Detection System of Decentralized Mixer using Transfer Learning from Centralized Mixers

Q: What does this research mean for the field?

Transfer learning can successfully apply behavioral patterns from centralized Bitcoin mixers to detect decentralized mixers and reduce labeling requirements, provided a minimum threshold of target data is used for fine-tuning. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Seongbin Kim; Hyunjun Jung

doi:10.14801/jkiit.2026.24.8.147

August 31, 2026The Journal of Korean Institute of Information Technology

Detection System of Decentralized Mixer using Transfer Learning from Centralized Mixers

View at Publisher

Ask AI

Bookmark

Share

Key Points

To determine whether behavioral patterns learned from centralized cryptocurrency mixers can be transferred to detect decentralized CoinJoin mixers on the Bitcoin network.
Extracted 10 behavioral features from the Bitcoin transaction graph.
Pre-trained a Graph Attention Network v2 (GATv2) architecture on centralized mixer data (BitcoinFog, ChipMixer) and fine-tuned it on varying label portions of decentralized Wasabi Wallet data.
Zero-shot transfer displayed a severe domain gap (Wasabi F1 score: 0.013 despite a source Macro F1 of 0.914), and fine-tuning with 10% labels resulted in negative transfer.
Fine-tuning with 25% target labels achieved a Wasabi F1 score of 0.984 ± 0.003, outperforming both the 25% randomly initialized model (F1: 0.972 ± 0.004) and the 100% randomly initialized model (F1: 0.978 ± 0.001).

Implication

Machine learning study reveals effective detection of decentralized CoinJoin mixers using centralized mixer pre-training, indicating significant label-saving efficiency across cryptocurrency...

Abstract

No abstract yet. The journal hasn't published one we can show.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Kim et al. (2026) studied this question.

synapsesocial.com/papers/6a9934b4636c6408cfa7c7b3 https://doi.org/10.14801/jkiit.2026.24.8.147

View at Publisher

Ask AI

Bookmark

Share

Detection System of Decentralized Mixer using Transfer Learning from Centralized Mixers

Key Points

Implication

Abstract

Discussion

Cite This Study

Detection System of Decentralized Mixer using Transfer Learning from Centralized Mixers

Key Points

Implication

Abstract

Discussion

Cite This Study

Keep reading

More related papers

More from The Journal of Korean Institute of Information Technology

Keep reading

More related papers

More from The Journal of Korean Institute of Information Technology