Evan Shelhamer

University of British Columbia · Canada

Unclaimed · AI-structured profileORCID OpenAlex

ML researcher at UBC spanning geospatial implicit neural representations and efficient sparse vision transformers

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—

The platform flags this axis as missing: normalized productivity measures such as total outputs and output trajectory are not yet computable. The snapshot lists partial aggregates (h-index 22, i10-index 27), but no publication-count aggregate is provided to score output, so no estimate is made here.

Impact—

Impact is also marked missing: field-normalized citation influence (citations per work, field-weighted reach) has not been computed in the source data. Raw aggregates report h-index 22 and i10-index 27, but the platform does not yet derive an impact score from them, and none is estimated here.

Focus0/100

Scores 0 because the field distribution is a perfect 50/50 split between geospatial machine learning/remote sensing and computer vision (efficient vision transformers), with no single-field concentration. Topic counts reinforce this breadth: Advanced Neural Network Applications (51) and Domain Adaptation and Few-Shot Learning (43) dominate, with smaller clusters in adversarial robustness (15), visual attention and saliency (11), and image/video retrieval (10).

Collaboration54/100

The score is 54 over 14 distinct coauthors in the snapshot. The frequent-coauthor list clusters around the geospatial community — Isaac Corley, Esther Rolf, Konstantin Klemmer, Nils Lehmann, Marc Rußwurm, Gengchen Mai, Nathan Jacobs, Hannah Kerner — alongside Arjun Rao, Sreehari Rammohan, Yousef Yassin, and Anthony Fuller. Both catalog papers are multi-authored, with positions of middle (MIND) and last (LookThere!), and neither is flagged as corresponding author.

Activity98/100

The highest radar score, 98, tracks a pronounced recent surge: 28 works in the last three years, with the per-year series climbing to 12 items in 2025 and 15 in 2026 — the two largest values in the 2014–2026 window — after earlier years that stayed between 1 and 8. Both Evolve-catalog papers are dated 2026, consistent with this acceleration.

Reproducibility78/100

Based on exactly two assessed papers, the reproducibility mean is 78. MIND the Gap rates high readiness, with Evolve noting a solid CoordBench evaluation while flagging US-heavy targets and teacher–task overlap for scrutiny; LookThere! rates medium, its Pareto-frontier claims tempered by test-set tuning and the absence of error bars. The split verdicts average into a solid but uneven 78.

Research summary

Evan Shelhamer, affiliated with the University of British Columbia (Canada), is a machine learning researcher whose recent record splits evenly — 0.5 and 0.5 in the snapshot's field distribution — between geospatial machine learning/remote sensing and efficient computer vision. The OpenAlex-resolved profile carries an h-index of 22 and an i10-index of 27, with 28 works in the last three years and a marked recent acceleration: the per-year series climbs to 12 items in 2025 and 15 in 2026, the highest values in the 2014–2026 window. Topic clustering is led by Advanced Neural Network Applications (51) and Domain Adaptation and Few-Shot Learning (43), followed by adversarial robustness (15), visual attention and saliency detection (11), and image/video retrieval (10). Evolve's catalog includes two 2026 arXiv papers: MIND the Gap, a Matryoshka-style geographic implicit neural representation distilled from geospatial teacher models (middle author; high reproducibility readiness), and LookThere!, reinforcement-learned token selection for sparse vision transformers (last author; medium readiness). Radar scores are very high for activity (98) and solid for reproducibility (78), moderate for collaboration (54 over 14 distinct coauthors), and 0 for focus, reflecting the even two-field split; the output and impact axes remain uncomputed in the source data.

Collaboration pattern

Collaboration is moderate (score 54) with 14 distinct coauthors in the snapshot. The recurring list is dominated by the geospatial ML community — Isaac Corley, Esther Rolf, Konstantin Klemmer, Nils Lehmann, Marc Rußwurm, Gengchen Mai, Nathan Jacobs, Hannah Kerner — plus Arjun Rao, Sreehari Rammohan, Yousef Yassin, and Anthony Fuller. On the two reviewed 2026 papers Shelhamer sits middle and last author respectively, neither as corresponding author, consistent with a senior collaborator contributing expertise rather than leading each write-up.

Research keywords

implicit neural representationsgeospatial machine learningMatryoshka representation learningknowledge distillationlocation embeddingsspatial cross-validationvision transformersadaptive computationtoken selectionreinforcement learningefficient inferencedomain adaptation

Highlights

MIND: Matryoshka-distilled geographic implicit neural representations
Distills geospatial teacher models into a coordinate-based implicit neural representation with adjustable spatial scale, evaluated on CoordBench. Evolve rates the evaluation solid while flagging US-heavy targets and teacher–task overlap for scrutiny; Shelhamer is a middle author and reproducibility readiness is high.
LookThere!: reinforcement-learned token selection for sparse vision
LookThere! trains GRPO-based reinforcement learning to select tokens, claiming an accuracy–compute Pareto frontier for sparse vision transformers. Evolve notes broad strong results but tempers them over test-set tuning and missing error bars; Shelhamer is the last author and reproducibility readiness is medium.
2025–2026 output surge drives activity score of 98
The per-year series jumps to 12 items in 2025 and 15 in 2026, the two highest values in the 2014–2026 window, with 28 works recorded over the last three years. Both Evolve-catalog papers fall in 2026, one in each of the researcher's two evenly split fields.

Timeline

2014Earliest year in the snapshot's per-year series, logging 8 items — an early level matched only again in 2018.
2018Counts return to 8, matching the 2014 level, before the series dips through quieter years (2019: 7; 2020: 3; 2023: 2; 2024: 1).
2021Part of a mid-period plateau at 6 items, repeated in 2022, between the early peak and the recent surge.
2025The series climbs to 12 items, the first clear break above the long-standing level of 8.
2026Highest count in the series at 15; both Evolve-catalog papers — MIND the Gap and LookThere! — are dated this year.

Papers in the Evolve catalog

Frequent co-authors

Isaac Corley×1 Arjun Rao×1 Esther Rolf×1 Konstantin Klemmer×1 Nils Lehmann×1 Marc Rußwurm×1 Gengchen Mai×1 Nathan Jacobs×1 Hannah Kerner×1 Sreehari Rammohan×1 Yousef Yassin×1 Anthony Fuller×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.

Bibliometric identity is resolved from OpenAlex (identity: openalex) against the listed ORCID (0000-0002-5228-3095); verify this record matches the intended Evan Shelhamer before citing.
All quoted statistics (h-index 22, i10-index 27, coauthor and topic counts, per-year counts) come verbatim from the provided OpenAlex-derived snapshot; no external data was added.
The output and impact radar axes are flagged missing in the source data; no productivity or citation-impact score is computed, and none is estimated in this profile.
Reproducibility (mean 78) rests on exactly two Evolve-reviewed papers, and the semantics of the countsByYear series (works vs. citations) are not specified in the snapshot.