Machine learning researcher framing long-horizon personalized AI through non-uniform geometric memory
1
papers here
1
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-20) 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 underlying publication-volume data is available yet, so this axis is unscored. Output would measure the number and cadence of indexed publications over time; it can be computed once more of this researcher's record enters the bibliographic snapshot.
Impact—
Citation-based data is not yet available for this profile, so impact is unscored. The axis would measure citations, citation velocity, and downstream influence of the work, and cannot be responsibly estimated from a single unreviewed 2026 preprint.
Focus100/100
Focus scores a perfect 100 because the field distribution lists exactly one field — machine learning / data mining with a personalized AI and user modeling emphasis — for the one reviewed paper. The 2026 preprint stays entirely inside that lane, uniting memory design, personalization, and geometric representation in a single agenda. With a sample of one paper, however, this concentration is best read as deliberate positioning rather than an established pattern.
Collaboration32/100
Collaboration scores 32, drawn from four distinct coauthors in the raw data. The recurring circle is small and specific — Wenhao Yu, Zexuan Qiu, Menglin Yang, and Irwin King — suggesting close-knit team science rather than broad networked collaboration. Because this rests on a single first-authored 2026 preprint, the score captures an early snapshot of working relationships, and the pattern may widen as more publications appear.
Activity—
Recent-activity data is insufficient to score this axis. Activity would track publication frequency and recency — for example outputs per year — but the current snapshot contains only one 2026 preprint, too thin to characterize a working cadence.
Reproducibility20/100
Reproducibility scores 20, a mean computed over one paper (reproPapers: 1). Evolve rated the 2026 preprint's reproducibility readiness as low, consistent with its verdict of a purely conceptual vision paper with no experiments yet. With no code, benchmarks, or empirical results in the reviewed record, the low score reflects an agenda-setting stage of work rather than failed replication, and should firm up if experimental follow-ups appear.
Research summary
Jiahong Liu is an early-stage machine learning researcher whose reviewed work centers on personalized AI and user modeling. The Evolve catalog currently contains one paper, the 2026 arXiv preprint 'Memory Has Geometry: Non-Uniform Geometric Memory for Long-Horizon Personalized AI', on which Liu is first author. Evolve's verdict characterizes it as a vision paper: it frames long-horizon memory as trajectory-conditioned estimation on non-uniform geometry — drawing on Riemannian geometry and hyperbolic embeddings — and judges the agenda compelling but purely conceptual, with no experiments yet. The bibliometric snapshot reflects this early posture: the field distribution lists a single field (machine learning / data mining, personalized AI and user modeling), giving a maximal focus score of 100; four distinct coauthors yield a collaboration score of 32; and reproducibility readiness for the reviewed paper is rated low, with a mean of 20 across one paper. Collaboration indicators point to a small recurring circle — Wenhao Yu, Zexuan Qiu, Menglin Yang, and Irwin King. Because the output, impact, and activity axes lack sufficient underlying data, this profile should be read as an early-stage, name-resolved view built from public bibliographic data, and future experimental or benchmark work would materially change the picture.
Collaboration pattern
The collaboration footprint is compact: four distinct coauthors drive a collaboration score of 32. The recurring names — Wenhao Yu, Zexuan Qiu, Menglin Yang, and Irwin King — indicate a small, stable working circle around the 2026 personalized-AI preprint on which Liu is first author. Because this rests on one reviewed paper, the pattern reads as an early team rather than an established network, and its breadth cannot yet be characterized.
Research keywords
Highlights
Memory as geometry: a non-uniform geometric memory for long-horizon personalization
An explicitly conceptual agenda, judged compelling
Early reproducibility profile leaves clear room to grow
Timeline
- 2026Posts the first-authored arXiv preprint 'Memory Has Geometry: Non-Uniform Geometric Memory for Long-Horizon Personalized AI', which enters the Evolve catalog as a first-author paper in machine learning / data mining (personalized AI, user modeling).
- 2026Evolve's review classifies the paper as a vision paper: long-horizon memory framed as trajectory-conditioned estimation on non-uniform geometry — compelling, but purely conceptual with no experiments yet.
- 2026Reproducibility readiness for the preprint is assessed as low, anchoring an initial reproducibility mean of 20 across one reviewed paper alongside focus 100 and collaboration 32.
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 are computed from an OpenAlex-derived snapshot only; the output, impact, and activity axes currently lack sufficient underlying data and are left unscored.
- Profile is name-resolved from public bibliographic data; Jiahong Liu is a common name, and no institutional affiliation or ORCID was available to confirm identity.
- Only one reviewed paper (a 2026 arXiv preprint) underpins this profile, so it is an early-stage view and all scores may shift substantially with new publications.
- The reproducibility score of 20 is a mean over a single paper and should not be read as a stable track-record measure.