Shanghai Jiao Tong researcher accelerating LLM and diffusion inference and managing agent memory across 2026 preprints
3
papers here
3
reviewed by Evolve
0
first-author
1
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—
This axis would measure publication volume and citation-derived counts such as h-index and i10-index. The OpenAlex record currently reports both indices as 0 with no aggregate totals, so output cannot be scored yet and is intentionally left unscored.
Impact—
Impact measures how far the work has been cited and taken up by others. Every catalogued paper is a 2026 preprint and the record holds no citation totals, so this axis cannot yet be scored.
Focus0/100
The focus score is low (0) because the field distribution is evenly split: computer vision — efficient generative model inference (33.3%), machine learning / efficient LLM inference (33.3%), and artificial intelligence — LLM agent memory systems (33.3%). No direction yet dominates the record, which reads as breadth-first exploration of inference-efficiency problems rather than an established specialty; concentration may emerge as the portfolio matures.
Collaboration65/100
Collaboration scores 65, supported by 25 distinct co-authors in the recent three-year window and a recurring core of 12 named collaborators, including Jinkui Ren, Zhirong Shen, Rui Huang, and Chang Zou. Author-position data shows a senior tilt: last author on MemForest and the feature-cache paper (also corresponding), and middle author on DFlow.
Activity61/100
Activity scores 61, backed by 7 works in the recent three-year window and a steep rise in OpenAlex counts-by-year from 1 work in 2025 to 6 in 2026, three of which are already reviewed in the Evolve catalog. The cadence points to an intensive 2026 publishing phase concentrated on inference-efficiency topics.
Reproducibility55/100
Reproducibility scores 55 (reproMean) across 3 assessed papers, all rated medium readiness. Evolve's verdicts flag concrete gaps: MemForest reports no significance tests with confounded AGPR attribution; DFlow is Qwen-only with no code release; and the feature-cache paper's abstract numbers do not match its tables.
Research summary
Linfeng Zhang is a researcher at Shanghai Jiao Tong University whose recent work concentrates on making large-model inference and agent systems more efficient. Identity is name-resolved from public OpenAlex bibliographic records; no ORCID is provided. The record lists 7 works in the recent three-year window, rising from 1 indexed work in 2025 to 6 in 2026, with 25 distinct co-authors. Three 2026 arXiv preprints have been reviewed in the Evolve catalog across three evenly weighted directions (roughly one third each): efficient generative model inference in computer vision, efficient LLM inference in machine learning, and LLM agent memory systems. MemForest introduces EventTree partitioning with progressive merging for agent memory, reporting 50% memory compression at 97-99.7% retention. DFlow relays verifier hidden states in block diffusion speculative decoding, lifting acceptance length by roughly 10-13.4% on Qwen3. A third preprint uses Gaussian process regression to correct diffusion transformer feature-cache drift with uncertainty-gated recomputation. Evolve rates reproducibility readiness as medium for all three papers (mean 55), citing missing significance tests, Qwen-only evaluation, no released code, and abstract-table discrepancies in one paper. All three are 2026 preprints, so citation-based output and impact scores are not yet available, and this profile should be read as early-stage.
Collaboration pattern
Across the three catalogued 2026 preprints, Zhang sits in middle and last author positions, including one corresponding authorship on the diffusion-transformer caching paper. The record shows 25 distinct co-authors in the recent three-year window, with a recurring core of 12 collaborators such as Jinkui Ren, Zhirong Shen, Rui Huang, and Chang Zou — a team-based pattern typical of systems-oriented ML groups.
Research keywords
Highlights
MemForest: EventTree-based agent memory compression
DFlow: verifier information flow for block diffusion speculative decoding
Uncertainty-gated feature caching for diffusion transformers
Timeline
- 2025First indexed work appears in OpenAlex counts-by-year (1 work), opening the visible record.
- 2026MemForest preprint on EventTree-based agent memory management (last author) is reviewed in the Evolve catalog.
- 2026DFlow preprint brings verifier information flow into block diffusion speculative decoding (middle author).
- 2026Gaussian-process rectified feature-cache paper (last and corresponding author); OpenAlex counts 6 works in 2026.
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 OpenAlex only; h-index and i10-index are recorded as 0, indicating incomplete citation coverage rather than measured zero impact.
- "Linfeng Zhang" is a common name; this profile is name-resolved from public bibliographic data and no ORCID was available for disambiguation.
- Only three 2026 arXiv preprints have been reviewed by Evolve; peer-review status and final venues are unknown.
- Focus is scored from an even three-way field split on a thin record and may change as more works are indexed.