KONTOGRAPH: Verified Point-in-Time Feature Consistency and Amortised Explanation for Real-Time Anti-Money Laundering under a 200 ms Decision Budget
Researchers have developed KONTOGRAPH, an anti-money laundering (AML) pipeline designed to operate within a tight 200 ms decision budget. The system uses a temporal graph network with per-node memory and achieves improved performance over traditional methods. An empirical study on simulated payments found that the pipeline's deployed tree ensemble can produce different results when converted to ONNX format due to numerical precision issues, highlighting the need for careful m
Researchers have developed KONTOGRAPH, an anti-money laundering (AML) pipeline designed to operate within a tight 200 ms decision budget. The system uses a temporal graph network with per-node memory and achieves improved performance over traditional methods. An empirical study on simulated payments found that the pipeline's deployed tree ensemble can produce different results when converted to ONNX format due to numerical precision issues, highlighting the need for careful model deployment and evaluation.
---
Why it matters: This matters because it addresses a critical challenge in real-time AML: making accurate decisions within extremely short timeframes. Engineers working on similar systems will be interested in the techniques used by KONTOGRAPH and the implications of its findings for model deployment and evaluation.
Source: https://arxiv.org/abs/2608.22389
This article was originally published at: https://arxiv.org/abs/2608.22389