts.
Reproducibility55/100
Scored 55, from reproMean across 2 assessed papers, both rated medium reproducibility readiness. Evolve's verdicts name concrete gaps: StePS evaluates a single base model with small margins and no significance tests, while SAPE-FL's convergence theory is idealized relative to its empirical claims. The methods described appear detailed enough for reimplementation, but independent verification of headline gains remains the bottleneck, and no released code or artifacts are reported in the reviews.
Research summary
Bao Duong is a machine learning researcher at Deakin University (Australia) whose OpenAlex-linked record splits evenly between two methodological streams — federated learning and Bayesian deep learning with parameter-efficient fine-tuning, each 0.5 of the field distribution. The visible output is recent and fast-moving: one indexed work in 2024 rises to seven in 2026, eight works across the recent three-year window, produced with 21 distinct coauthors. Topic clusters centre on Bayesian modeling and causal inference (3 works) and domain adaptation and few-shot learning (3), with statistical methods and inference, data mining algorithms, and stochastic gradient optimization contributing two each. Two 2026 arXiv preprints anchor the profile: StePS, which brings Riemannian Stein variational gradient descent to Stiefel-manifold LoRA for calibrated parameter-efficient fine-tuning, and SAPE-FL, a similarity-aware route to personalized federated learning under statistical heterogeneity. A third paper applies a novel oversampling method to predict amotivation and anhedonia symptoms among university students. Evolve rates both flagship preprints at medium reproducibility readiness (mean 55 across 2 assessed papers), citing single-model evaluation and idealized convergence theory. With h-index 0 and i10-index 0, this is an early-stage record whose methodological range currently outruns its citation footprint.
Collaboration pattern
Duong works in tightly coupled teams: 8 recent three-year works span 21 distinct coauthors. A recurring core — Sunil Gupta, Kien Do, Dang Nguyen, Dat Phan-Trong, Julian Berk, Debopriyo Bal — supports the Bayesian/PEFT and federated strands, while Jill Newby, Helen Christensen, Taylor A. Braund, and Leonard Hoon appear on the applied student mental-health paper. All cataloged papers list Duong as a middle author and none as corresponding, indicating a contributor role within an established group.
Research keywords
Highlights
StePS: Riemannian SVGD on the Stiefel Manifold for calibrated LoRA fine-tuning
SAPE-FL: similarity-aware personalized federated learning under heterogeneity
Screening amotivation and anhedonia in university students with a novel oversampling method
A third 2026 arXiv paper applies machine learning with a new oversampling technique to predict symptoms of amotivation and anhedonia among university students, extending the methodological core toward digital mental health. It is not yet reviewed in the Evolve catalog, so no field, keywords, verdict, or readiness rating exist for it.
Timeline
- 2024Earliest entry in the OpenAlex year counts: one indexed publication opens the visible record.
- 2026Output jumps to seven indexed works in a single year, within a recent three-year total of eight — the most active year on file.
- 2026Posts the StePS preprint on arXiv: Stiefel-manifold LoRA fine-tuned with Riemannian Stein variational gradient descent for calibrated, Bayesian parameter-efficient adaptation.
- 2026Posts SAPE-FL on arXiv, proposing similarity-aware anchoring of clients to global and peer models for personalized federated learning under non-IID heterogeneity.
- 2026Contributes in a middle authorship position to a third arXiv paper predicting amotivation and anhedonia symptoms in university students via a novel oversampling method.
Papers in the Evolve catalog
Frequent co-authors
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.
- Bibliometrics come from OpenAlex only (identity: openalex); no ORCID is linked, and 'Bao Duong' is a common Vietnamese name, so the record may omit works or absorb homonyms — the profile is name-resolved from public bibliographic data.
- Citation-impact signals are absent or zero in this snapshot (h-index 0, i10-index 0; output and impact axes unscored), so this is an early-stage profile.
- Only 3 papers carry Evolve catalog data; the third lacks field, keywords, verdict, reproducibility rating, and slug.
- All cataloged papers are arXiv preprints with middle authorship and no corresponding-author role, so peer-review status and exact contribution shares are unknown.