Imperial College London researcher in inference-time LLM alignment and efficient sampling for generative models
2
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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 yet: every catalogued work is dated 2026, so volume, venue-tier, and authorship-share signals are too fresh to score reliably. This axis would measure total publications, venue quality, and first/corresponding-author proportion.
Impact—
Impact cannot be scored yet: the snapshot reports h-index 0 and i10-index 0 on an all-2026 arXiv record, too early for citation accumulation. This axis would measure citations, citation velocity, and uptake of the alignment and sampling methods.
Focus0/100
The focus score is 0, driven by an even 50/50 field split between generative-model efficient inference and inference-time LLM alignment, and five OpenAlex topics (topic modeling, neural network applications, NLP, video coding, GAN image synthesis) each appearing once. In practice the two halves are adjacent — both target sampling and inference efficiency — so the breadth looks transitional rather than scattered.
Collaboration41/100
Collaboration scores 41 on 7 distinct coauthors in the snapshot. Recurring names — Deniz Gündüz, Marcello Bullo, Öykü Sıla Güner, Arpan Mukherjee, Sicheng Wan — suggest a stable small-team pattern: first-author role on the ExpBoN alignment paper, a middle position on the speculative-diffusion paper, both arXiv-first 2026 releases.
Activity40/100
Activity scores 40, consistent with an emerging record: recent-3-year works total 3, and the counts-by-year data places all 3 in 2026. Both catalogued papers are 2026 arXiv preprints, so the entire visible output is very recent and an annual publication cadence cannot yet be established.
Reproducibility55/100
Reproducibility scores 55, the mean across 2 assessed papers, both rated medium readiness. Both contributions are theory-first — exact tilted decompositions with geometric convergence, and exact coupled speculative sampling — so code and empirical harnesses appear to lag the mathematics; the ~8% NFE-only gains on diffusion sampling also invite wider empirical validation.
Research summary
Yanxiao Liu is an early-career machine learning researcher at Imperial College London whose recorded output clusters entirely in 2026, with three works in the OpenAlex snapshot and two papers reviewed in the Evolve catalog. The profile splits evenly between two closely related subfields: generative models with efficient inference (50%) and LLM alignment via inference-time methods (50%). The first-authored ExpBoN paper introduces an exponential-noise soft best-of-n sampler for test-time LLM alignment, with an exact finite-n tilted decomposition and geometric convergence, plus an ExpGSI variant that saves GSI compute at comparable accuracy. A co-authored paper accelerates diffusion sampling with speculative draft trees, using greedy rejection sampling and reflection maximal coupling for exact sampling; Evolve notes sound theory but modest (~8%) gains over RMC measured in NFE only. Citation indices (h-index 0, i10-index 0) reflect how recent this output is, and the focus score of 0 tracks the even field split and five single-count OpenAlex topics. Recurring coauthors include Deniz Gündüz, Marcello Bullo, Öykü Sıla Güner, Arpan Mukherjee, and Sicheng Wan, with 7 distinct coauthors recorded. Both reviewed papers carry medium reproducibility readiness (mean 55), consistent with theory-heavy contributions whose empirical tooling is still consolidating.
Collaboration pattern
Records show 7 distinct coauthors and a compact recurring core: Deniz Gündüz, Marcello Bullo, Öykü Sıla Güner, Arpan Mukherjee, and Sicheng Wan. The pattern is small-team, arXiv-first work with role variety — first author on the ExpBoN alignment paper, middle author on the speculative diffusion sampling paper. Both collaborations sit inside machine-learning inference efficiency, suggesting a single Imperial College London cluster rather than wide cross-institution spread.
Research keywords
Highlights
ExpBoN: exponential-noise Best-of-n for efficient test-time LLM alignment
Speculative draft trees for exact, faster diffusion sampling
Timeline
- 2026First-authored ExpBoN, introducing exponential-noise soft Best-of-n for efficient test-time LLM alignment (arXiv).
- 2026Co-authored the speculative draft tree paper accelerating diffusion sampling via exact greedy-rejection-sampling coupling (arXiv).
- 2026OpenAlex snapshot records 3 works, all dated 2026 — the entire visible output so far.
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 an OpenAlex snapshot only; the reported h-index (0) and i10-index (0) reflect a 2026-dated arXiv record, not certified peer-review impact.
- "Yanxiao Liu" is a common name; this profile is name-resolved from public bibliographic data (identity: openalex) and no ORCID was provided for confirmation.
- Only 2 papers were reviewed by Evolve; the OpenAlex topics (video coding, GANs, topic modeling, etc.) come from the broader record and may not reflect current research directions.
- Coauthor strings contained encoding variants and the researcher's own name; duplicates were merged and self-entries excluded during cleaning.