Day 3: Thorsten Neumann - AI Driven Risk Assessment in Digital Asset Markets | ADIA Lab Symposium

Day 3: Thorsten Neumann - AI Driven Risk Assessment in Digital Asset Markets | ADIA Lab Symposium

by Thorsten Neumann

Applied Sciences & Engineering Economics & Finance KFFFinance and the finance industryKFFFFinancial technology (fintech)
🎙 Thorsten Neumann 👥 824 📅 November 5, 2025 ⏱ 24 min 👁 40 📄 expert opinion 🧭 2026-08-16
Available in: English (current) Français

Keywords

AIrisk assessmentdigital assetsdata distillationfinancial crime

Summary

Thorsten Neumann, from Standard Chartered, presents a talk on AI-driven risk assessment in digital asset markets. He begins by outlining the challenges of financial crime monitoring in traditional banking, emphasizing the limitations of rule-based systems and the need for more adaptive, data-driven approaches. He then introduces the complexities introduced by DeFi and digital assets, such as pseudonymity, real-time flows, and new types of fraud. The core of the talk focuses on a data distillation technique using transformer models to create synthetic datasets that can be shared between institutions without revealing sensitive customer information. Neumann presents experimental results showing that distilled datasets, even at 12% of the original size, can enable other institutions to achieve up to 93% accuracy in fraud detection. He discusses various sharing mechanisms, including central registries and blockchain-based infrastructures. He also highlights the importance of explainability in AI models for regulatory compliance and the need for further research in this area. The talk concludes with future directions, including extending the research to other blockchain networks, enriching metadata, and applying risk assessment to tokenized assets themselves.

178 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides valuable insights into the practical application of AI for financial crime detection in digital asset markets. The speaker presents a novel approach using data distillation to enable collaborative learning without compromising privacy. The argumentation is coherent, moving from problem definition to solution and experimental validation. However, the presentation is high-level and lacks detailed technical depth, and the experimental results are presented without rigorous statistical analysis. The speaker’s expertise is evident, but the claims would benefit from peer-reviewed publication.

Scientific Rigor, Source Quality, Title Accuracy

The talk references several public datasets from companies like Elliptic, Chainalysis, and Binance, which are credible sources in the blockchain analytics space. The speaker also mentions academic concepts like graph neural networks and data distillation, but does not provide specific citations. The title accurately reflects the content, and the talk is well-structured. However, the lack of formal citations and the reliance on anecdotal evidence reduce the scientific rigor. The speaker’s position at a major bank lends credibility, but the presentation is more of an industry perspective than a rigorous academic study.

187 words

Title / Content Match

The title accurately reflects the content, which focuses on AI-driven risk assessment in digital asset markets, as presented by Thorsten Neumann.

Quality & Reliability

7/10

The speaker is a practitioner from a major financial institution, presenting applied research and industry insights. The talk includes specific technical details and references to public datasets, but lacks formal peer-reviewed validation. The claims about data distillation are plausible but not independently verified in this context.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

  • Elliptic Data Sets — Mentioned as a source of flagged transaction data sets for training models.
  • Chainalysis — Mentioned as a provider of blockchain data and analytics.
  • Binance Data Sets — Mentioned as a source of transparent on-chain transaction data.

Concurring Sources

  • Elliptic Data Sets — Mentioned as a source of flagged transaction data sets for training models.
  • Chainalysis — Mentioned as a provider of blockchain data and analytics.
  • Binance Data Sets — Mentioned as a source of transparent on-chain transaction data.

Contribution & Novelties

The talk presents a practical application of data distillation for privacy-preserving collaboration in financial crime detection. The speaker demonstrates that distilled datasets can be shared between institutions to improve model accuracy without revealing sensitive information. This approach has potential to overcome data-sharing barriers in the financial industry. The talk also highlights the importance of explainability and the need for further research in this area.

Pour aller plus loin :

  • Data Distillation — Provides background on the concept of distilling knowledge from models.
  • Graph Neural Networks — Relevant to the use of graph-based models for transaction analysis.
  • Federated Learning — An alternative approach to collaborative learning without sharing raw data.

109 words

Radar Profile

The radar profile shows a balanced performance across all dimensions, with slightly higher scores in information quantity and technical level, indicating a content-rich and technically oriented presentation. The lower score in reliability suggests that while the information is valuable, it lacks formal verification.

Reliability 6/10