Tongji University researcher bridging nonlinear dynamics and mechanistic interpretability of generative video models
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
21
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—
The output axis is not yet scored: this snapshot lacks a normalized publication-volume measure. It would compare cumulative and per-year output against field norms, with the 26 recent three-year works as its base signal.
Impact—
The impact axis is not yet scored: no citation-normalized impact measure (e.g. field-weighted citations) is available. It would gauge scholarly influence; the snapshot only reports raw h-index 21 and i10-index 34, which are not converted into a score here.
Focus0/100
The radar records a focus score of 0, consistent with the raw field distribution: computational nonlinear dynamics and machine learning (interpretability of generative video models) each carry 0.5, an even split rather than concentration. Topic data reinforce the breadth — 21 works tagged Advanced Neural Network Applications, 15 Neural Networks and Applications, 13 Domain Adaptation and Few-Shot Learning, 9 Model Reduction and Neural Networks, and 9 Stochastic Gradient Optimization Techniques. This reads as deliberate dual-track work, not drift.
Collaboration39/100
Collaboration scores 39/100. Over the last three years the OpenAlex record lists 26 works with only 6 distinct co-authors, a small, stable pool (Leo Yao, Max Tegmark, Xingyun Wang, Haomin Zheng, Man Yuan, Leqian Yang). Both reviewed 2026 preprints are multi-author efforts in which this researcher holds a middle or last author position without the corresponding-author role, suggesting contribution within a team rather than an independent lab signature.
Activity96/100
Activity is the strongest axis at 96/100. The counts-by-year series climbs from 5 works in 2019 through 8 in 2020, 11 in 2021 and 11 in 2022 to a peak of 18 in 2023, then 14 in 2024, with 26 works in the most recent three-year window and 7 already logged for 2025. Output continues into 2026, when both reviewed arXiv preprints appeared.
Reproducibility55/100
Reproducibility averages 55 across the 2 papers Evolve has assessed so far; both received a medium readiness rating. The verdicts note concrete gaps — the double-pendulum continuation study lacks head-to-head baselines, released code and stability analysis, while the video-model interpretability paper rests on synthetic dynamics only. A small sample, but consistent: rigorous evidence paired with unmet artifact-release expectations.
Research summary
Ziming Liu is a researcher at Tongji University whose OpenAlex-indexed profile spans two deliberately balanced tracks: computational nonlinear dynamics and machine learning research on the interpretability of generative video models, each carrying 0.5 of the field distribution. The bibliometric record shows steady acceleration — from a first indexed work in 2014, through 11 works in 2021, to a peak of 18 works in 2023, with 26 works in the most recent three-year window and an activity score of 96/100, the strongest axis in the radar. Cumulative indicators include an h-index of 21 and an i10-index of 34, though output and impact axes remain unscored for lack of normalized data. Topic clustering is led by Advanced Neural Network Applications (21 works), followed by Neural Networks and Applications (15) and Domain Adaptation and Few-Shot Learning (13). Two 2026 arXiv preprints have been reviewed by Evolve: a last-author study of causal writability in video models, which activation-patching experiments show stays rewritable until a sharp depth commitment boundary, and a middle-author variational continuation of double-pendulum periodic orbits. Both are rated medium for reproducibility, giving a mean of 55 across two assessed papers. Collaboration scores 39/100, built on a compact pool of 6 distinct recent co-authors. This is an early-stage, data-scoped profile; 'Ziming Liu' is a common name, so records are name-resolved from public bibliographic data tied to ORCID 0000-0002-7090-8853.
Collaboration pattern
Collaboration is moderate (39/100): 26 works in the last three years involve only 6 distinct co-authors, indicating a small, stable circle rather than a broad network. Recurring partners include Leo Yao, Max Tegmark, Xingyun Wang, Haomin Zheng, Man Yuan and Leqian Yang. In both reviewed 2026 preprints this researcher appears in a middle or last author position without the corresponding-author role, consistent with contributing inside an established team.
Research keywords
Highlights
Causal writability in video models: the chosen future stays rewritable until a sharp depth commitment boundary
Variational continuation of double-pendulum periodic orbits via Hessian eigenanalysis and automatic differentiation
A dual-track agenda bridging dynamical systems computation and deep-learning interpretability
The field distribution splits evenly (0.5/0.5) between computational nonlinear dynamics and machine learning research on generative video model interpretability, and the two reviewed 2026 preprints — one per track — share a quantitative, intervention-based style, with Hessian eigenanalysis and activation patching as methodological anchors.
Timeline
- 2014First work in the OpenAlex counts-by-year series (1 work), the start of the indexed publication record.
- 2021Annual output reaches 11 works, up from 5 in 2019, beginning a sustained double-digit run that repeats in 2022.
- 2023Peak recorded annual output: 18 works, with 14 following in 2024.
- 2026Both Evolve-reviewed arXiv preprints appear: causal writability in video models and variational continuation of double-pendulum periodic orbits.
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 come from OpenAlex only and are name-resolved via ORCID 0000-0002-7090-8853; 'Ziming Liu' is a common name, so conflation with same-name researchers cannot be fully excluded.
- Output and impact radar axes are unscored; only raw h-index (21) and i10-index (34) are reported, without citation-level verification.
- The reproducibility score (mean 55) rests on only 2 assessed papers, both 2026 arXiv preprints rated medium; coverage is thin.
- Counts for 2025-2026 reflect recent or in-progress periods and may change; topics and field shares are OpenAlex-derived labels, not self-declared expertise.