Tel Aviv University researcher in LLM prompt optimization and efficient inference, incl. speculative decoding
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
2
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) 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 reliable total-publication volume is available for this OpenAlex record, so output cannot be scored. This axis would measure cumulative publication count and yearly output cadence across journals, conferences, and preprint servers.
Impact—
The snapshot's only influence signals — h-index 2 and i10-index 1 — are too sparse to score impact meaningfully. This axis would measure citation counts, citation velocity, and venue-level influence across the researcher's published outputs.
Focus0/100
Focus scores 0, the lowest axis. The OpenAlex field distribution splits exactly 50/50 between LLM prompt optimization and efficient LLM inference / speculative decoding, and the topic profile spreads across five clusters — GANs and image synthesis (7), topic modeling (4), multimodal applications (4), NLP techniques (3), and neural network applications (2) — reading as broad rather than concentrated.
Collaboration39/100
The collaboration axis scores 39 on a compact network: the snapshot records 6 distinct co-authors, with Omri Bar Haim, Shahar Katz, Roy Eisenstadt, Ido Cohen, Edo Cohen-Karlik, and Itamar Zimerman recurring across works. Both cataloged 2026 papers place this researcher in middle or last (senior) positions without corresponding-author status, pointing to steady small-team supervision rather than a large, distributed collaboration web.
Activity83/100
Activity is the profile's strongest axis at 83. The snapshot records 16 works in the recent three-year window, with countsByYear listing 16 works in 2026 against only 2 in 2022 — a sharp recent acceleration. Both cataloged papers are 2026 arXiv preprints, consistent with a currently active, fast-producing research phase.
Reproducibility78/100
Reproducibility scores 78, the mean over the 2 papers Evolve has reviewed. ClusterFewshot carries a high repro-readiness rating, while SwitchSD is rated medium after reviewers flagged unreported variance. Both are arXiv preprints describing concrete, implementable pipelines — DSPy-based clustered demo selection and a learned linear copy-intent probe — supporting an above-average reproducibility signal.
Research summary
Lior Wolf is a researcher at Tel Aviv University whose recent work, as indexed by OpenAlex, centers on making large language models easier to optimize and cheaper to run. The profile's field distribution splits evenly between natural language processing (LLM prompt optimization) and efficient LLM inference / speculative decoding. Two 2026 arXiv preprints anchor the catalog: ClusterFewshot, which proposes semantic clustering plus one-shot-scored demonstration selection inside DSPy-style workflows and reports beating BFRS and MIPROv2 with large cost reductions; and SwitchSD, which adds a learned linear copy-intent probe to gate copying against neural drafting in speculative decoding, reporting up to 15% throughput improvement over EAGLE3. Evolve's reviews note strengths — high repro-readiness for ClusterFewshot, consistent gains for SwitchSD — alongside untested statistical significance and unreported variance. The bibliometric snapshot is thin: h-index 2, i10-index 1, 16 works in the recent three-year window concentrated in 2026, 6 distinct co-authors, and a topic profile spanning GANs and image synthesis, topic modeling, multimodal learning, and NLP. Activity scores 83 and reproducibility 78, while output and impact lack sufficient data to score. This profile is name-resolved from public bibliographic data; the underlying OpenAlex record may be partial.
Collaboration pattern
The collaboration footprint in the snapshot is compact: 6 distinct co-authors, with Omri Bar Haim, Shahar Katz, Roy Eisenstadt, Ido Cohen, Edo Cohen-Karlik, and Itamar Zimerman recurring. On both cataloged papers this researcher holds a middle or last (senior) author position without corresponding-author status, consistent with a supervisory role in a small Tel Aviv University team. The collaboration dimension scores 39, mid-range for this profile.
Research keywords
Highlights
ClusterFewshot: clustering-based few-shot demo selection for LLM workflows
SwitchSD: gating speculative decoding with intrinsic copy-intent signals
A 2026 concentration on LLM efficiency and optimization
Both cataloged works appear in 2026 and target the same practical problem: getting more out of LLMs for less — cheaper prompt workflows and faster decoding — within a recent three-year window of 16 recorded works.
Timeline
- 2022The OpenAlex snapshot records 2 works in 2022, the earliest year present in the available countsByYear data.
- 2026countsByYear lists 16 works in 2026, and the snapshot reports 16 works over the recent three-year window — the profile's densest recorded activity.
- 2026ClusterFewshot appears on arXiv, with this researcher as last author; Evolve rates its reproducibility readiness high.
- 2026SwitchSD, on controlling speculative decoding via intrinsic model signals, appears on arXiv with this researcher as a middle author.
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.
- All bibliometrics come from an OpenAlex snapshot; no Google Scholar, Scopus, or DBLP data is included.
- The record shows h-index 2 and i10-index 1, likely reflecting a partial or name-conflated OpenAlex profile; citation figures should be read with caution.
- This profile is name-resolved from public bibliographic data for the name Lior Wolf; identity certainty is not guaranteed.
- Evolve's qualitative assessments cover only the 2 cataloged papers; the broader body of work is not reviewed here.