Efthymios Georgiou

University of Bern · Switzerland

Unclaimed · AI-structured profileORCID OpenAlex

University of Bern researcher in multimodal sentiment analysis, efficient MoE LLMs, and low-resource Greek TTS

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—

Output cannot be scored yet: the OpenAlex snapshot is too thin to measure total publication volume or per-year productivity reliably. The axis would count indexed works by year and venue; only fragmentary signals exist, such as an h-index of 4 and a counts-by-year series starting in 2019.

Impact—

Impact cannot be scored yet: citation data is not available in sufficient depth. The axis would measure accumulated citations, citation velocity, and downstream scholarly influence; the lone available indicator, an i10-index of 3, is too small to support a meaningful score.

Focus0/100

Focus scores 0, the lowest axis, because output is deliberately spread: the field distribution is an even one-third split across multimodal machine learning (affective computing/NLP), speech synthesis, and efficient LLM architectures (Mixture-of-Experts routing). Topic counts mirror this breadth — multimodal ML applications (12), sentiment analysis (9), emotion recognition (9), topic modeling (7) — marking a generalist profile advancing three fronts in parallel rather than concentrating on one.

Collaboration44/100

The collaboration score of 44 reflects a modest network: 8 distinct coauthors over the last three years, matching the eight recurring collaborators on record. Georgiou appears in a middle author position without corresponding role on all three reviewed 2026 papers, consistent with contributing within a team rather than leading it.

Activity75/100

Activity is the strongest axis at 75: the OpenAlex record shows 12 works in the last three years, with annual counts rising from 4 in 2025 to 6 in 2026, matching the previous peak of 2022. Production is current and accelerating, concentrated in 2026 arXiv preprints across all three of the researcher's fields.

Reproducibility55/100

Reproducibility scores 55, averaged over the 3 papers Evolve has reviewed, all rated medium readiness. Reviews credit credible recipes — frozen-weight MoE gains of +3.37pp on GSM8K, late-fusion diagnostics, a 3.5-hour Greek TTS pipeline — while flagging missing shuffled-routing and capacity controls, absent significance tests, and key statistics reported only as trends.

Research summary

Efthymios Georgiou is a researcher at the University of Bern whose work spans three evenly weighted fronts: multimodal machine learning for affective computing and NLP, speech synthesis, and efficient LLM architectures built on Mixture-of-Experts routing. The OpenAlex record (h-index 4, i10-index 3) is early-stage, but recent activity is strong: 12 works in the last three years, rising from 4 publications in 2025 to 6 in 2026. Three 2026 arXiv preprints have been reviewed by Evolve. Attention-Aware Routing couples MoE routing with attention-window features and lifts OLMoE's GSM8K score by +3.37 percentage points with frozen weights. The Illusion of Balanced Multimodal Sentiment Analysis shows optimization-based balancing never reliably beats late fusion, separating fitting from utility. Deterministic Prompting for Greek TTS combines prompt conditioning with LoRA on 3.5 hours of data to obtain speaker-stable Modern Greek synthesis. Across reviews, the recipes are judged credible while statistical rigor — significance tests, shuffled and capacity controls — lags behind, yielding a reproducibility score of 55 with medium readiness on all three papers. Topic counts concentrate on multimodal ML applications (12), sentiment analysis (9), and emotion recognition (9). He works within a compact network of 8 recent coauthors and appears in middle author positions on all reviewed papers.

Collaboration pattern

Collaboration runs through a compact, recurring cluster: the eight listed frequent coauthors — Alexandros Potamianos, Ioanna Kaffeza, Georgios Syllas, Kosmas Kritsis, Despoina Kosmopoulou, Anastasios Tsetsilas, Giannis Karamanolakis, and Swastik Roy — match the 8 distinct coauthors recorded over the last three years. The collaboration score of 44 signals a moderate network rather than a large one, and on all three reviewed 2026 papers Georgiou appears in a middle author position, not as corresponding author.

Research keywords

Multimodal sentiment analysisEmotion recognitionModality imbalanceLate fusionMixture-of-Experts routingEfficient LLM architecturesLow-resource text-to-speechParameter-efficient fine-tuning (LoRA)Prompt conditioningTopic modelingMathematical reasoning (GSM8K)

Highlights

Attention-aware routing lifts OLMoE math reasoning with frozen weights
The 2026 AAR paper couples MoE routing with attention-window features, raising OLMoE's GSM8K score by +3.37 percentage points without weight updates; Evolve rates the gains credible but notes missing shuffled-routing and capacity controls and significance tests.
Arguing that balanced multimodal sentiment training is an illusion
This 2026 study shows optimization-based balancing in multimodal sentiment analysis never reliably beats late fusion, separating fitting from utility; Evolve calls the diagnostics strong and the agenda credible, with thin statistics.
Speaker-stable Modern Greek TTS from 3.5 hours of data
A 2026 recipe combines deterministic prompting with LoRA fine-tuning on Parler-TTS to obtain speaker-stable low-resource Greek TTS near human quality; Evolve finds the recipe credible but the key statistics only trends.

Timeline

2019First work appears in the OpenAlex counts-by-year record, with 1 publication.
2021Output resumes at 3 publications after the 2020 gap in the indexed record.
2022Productivity peaks at 6 publications, the highest annual count before 2026.
20254 publications; the recent three-year output reaches 12 works with 8 distinct coauthors.
2026Most productive year on record with 6 works; three arXiv preprints enter the Evolve catalog, spanning MoE routing, multimodal sentiment analysis, and Greek TTS.

Papers in the Evolve catalog

Frequent co-authors

Alexandros Potamianos×3 Ioanna Kaffeza×1 Georgios Syllas×1 Kosmas Kritsis×1 Despoina Kosmopoulou×1 Anastasios Tsetsilas×1 Giannis Karamanolakis×1 Swastik Roy×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.

Bibliometrics come from a single OpenAlex snapshot (identity resolved via ORCID 0000-0002-6042-9584); output and impact axes could not be scored, and no citation totals beyond h-index 4 and i10-index 3 are available.
Only the three reviewed 2026 arXiv preprints are assessed by Evolve; the reproducibility score of 55 averages exactly these papers, and preprint status means peer review is not yet established.
All catalog papers list Georgiou in a middle, non-corresponding author position, so the exact individual contribution level cannot be established from the available data.
The counts-by-year series has a gap at 2020, and the 2026 count of 6 covers only part of the year as of October 2026.