ML researcher at Amazon: efficient LLM reasoning, multimodal latent diffusion, and adaptive LLM agents
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
2
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—
Output cannot be scored yet: the productivity pipeline lacks sufficient coverage for this OpenAlex-resolved profile. This axis would measure publication volume, consistency, and venue spread over time, complementing the raw counts-by-year signal (e.g., 15 outputs in 2026).
Impact—
Impact cannot be scored yet: citation signals are too sparse (h-index 2, i10 index 1), and 2026 arXiv work has had little time to accrue citations. This axis would measure citations, citing works, and downstream influence.
Focus0/100
Focus scores 0 because the field distribution is perfectly flat: the profile splits into three equal thirds — multimodal machine learning, efficient reasoning in large language models, and LLM agents/prompt optimization (0.333 each). The three reviewed 2026 papers confirm deliberate breadth rather than a single specialty, spanning GRPO credit assignment, latent-diffusion multimodal reasoning, and replay-gated prompt adaptation.
Collaboration63/100
Collaboration scores 63, built on 23 distinct coauthors in the raw record. The frequent-coauthor list is anchored by Manqing Mao, Zhuoer Wang, Samson Koelle, Haoqiang Kang, Yizhe Zhang, Yian Ma, Lianhui Qin, Yuqing Zhou, Hong Wang, and Jie Yuan, suggesting a small recurring core team behind the 2026 arXiv papers; all three reviewed papers place Kuang in a middle-author position.
Activity87/100
Activity is the strongest axis at 87. The raw counts-by-year show sustained output since 2019 (3 in 2019, 1 in 2020, 1 in 2022, 2 in 2023, 2 in 2024, 2 in 2025) and a sharp acceleration to 15 outputs in 2026, with 19 works in the last three years — an intensive current production phase.
Reproducibility43/100
Reproducibility scores 43 as the mean readiness across the three reviewed papers. Readiness is medium for the GRPO credit-assignment paper and Uni-LaDiR — the latter has strong ablations but no significance tests — and low for the RIPPLE workflow-synthesis work, whose gains come on a 39-task synthetic benchmark described as small-scale; RECAP also hinges on annotation.
Research summary
Nikki Lijing Kuang is a machine learning researcher at Amazon (United States) whose profile, resolved from OpenAlex bibliographic records, splits evenly across three directions — multimodal machine learning, efficient reasoning in large language models, and LLM agents with prompt optimization, each about 0.333 of the field distribution. Three 2026 arXiv papers in the Evolve catalog illustrate this range: redundancy-aware credit assignment for GRPO that cuts reasoning tokens by 8-31% on 7B math models while lifting pass@1; Uni-LaDiR, a shared-latent diffusion approach unifying multimodal reasoning with reported relative gains of 7.3% on a math VLM benchmark and 6.1% on RLBench; and RIPPLE, replay-gated prompt adaptation that raises Haiku validation by 23.1 points on a 39-task workflow-synthesis benchmark. Activity is the standout signal (radar 87): 19 works in the last three years and a 2026 burst of 15 outputs, after sparser years back to 2019. Across the reviewed papers Kuang appears as a middle author, with 23 distinct coauthors, a modest citation footprint (h-index 2, i10 index 1), and a reproducibility mean of 43, reflecting medium readiness for two papers and low for the small-scale agent benchmark.
Collaboration pattern
With 23 distinct coauthors and a collaboration score of 63, Kuang works within a recurring core group: Manqing Mao, Zhuoer Wang, Samson Koelle, Haoqiang Kang, Yizhe Zhang, Yian Ma, Lianhui Qin, Yuqing Zhou, Hong Wang, and Jie Yuan appear as frequent coauthors. All three 2026 catalog papers list Kuang as a middle author rather than lead or corresponding, consistent with a contributor role inside a larger team spanning efficient reasoning, multimodal diffusion, and agent workflows.
Research keywords
Highlights
Redundancy-aware credit assignment trims 8-31% of reasoning tokens
Shared-latent diffusion unifies multimodal reasoning
Replay-gated prompt edits lift workflow synthesis by 23.1 points
Timeline
- 2026Publishes 'Giving Credit Where It's Due', introducing redundancy-aware credit assignment for GRPO that cuts 8-31% of reasoning tokens on 7B math models.
- 2026Publishes 'Uni-LaDiR', unifying multimodal reasoning with a shared latent diffusion model and reporting gains on math VLM and RLBench.
- 2026Publishes 'Local Edits, Global Ripples', applying replay-gated prompt adaptation to workflow synthesis with a +23.1pp Haiku validation gain.
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 data is resolved from OpenAlex; the identity 'Nikki Lijing Kuang' is name-resolved from public bibliographic records, and the frequent-coauthor list derives from the same source.
- Output and impact dimensions are unscored because the underlying data is insufficient; h-index (2) and i10 index (1) are quoted as-is from the snapshot and likely reflect thin citation coverage of very recent arXiv work.
- Only the three papers reviewed by Evolve are covered here; reproducibility reflects Evolve's readiness assessments (medium/medium/low) rather than independent replication.
- The 2026 count (15 outputs) covers a partial, still-accumulating publication year and may be revised; sparse earlier years (2021 absent) suggest incomplete coverage.