I am an Assistant Professor of Artificial Intelligence in Finance. My research focuses on creating secure and transparent AI systems, particularly for the finance, law, and insurance industries. Most of my research repositories are private, you can find an overview of my key projects below.
- Development of explainable AI and machine learning algorithms
- Human-AI augmented decision-making in finance, insurance, and law
- Blockchain technology for automated auditing of AI decisions
- Cryptographic security for AI systems
🔭 I’m currently working on projects involving Explainable AI (XAI) for financial inclusion and Blockchain to enhance the cryptographic security of AI systems.
Link to PAPER:
This work validates Daniel Kahneman's decision theory on "Noisy" Human Judgment and shows how AI and Human collaboration can offset the gap between past inconsistent decisions and the ultimate true decisions by the Evidential Reasoning-eXplainer (ER-X) algorithm.
Link to PAPER:
Consistency analysis of 36 recurrently funded cases. The line plot shows the probability mass for the fund {F} decision. The remaining probability mass is assigned to reject {R} decision by data and not-sure {F, R} by underwriters.
The probability mass through line plots and degree of credibility (relative consistency) among four domain experts and data by heat maps for the clear-cut decline, recurrently funded, and rejected cases, respectively. In these figures, the junior underwriters are designated with labels '1' and '2', while senior underwriters are marked as '3' and '4'. The scattered judgment for single pieces of evidence in clear-cut cases and multiple pieces of evidence in recurrent cases is visually represented by variations in color intensity. The degree of credibility does not vary much among underwriters and data for clear-cut decline criteria and recurrently rejected loan applications compared to funded loan applications. The probability mass for an outcome by data reflects the collective judgment by multiple underwriters in the past at time t, whereas the subjective judgment by underwriters through online assessment was obtained at the time t^', such that t
Link to PAPER:
4. Human-in-the-Loop Framework to mitigate human error and malicious intent by eXplainable Deep Neural Network (xDNN) and cryptographic security through blockchain
Link to PAPER:
Internal and External Legal Data Management by Blockchain
5. Blockchain Framework in Compliance with Data Protection Law to Manage and Integrate Human Knowledge by Fuzzy Cognitive Maps: Small Business Loans
New firms have minimum accessibility to data, unlike large data repositories of businesses operating for longer periods. They rely on the knowledge exchange and perceptions of domain experts outside the organisation to enhance the capability of the decision support systems. This paper presents a framework on blockchain technology to develop a secure knowledge management service which addresses the challenges imposed by European and US data protection laws and high on-chain computational costs. It presents a hybrid off-chain and on-chain smart contract computation methodology for the scalable application of blockchain technology. It requires low on-chain storage cost and low cost of frequent data imports for off-chain AI computations. The framework demonstrates a real-world application of knowledge management by blockchain to integrate the judgment of multiple financial experts to design loan eligibility policies.
Link to PAPER:
Architecture to Manage Knowledge-Base by Blockchain
Average latency and throughput of 800 transactions
Off-chain knowledge integration by FCM to assess loan eligibility dynamics by 12 small business loan experts in four organisations
6. Optimal data-driven strategy for in-house and outsourced technological innovations by open banking APIs
Link to PAPER:
