Boston University machine learning researcher in RL-based LLM post-training and tabular foundation models
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
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 metrics are not yet available: beyond 6 works in the last three years, the publication-volume measures this axis needs are insufficient in the current OpenAlex snapshot.
Impact—
Impact is not yet scorable: citation counts, h-index and i10-index are not reliably available in the current snapshot. This axis would aggregate citations, h-index and i10-index to measure how widely the work is taken up.
Focus0/100
A focus score of 0 reflects breadth rather than concentration: OpenAlex topics spread across topic modeling, reinforcement learning in robotics and domain adaptation/few-shot learning (3 mentions each), with NLP techniques (2) and advanced neural network applications (1) added on top, and the field distribution splits exactly one-third across LLM post-training/RL and tabular deep learning/foundation-model tracks, matching the three Evolve-reviewed papers.
Collaboration62/100
Collaboration scores 62, supported by 21 distinct coauthors across 6 works in the last three years — a broad but not dense network. All three cataloged 2026 papers place Chris Kong in the last author position with no corresponding-author designation, consistent with senior-listed team membership rather than sole leadership. Recurring names include Gerald Friedland, Yefan Tao, Mingyu Chen, Xuezhou Zhang and Danielle Maddix, spanning both the RL and tabular foundation-model tracks.
Activity57/100
Activity scores 57 on a steady recent cadence: 6 works over the last three years, evenly split at 3 in 2025 and 3 in 2026, with all three Evolve-reviewed papers landing in 2026. The profile is compact and current — consistent year-over-year production rather than a large cumulative body — and overall volume remains modest relative to the mid-range score.
Reproducibility70/100
Reproducibility scores 70, the mean across the 3 Evolve-reviewed papers. Mitra-v2 rates high for unusually transparent evaluation, though its top rank is a statistical tie and v1-to-v2 gains are unattributed; MInTRL and Cliff rate medium, with reviewers flagging missing seed variance and single-run results without significance tests. The pattern suggests above-average openness on evaluation while statistical robustness reporting remains a gap.
Research summary
Chris Kong is a researcher at Boston University whose recent work centers on machine learning. OpenAlex-derived topic signals span topic modeling, reinforcement learning in robotics, and domain adaptation and few-shot learning (3 mentions each), with secondary activity in natural language processing techniques (2) and advanced neural network applications (1). The Evolve catalog links three 2026 arXiv papers, all listing Chris Kong in the last author position and none as corresponding author: MInTRL, which injects sparse teacher corrections into on-policy RL rollouts for LLM reasoning; Cliff, which learns process rewards from the first mistake; and the Mitra-v2 technical report on a 77M tabular foundation model pretrained on synthetic data. Count data show 6 works in the last three years, split 3 in 2025 and 3 in 2026, with 21 distinct coauthors. Field distribution splits evenly across LLM post-training/reinforcement learning and tabular deep learning/foundation models. Evolve's reproducibility assessment averages 70 across the three reviewed papers — Mitra-v2 rated high, the two RL papers medium — with reviewers flagging missing seed variance, single-run results, and unattributed v1-to-v2 gains. Citation-based output and impact measures are not yet available for this profile.
Collaboration pattern
Chris Kong works in team settings: 21 distinct coauthors across 6 recent works, with a recurring core of Gerald Friedland, Yefan Tao, Mingyu Chen, Xuezhou Zhang, Peixuan Han, Boran Han and Danielle Maddix. On all three Evolve-cataloged 2026 papers he appears in the last author position without corresponding-author designation, consistent with a senior collaborative role on LLM post-training and tabular foundation-model projects rather than solo-led efforts.
Research keywords
Highlights
MInTRL: sparse teacher corrections lift on-policy RL
Mitra-v2: a 77M tabular foundation model on synthetic data
Cliff: process rewards learned from the first mistake
Timeline
- 2025OpenAlex counts record 3 works in 2025, opening the current two-year run of output.
- 2026MInTRL appears on arXiv, reporting that sparse off-policy teacher interventions boost on-policy RL and outperform GRPO, OPD and MENTOR.
- 2026Cliff appears on arXiv, proposing process rewards learned from a trajectory's first mistake, reported to beat GRPO and OPD.
- 2026The Mitra-v2 technical report appears on arXiv, claiming SOTA-level performance for a 77M tabular foundation model trained on synthetic data.
- 2026OpenAlex records 3 works for 2026 overall; all three Evolve-reviewed papers list Chris Kong as last author without corresponding designation.
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, h-index and i10-index are not available, leaving output and impact unscored.
- No ORCID is linked, so the profile is name-resolved from public bibliographic data and attribution carries some ambiguity.
- Only the three 2026 arXiv papers reviewed by Evolve are individually assessed; other works appear only in aggregate counts.
- Reproducibility scores reflect Evolve reviewer judgment on 3 papers, not independent replication attempts.