Sapienza University of Rome researcher in model merging, emergent capabilities, and quantization effects on retrieval
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) 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 metrics — such as total publication count, venue distribution, and yearly productivity — are not yet available in this profile's data snapshot, so this axis cannot be scored; it would measure the volume and consistency of Luca Zhou's published research output.
Impact—
Impact signals — citations, h-index-style aggregates, and downstream uptake of the two catalogued papers — are not yet available; this axis would measure how widely the work on model merging and quantization-induced retrieval damage is being cited and built upon.
Focus0/100
The focus score is 0, and the underlying field distribution explains it directly: the catalogued work splits evenly, 50% in machine learning (LLM interpretability and safety) and 50% in machine learning (model compression and information retrieval). Rather than a single-track agenda, the profile shows two parallel lines — model merging and emergent capabilities on one side, post-training quantization and retrieval stability on the other — with no dominant concentration yet.
Collaboration32/100
The collaboration score is 32, based on 4 distinct coauthors in the snapshot — a compact author network. Emanuele Rodolà, Alessandro Zirilli, Daniele Solombrino, and Roberto Dessì recur as frequent coauthors, and both 2026 papers are first-authored (non-corresponding), consistent with a small-team working style around a Rome-based group rather than large multi-institution consortia.
Activity—
Activity data — publication cadence over time, recent-year output, and venue submission rhythm — is insufficient in the current snapshot, which contains only two 2026 papers; this axis would measure how actively Luca Zhou publishes and sustains research output across years.
Reproducibility55/100
The reproducibility score averages 55 across 2 assessed papers, both rated medium reproducibility readiness by Evolve. The model merging paper is described as well-designed and robust within its LoRA testbeds, while the quantization paper is backed by large experimental sweeps; Evolve nevertheless notes its core theory restates a known condition, a factor tempering the axis.
Research summary
Luca Zhou is a machine learning researcher at Sapienza University of Rome whose full-tier Evolve profile is built from two 2026 first-authored arXiv papers reviewed by Evolve. The catalogued work splits evenly across two subfields: 50% in machine learning (LLM interpretability and safety) and 50% in machine learning (model compression and information retrieval), a distribution that underlies the focus score of 0. The first paper, On Emergent Capabilities and Model Merging, maps three regimes of how model merging preserves, cannot create, or preferentially dilutes emergent capabilities; Evolve judged it well-designed and robust within its LoRA testbeds. The second, The Undetected Damage of Quantization on Retrieval and How to Fix It, argues that post-training quantization can preserve accuracy while flipping 14-46% of retrieval top-1 results; Evolve found the claim well supported by large sweeps while noting the core theory restates a known condition. Both papers carry medium reproducibility readiness, and the reproducibility axis averages 55 across the two assessed papers. The collaboration network is compact: 4 distinct coauthors appear in the snapshot — Emanuele Rodolà, Alessandro Zirilli, Daniele Solombrino, and Roberto Dessì recur — yielding a collaboration score of 32. Output, impact, and activity axes are not yet computable from the available data.
Collaboration pattern
The visible collaboration pattern is compact: 4 distinct coauthors appear in the bibliometric snapshot, giving a collaboration score of 32. Emanuele Rodolà, Alessandro Zirilli, Daniele Solombrino, and Roberto Dessì recur as frequent coauthors, and both 2026 papers are first-authored though not corresponding, suggesting a small, stable Sapienza-centered working group rather than large cross-institution teams.
Research keywords
Highlights
Mapping how model merging shapes emergent capabilities
Quantization silently flips retrieval rankings
Timeline
- 2026Published first-authored arXiv paper On Emergent Capabilities and Model Merging, mapping three regimes of how merging preserves, cannot create, or preferentially dilutes emergent capabilities.
- 2026Published first-authored arXiv paper The Undetected Damage of Quantization on Retrieval and How to Fix It, reporting that quantization flips 14-46% of retrieval top-1 results.
- 2026Both catalogued papers landed in the same year, splitting evenly between LLM interpretability/safety and model compression/information retrieval.
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.
- Bibliometric dimensions come from an OpenAlex-based snapshot only; the output, impact, and activity axes lack sufficient data and are left unscored.
- No ORCID or country is attached to this record, so identity is name-resolved from public bibliographic data (Luca Zhou is a relatively common name).
- The profile rests on 2 Evolve-reviewed arXiv papers (both 2026, first-authored); other publications, if any, are outside the catalog.
- The reproducibility mean of 55 is computed from only 2 assessed papers, so the axis carries high uncertainty.