Tongji University researcher on data augmentation and prior distillation for offline and online reinforcement learning
2
papers here
2
reviewed by Evolve
0
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 counts, venue list, or yearly output totals are available in this snapshot, so the output axis is unscored. It would measure cumulative research productivity across venues and years once a fuller bibliographic record is linked.
Impact—
Citation counts, h-index, and citing works are not yet available, so impact is unscored. This axis would measure how far the BADA and B2PD methods have been adopted by subsequent reinforcement learning research.
Focus0/100
The focus axis is scored 0 in the current snapshot. The underlying field distribution contains only two closely related labels — reinforcement learning (0.5) and machine learning (reinforcement learning) (0.5) — drawn from just two reviewed papers, so topical breadth cannot yet be meaningfully assessed; the visible record points to a single concentrated agenda around RL data augmentation and prior distillation.
Collaboration41/100
Collaboration scores 41 in the snapshot, drawn from 7 distinct coauthors. The recurring partners are Gong Gao and Xianhui Liu, with Xiao Lai, Jiaji Shen, and Ning Jia also frequent, indicating a compact, stable core team typical of a single lab rather than broad cross-institution collaboration; across the two reviewed 2026 papers he is a middle author once and last author once, never corresponding.
Activity—
Annual publication rates and recency streaks are not available in the current data, so activity is unscored. It would measure publishing tempo and continuity over recent years from a complete, disambiguated publication record.
Reproducibility38/100
Reproducibility averages 38 across the 2 papers scored in the catalog. The offline RL data-augmentation paper (BADA) reaches medium readiness, while the online B2PD distillation paper stays low, indicating that benchmark-level empirical claims are not yet matched by released code or complete experimental detail; raising code and hyperparameter documentation transparency would be the fastest route to a stronger score.
Research summary
Weidong Zhao is a reinforcement learning researcher at Tongji University (ORCID 0009-0009-2091-3900) whose work targets the sample-efficiency and generalization limits of modern RL in both offline and online settings. Evolve's catalog includes two of his 2026 arXiv preprints. The first, Boundary-Aware Data Augmentation (BADA), introduces neighbor-bounded interpolation for offline RL and delivers broad D4RL gains under scarce data; Evolve judges the empirical results credible while noting thin theory and overstated SOTA claims. The second, Bidirectional Behavior Prior Distillation (B2PD), distills Q-guided CVAE behavior priors into SAC and TD3 agents, reporting strong online RL benchmark gains and improved sample efficiency; the verdict again praises the empirics but flags thin, self-contradictory theory. His listed fields split evenly between reinforcement learning and machine learning (reinforcement learning), and the catalog work shows one consistent pattern: reusing generative or interpolation machinery to regularize RL value learning. He publishes in small teams — 7 distinct coauthors appear in his record, with Gong Gao and Xianhui Liu the most frequent — typically as a middle or last author rather than corresponding author. Reproducibility scoring covers 2 papers with a mean of 38, indicating early-stage code and documentation practices.
Collaboration pattern
Zhao works in small, stable teams: the snapshot lists 7 distinct coauthors, with Gong Gao and Xianhui Liu appearing most frequently, followed by Xiao Lai, Jiaji Shen, and Ning Jia. His collaboration score is 41, consistent with a compact core group rather than a large distributed network; on the two reviewed 2026 papers he appears as a middle author on the offline RL study and last author on the online RL study, neither as corresponding author.
Research keywords
Highlights
BADA: boundary-aware data augmentation for offline RL
B2PD: bidirectional behavior prior distillation for online RL
Timeline
- 2026Posts Boundary-Aware Data Augmentation (BADA) on arXiv, extending offline RL with neighbor-bounded interpolation validated on D4RL benchmarks.
- 2026Posts Bidirectional Behavior Prior Distillation (B2PD) on arXiv, distilling Q-guided CVAE priors into SAC/TD3 for online RL sample efficiency.
- 2026Both preprints enter the Evolve reviewed catalog, which scores reproducibility across 2 papers (mean 38) and records 7 distinct coauthors.
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 indicators come from a small OpenAlex-derived snapshot; no citation counts, h-index, or total publication counts were available, so the output, impact, and activity axes are unscored.
- The evidence base is two 2026 arXiv preprints; peer-reviewed venue acceptance is not yet recorded.
- Reproducibility scoring covers only the 2 reviewed papers (mean 38) and may not reflect unreleased code or supplementary materials.
- The profile is name-resolved from public bibliographic data (ORCID 0009-0009-2091-3900); Weidong Zhao is a common name, so identity matching beyond the ORCID and Tongji University affiliation is not guaranteed.