Tel Aviv University ML researcher: mechanistic interpretability of sequence models and efficient LLM inference
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
3
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—
Not scored: this axis measures total publication volume and venue-normalized productivity. The snapshot lists 30 works in the last three years, but the underlying output data is still incomplete for computing a reliable score, so no estimate is given here.
Impact—
Not scored: this axis measures citation uptake — total citations, h-index trajectory, and field-normalized influence. The available signals (h-index 3, i10-index 1) are too thin for recently posted 2025–2026 arXiv work, so no score is assigned yet.
Focus0/100
Scores 0/100 — the most dispersed axis in this profile. OpenAlex topic tags scatter across five areas (topic modeling 23, advanced neural network applications 18, XAI 12, NLP techniques 11, cryptography and data security 9), and the field distribution splits evenly at 0.5 for sequence-model theory/mechanistic interpretability and 0.5 for efficient LLM inference/speculative decoding. This reads as breadth across adjacent deep-learning themes; name-resolution noise in the OpenAlex record may also contribute.
Collaboration44/100
Scores 44/100, reflecting a deliberately small collaboration footprint: only 8 distinct co-authors across 30 recent works. The recurring group includes Yuval Koren, Assaf Ben-Kish, Raja Giryes, Lior Wolf, Roy Eisenstadt, Ido Cohen, and Edo Cohen-Karlik. On both 2026 catalog papers Zimerman is the last-listed, non-corresponding author, consistent with a senior seat in a compact team around his Tel Aviv University base rather than large multi-institution consortia.
Activity100/100
Scores 100/100, the strongest axis of this profile. The snapshot logs 30 works in the last three years, with countsByYear moving from 9 in 2023 and 4 in 2024 to 14 in 2025 and 12 already in 2026 — an accelerating, near-uninterrupted cadence, and both reviewed papers appeared on arXiv in 2026. One caution: OpenAlex may count preprint versions separately, so raw output can overstate finalized publications.
Reproducibility78/100
Scores 78/100, averaged over the 2 papers Evolve has scored (reproMean 78). The Mamba recall-scaling paper rates high readiness: theory built on Johnson–Lindenstrauss hashing that matches MQAR experiments. The SwitchSD paper rates medium: its reported up-to-15% throughput gain over EAGLE3 comes without variance reporting. With only 2 scored papers this is the least statistically robust axis, but it currently indicates above-average reproducibility discipline.
Research summary
Itamar Zimerman is a machine-learning researcher at Tel Aviv University whose OpenAlex-resolved profile spans two complementary lines of work: sequence-model theory with mechanistic interpretability, and efficient LLM inference via speculative decoding — his field distribution splits evenly (0.5/0.5) between them. The snapshot indexes 30 works over the last three years, with yearly counts moving from 9 (2023) and 4 (2024) to 14 (2025) and 12 (2026); citation metrics remain early (h-index 3, i10-index 1), consistent with a fast-moving arXiv-first record. Topic tags concentrate on topic modeling (23), advanced neural network applications (18), explainable AI (12), NLP techniques (11), and cryptography and data security (9). Two 2026 arXiv papers anchor the profile: a mechanistic study deriving recall scaling laws for Mamba from implicit Johnson–Lindenstrauss hashing (ND = Ω(Nf log V), high reproducibility readiness), and SwitchSD, a learned linear copy-intent probe that gates speculative decoding and reports up to 15% throughput over EAGLE3 (medium readiness). On both he is the last-listed, non-corresponding author, working within a compact pool of 8 distinct co-authors. Activity scores 100/100; reproducibility 78/100 across 2 scored papers, while output and impact are not yet scoreable.
Collaboration pattern
Collaboration is compact and stable: 8 distinct co-authors across 30 recent works (axis score 44/100). The recurring circle — Yuval Koren, Assaf Ben-Kish, Raja Giryes, Lior Wolf, Roy Eisenstadt, Ido Cohen, and Edo Cohen-Karlik — spans both research lines. On the two 2026 catalog papers Zimerman takes the last author slot without corresponding-author designation, suggesting a senior role in small Tel Aviv University-based teams rather than large distributed consortia.
Research keywords
Highlights
Recall scaling laws for Mamba from implicit hashing circuits
SwitchSD: a learned probe that decides when to copy during speculative decoding
Timeline
- 2021First year with works indexed in the OpenAlex snapshot (1 work) — the bibliographic record begins.
- 2023Output jumps to 9 indexed works, the strongest year of the early record.
- 2024A quieter year with 4 indexed works.
- 2025Output peaks at 14 indexed works.
- 202612 works indexed so far; both reviewed papers appear on arXiv — the Mamba recall-scaling study and SwitchSD for speculative decoding.
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; citation counts for very recent 2025–2026 arXiv work typically lag publication.
- The output and impact radar axes are unscored because the underlying bibliometric data is insufficient in this snapshot.
- The profile is name-resolved from public bibliographic records (ORCID 0000-0001-8321-0609); Evolve has reviewed only 2 papers, and no 2022 works appear in the indexed counts.
- Author roles reflect catalog metadata (last-listed, non-corresponding); affiliation details beyond Tel Aviv University are not available.