Dunamu places three blockchain AI papers at ICAIF 2026
Dunamu's research covers exchange wallet detection on Ethereum, AI-based asset classification and on-chain fund tracing, with one paper in the main conference.
Dunamu announced that three of its research papers on AI-based blockchain data analysis were accepted at the ICAIF 2026 international financial AI conference. One paper was accepted for the main conference and two for specialized workshops, according to techm.kr.
ICAIF is organized by ACM and was held this year at Bocconi University in Milan, Italy, from the 14th to the 17th. More than 1000 papers were submitted to this year's conference, and the main conference acceptance rate was about 22%.
The main-conference paper presents a machine-learning method that uses exchanges' Proof-of-Reserves data to identify exchange-owned wallets on Ethereum. The research team analyzed about 3.03 million Ethereum wallet addresses based on Proof-of-Reserves data from 15 exchanges, extracting 58 behavioral features per wallet, including transaction patterns, token transfers and gas usage. The model achieved an average PR-AUC of 0.9594 when tested on exchange wallets not used in training, and part of the technology has been applied to the blockchain analysis service of Upbit Data Lab.
Asset classification and fund tracing
A second paper uses large language models to analyze smart contract code and classify digital assets by their actual financial function. In experiments on Ethereum-based DeFi protocols including Uniswap, Aave and Curve, 16 of 17 analyzed assets matched existing classification criteria, about a 94% match rate. Dunamu plans to use the technology to improve its Upbit digital asset classification system (UBCS).
A third paper analyzes the range within which a specific deposit's later use in payments can be estimated when transaction records are incomplete, and will be used to improve the analytical reliability of Dunamu's on-chain fund tracing solution OTS (Onchain AI Tracer System).
Dunamu's Chief Data Officer Kim Dae-hyun said the research confirmed the potential of using AI to analyze blockchain data more precisely and improve the reliability of results.
Quick answers
What did Dunamu present at ICAIF 2026?
Three research papers on AI-based blockchain data analysis: one for the main conference and two for specialized workshops.
What was the main conference acceptance rate at ICAIF 2026?
About 22%, from more than 1000 submitted papers.
Where will Dunamu apply the research?
The asset classification technology will feed into the Upbit digital asset classification system (UBCS), and the fund-tracing research into the OTS solution.