Columbia University researcher spanning image quality assessment, video reward modeling, and rare-event AI safety
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-09) 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 aggregate output data — total papers, preprint counts, or venue record — is available in this snapshot, so this axis is unscored. It would measure overall publication volume and its growth over time.
Impact—
Citation counts, h-index, and venue acceptance are not yet available for this profile, so impact is unscored. This axis would measure citations to, and downstream uptake of, the reviewed papers.
Focus0/100
The focus score is 0 because the field distribution is perfectly even: each of the three reviewed 2026 papers sits in a different area — computer vision image quality assessment, machine learning for video generation evaluation, and AI safety rare-event Monte Carlo simulation — at exactly one-third weight each. The evidence therefore shows deliberate breadth across evaluation and safety problems rather than concentration in a single subfield.
Collaboration55/100
Collaboration scores 55, with 15 distinct coauthors across the three reviewed papers. Recurring names include Bo Peng, Songlin Yang, Zhenchen Tang, Zichuan Wang, Leilei Cao, and Fengjie Zhu. Jing Dong appears in middle position on the rare-event paper and last position on the other two, none as corresponding author — a pattern consistent with contributing expertise within moderately sized, multi-author teams rather than leading solo efforts.
Activity—
This snapshot contains only three 2026 preprints and no multi-year publication record, so research cadence cannot be scored. The axis would measure output frequency and sustained engagement across recent years.
Reproducibility43/100
Reproducibility scores 43, the mean readiness across all three reviewed papers (reproPapers = 3). Two papers rate medium — the rare-event importance-sampling work and the agentic image quality study — while RewardVerse rates low, with Evolve noting missing headline tables and residual scalar instability. The agentic IQA paper's central evaluation rests on a self-built 180-image set without statistical tests, further tempering confidence.
Research summary
Jing Dong is a researcher at Columbia University whose reviewed work spans three adjacent areas of machine learning: image quality assessment, video generation evaluation, and AI safety through rare-event Monte Carlo simulation. The Evolve catalog holds three 2026 arXiv preprints. Rare Event Estimation via Iterative Unalignment applies importance sampling over language-model weight space, with probability estimates verifiable down to the 1e-9 scale, though demonstrated only on proxy events rather than real agent failures. RewardVerse introduces rubric-guided policy optimization for video reward modeling, using GRPO and soft-logits readout to curb scalar drift; ablations are convincing, but headline tables are missing and residual instability remains. An Evolutionary Agentic Approach for Open-ended Image Quality Perception uses evolutionary multi-agent protocols to reduce MLLM holistic bias in IQA, cutting the holistic-bias rate from 44.4% to 8.6%, although the central evaluation rests on a self-built 180-image set without statistical tests. Across the three papers Jing Dong holds middle or last authorship without a corresponding role, collaborating with 15 distinct coauthors including Bo Peng, Songlin Yang, and Zhenchen Tang. Output, impact, and activity metrics are not yet available; the even one-third split across fields yields a focus score of 0, and reproducibility averages 43. The profile is name-resolved from public bibliographic data, marking an early-stage, breadth-first research trajectory.
Collaboration pattern
Across three 2026 preprints Jing Dong works with 15 distinct coauthors, giving a collaboration score of 55. Recurring collaborators include Bo Peng, Songlin Yang, Zhenchen Tang, Zichuan Wang, Leilei Cao, and Fengjie Zhu. Authorship sits in the middle on the rare-event paper and last on the video reward and agentic IQA papers, none as corresponding author — a pattern of contributing specialist expertise within moderately sized teams rather than single-author or corresponding-lead work.
Research keywords
Highlights
Importance sampling in LM weight space for rare-event estimation
Rubric-guided policy optimization for video reward modeling
Evolutionary agentic protocols cut holistic bias in image quality assessment
Timeline
- 2026Co-authored 'Rare Event Estimation via Iterative Unalignment' (arXiv), applying weight-space importance sampling to rare-event estimation in a middle authorship position.
- 2026Contributed to 'RewardVerse: Rubric-Guided Policy Optimization for Video Reward Modeling' (arXiv) in last authorship position, targeting scalar drift in video reward models.
- 2026Contributed to 'An Evolutionary Agentic Approach for Open-ended Image Quality Perception' (arXiv) in last authorship position, reducing MLLM holistic bias in image quality assessment.
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 signals (coauthor counts, field distribution, reproducibility scores) come from OpenAlex-based data and Evolve's review of only three arXiv preprints.
- Citation counts, h-index, and venue acceptance are unavailable, so output, impact, and activity dimensions are unscored.
- All three reviewed papers are 2026 arXiv preprints; peer-review status is unknown.
- 'Jing Dong' is a common name; this profile is name-resolved from public bibliographic data and may not reflect a single individual.