Counting HyperGraphlets via Color Coding: a Quadratic Barrier and How to Break It
2026-04-15T08:52:24Z•a57050be6b1bf11e6a2e9dc78f5003814e68da4bfc14749b2ca32a0f5e24b665
DeFiLLMalgorithmsblockchaincross-chaindataset-releasedeanonymizationdetection-evasiondual-usefraud-detectiongraph-analysishypergraphmobility-datamultimodalopen-sourceprivacy
What happened
This feed contains multiple CS/AI papers with one security-relevant contribution: “UniDetect” (arXiv:2604.12329) is an LLM-driven, multimodal method for multi-chain cryptocurrency fraud-account detection that (1) generates human-readable transaction summaries as evidence, (2) combines textual evidence with transaction graph patterns via a two-stage alternating training strategy, (3) reports high detection (≈94.6% cross-chain zero-shot) and KS/F1 improvements, and (4) releases dataset and source code. Relevant secondary issues across the feed: mobility and place-exposure studies using large, ge
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_si
- Record identifier
- a57050be6b1bf11e6a2e9dc78f5003814e68da4bfc14749b2ca32a0f5e24b665
- Enrichment time
- 2026-04-15T08:52:24Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.