RMIT researcher bridging LLM-agent privacy and multimodal content moderation; last author on two 2026 arXiv papers
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 output score is available yet: the current snapshot lacks sufficient publication-volume data for this researcher. This axis would measure total papers, venue spread and yearly productivity once a fuller bibliographic record is linked.
Impact—
No impact score is available: citation counts, h-index and venue influence are absent from the supplied data. This axis would measure citation-based scholarly influence and reach; it is left unscored rather than estimated.
Focus0/100
The focus score is 0, derived directly from an even field distribution: 50% AI security and privacy around LLM agents and 50% multimodal natural language processing for content moderation. The two catalog papers mirror this split exactly — CIPL on agent privacy, ProKDA on hateful meme detection — so the portfolio reads as a deliberately bifurcated two-line agenda rather than concentration in one area.
Collaboration61/100
Collaboration scores 61, grounded in 20 distinct co-authors across the record. The frequent-coauthor list reveals a compact core: Chen Hou recurs three times, Tao Huang, Guosen Wu and Guolong Zheng each appear twice, alongside Jiayang Meng, Xu Yang and Xuechao Yang. On both catalog papers Feng Xia holds the last-author position, consistent with a senior supervisory role inside a stable, recurring team.
Activity—
No activity score is available: recent publishing cadence cannot be computed from a snapshot that resolves only two 2026 papers. This axis would measure ongoing publication tempo and research engagement; it is left unscored.
Reproducibility55/100
Reproducibility scores 55, the mean across the two reviewed papers, each rated medium readiness. Evolve's verdicts flag the same gap in both: CIPL's results are coherent across five providers but lack statistical testing, while ProKDA's gains over coupled explain-then-detect SFT are strong but lack significance tests. Both entries are arXiv preprints, so this catalog reflects no peer-review verification.
Research summary
Feng Xia is a researcher at RMIT University whose work spans two equally weighted areas: AI security and privacy around LLM agents, and multimodal natural language processing for content moderation. The Evolve catalog carries two 2026 arXiv papers, on both of which Feng Xia is the last author. CIPL formalizes channel-aware, recoverable privacy leakage in LLM agents, framing exposure versus recovery for black-box evaluation across five providers. ProKDA targets explainable hateful meme detection, progressively aligning knowledge to decision by decoupling knowledge, detection and DPO stages. The field distribution splits exactly 50/50 between these two areas, which yields a focus score of 0 — a deliberately bifurcated portfolio rather than a single-line agenda. Collaboration scores 61, built on 20 distinct co-authors with a recurring core of Chen Hou, Tao Huang, Guosen Wu and Guolong Zheng. Reproducibility averages 55 across the two reviewed papers, both rated medium readiness: results are coherent, but neither reports statistical significance testing. Output, impact and activity scores are not yet computable. With only two reviewed entries and no citation data, this remains an early-stage profile with room to deepen.
Collaboration pattern
Feng Xia's collaboration footprint spans 20 distinct co-authors, with a stable inner core: Chen Hou is the most frequent partner (three appearances on the frequent-coauthor list), followed by Tao Huang, Guosen Wu and Guolong Zheng (two each), plus Jiayang Meng, Xu Yang and Xuechao Yang. Both catalog papers are last-authored, suggesting a senior, supervisory position within a consistent team rather than loose, one-off partnerships.
Research keywords
Highlights
CIPL: formalizing recoverable privacy leakage in LLM agents
ProKDA: staged knowledge-to-decision alignment for explainable hateful meme detection
Timeline
- 2026CIPL, a channel-aware framework for recoverable privacy leakage in LLM agents, released on arXiv with Feng Xia as last author.
- 2026ProKDA, progressive knowledge-to-decision alignment for explainable hateful meme detection, released on arXiv with Feng Xia as last author.
- 2026Both papers enter the Evolve catalog with medium reproducibility readiness, anchoring a 50/50 split between LLM-agent privacy and multimodal content moderation.
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.
- No ORCID is provided and 'Feng Xia' is a fairly common name; this profile is name-resolved from public bibliographic data and identity consolidation may be imperfect.
- Output, impact and activity dimensions lack underlying data: citation counts and a full publication history are not part of the current snapshot.
- Only two papers have been reviewed in the Evolve catalog, so the profile reflects a narrow slice of this researcher's output.
- Both catalog entries are 2026 arXiv preprints; peer-review status and code or data availability are not captured.