Reinforcement learning researcher strengthening offline and online RL via data augmentation and prior distillation
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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 publication-volume data is available for this profile yet. This axis would measure total research output — number of papers, venues, and publication types — across the researcher's career; only two reviewed 2026 arXiv preprints are on record here.
Impact—
Citation and influence metrics are not available, so this axis is unscored. It would measure academic impact — citations, downstream adoption, and influence on later work — which cannot yet be assessed from the two recently posted preprints alone.
Focus0/100
Focus scores 0 on the 0-100 radar. The underlying field distribution splits evenly between two labels — reinforcement learning (0.5) and machine learning (reinforcement learning) (0.5) — so although the reviewed work sits entirely within reinforcement learning broadly, the even split across two overlapping field labels yields the minimum score. Both papers share the data-reuse theme, suggesting topical concentration beneath the label split.
Collaboration41/100
Collaboration scores 41 out of 100, based on 7 distinct coauthors recorded in the bibliometric data. The frequent-coauthor list centers on a small recurring team — Weidong Zhao and Xianhui Liu (the latter appearing twice in the records), plus Xiao Lai, Jiaji Shen, and Ning Jia — consistent with a compact working group rather than a wide collaboration network.
Activity—
No multi-year publication series is available, so activity is unscored. This axis would measure research cadence — papers per year, submission continuity, and sustained engagement over time; the record here covers only two 2026 preprints, which cannot establish a tempo.
Reproducibility38/100
Reproducibility averages 38 out of 100 across the 2 papers assessed. Evolve rates the BADA offline-RL paper's reproducibility readiness as medium and the B2PD online-RL paper as low; both papers are first-authored rather than corresponding, and the reviews flag thin or self-contradictory theory, limiting independent replication confidence.
Research summary
Gong Gao is a reinforcement learning researcher whose reviewed work to date consists of two first-authored 2026 arXiv preprints, both focused on helping RL agents learn more reliably from limited or biased experience. The first, BADA (Boundary-Aware Data Augmentation), targets offline RL: neighbor-bounded interpolation of logged data yields broad gains on the D4RL benchmark under scarce-data conditions, though Evolve's review notes the theory is thin and state-of-the-art claims overreach. The second, B2PD (Bidirectional Behavior Prior Distillation), addresses online RL by distilling Q-guided CVAE behavior priors into SAC and TD3, delivering strong benchmark gains and better sample efficiency, again with promising empirics but thin, partly self-contradictory theory. Together the two papers form one coherent methodological thread: using interpolation or generative data manipulation — Mixup-style interpolation in one, CVAE prior distillation in the other — to stretch limited experience further. Bibliometrically the profile is thin: radar scoring covers three axes only, with collaboration at 41 (7 distinct coauthors), reproducibility at 38 (mean readiness across 2 reviewed papers, one medium and one low), and focus scored 0 on an even 0.5/0.5 split between the reinforcement learning and machine learning (reinforcement learning) field labels. Output, impact, and activity are unscored for lack of data. Institution and country are not recorded; the profile is name-resolved from public bibliographic records tied to ORCID 0000-0003-0469-6028.
Collaboration pattern
The bibliometric record lists 7 distinct coauthors and a collaboration score of 41 out of 100. Recurring names — Weidong Zhao and Xianhui Liu (recorded twice), with Xiao Lai, Jiaji Shen, and Ning Jia also frequent — suggest a small, stable working group. Gong Gao appears as first author (not corresponding) on both reviewed papers, consistent with an early-stage position inside that team; the researcher's own name also surfaces in the frequent-coauthor records, likely a self-matching artifact.
Research keywords
Highlights
BADA: boundary-aware data augmentation lifts offline RL on scarce data
B2PD: Q-guided CVAE behavior priors distilled into SAC and TD3
One thread: stretching limited experience
Across both 2026 preprints, Gong Gao attacks the same problem from two sides — reusing logged data via interpolation for offline RL and compressing a learned behavior prior for online RL — a coherent agenda around generalization, robustness, and sample efficiency.
Timeline
- 2026First-authored arXiv preprint introducing BADA, boundary-aware data augmentation for offline RL, with broad D4RL gains on scarce data.
- 2026First-authored arXiv preprint introducing B2PD, bidirectional behavior prior distillation into SAC/TD3 for online RL.
- 2026Both papers enter the Evolve catalog, establishing a full profile scored on collaboration (41), reproducibility (38), and focus (0).
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 scores are computed from the provided public bibliometric snapshot only; no external citation or venue databases were consulted.
- No institution, country, or career history is recorded; the profile is name-resolved from public bibliographic data linked to ORCID 0000-0003-0469-6028.
- Output, impact, and activity radar axes are unscored because no publication counts, citation metrics, or multi-year publication series were available.
- Both assessed papers are 2026 arXiv preprints; peer-review status and citation counts are unknown, and only Evolve-reviewed papers were evaluated in depth.