Danny Butvinik

As technological advancement meets regulatory compliance imperatives, financial institutions face a pivotal question: How can they harness Generative AI (GenAI) to enhance innovation, efficiency, and customer experience while adhering to strict governance and compliance standards? The EU’s AI Act, a critical piece of legislation, intensifies this query by demanding safety, transparency, and the protection of individual rights.

The article articulates the complex multi-head attention mechanism in LLMs and its unintended consequence of hallucinations. The focus is on three critical aspects—balancing, regularization, and optimization of information integration—and their impact on the efficacy and reliability of LLM outputs. Through a blend of theoretical analysis and practical examples, we explore how these mechanisms can either contribute to or mitigate the risk of hallucinations, providing a nuanced understanding of… Show more

The article articulates the complex multi-head attention mechanism in LLMs and its unintended consequence of hallucinations. The focus is on three critical aspects—balancing, regularization, and optimization of information integration—and their impact on the efficacy and reliability of LLM outputs. Through a blend of theoretical analysis and practical examples, we explore how these mechanisms can either contribute to or mitigate the risk of hallucinations, providing a nuanced understanding of the challenges and solutions in deploying LLMs effectively. This article is crafted for both theorists intrigued by the mathematical underpinnings and practitioners seeking to apply these insights in real-world scenarios. Show less

This article explores the nuanced impacts of temperature scaling on hallucinations within Large Language Models, categorizing them into factual and non-factual types. It highlights how temperature adjustments can influence neural network behavior, enhancing creativity or ensuring factual accuracy. Additionally, it introduces 'Silver Lining' hallucinations, where creatively valuable outputs emerge from otherwise inaccurate data. This comprehensive analysis provides foundational knowledge for… Show more

This article explores the nuanced impacts of temperature scaling on hallucinations within Large Language Models, categorizing them into factual and non-factual types. It highlights how temperature adjustments can influence neural network behavior, enhancing creativity or ensuring factual accuracy. Additionally, it introduces 'Silver Lining' hallucinations, where creatively valuable outputs emerge from otherwise inaccurate data. This comprehensive analysis provides foundational knowledge for optimizing LLM outputs, which is crucial for developers and researchers in AI and machine learning. Show less

Each year, financial crime costs the global economy trillions of dollars due to increasingly sophisticated deception. A technological paradox is emerging. Generative AI, a tool known for its creation ability, may hold the key to exposing and enabling fraud. Fueled by ever-evolving technology, fraudsters constantly devise new ways to exploit the system. Now, Generative AI enters the arena – a tool with the potential to both illuminate and obfuscate. This article delves into the heart of this… Show more

Each year, financial crime costs the global economy trillions of dollars due to increasingly sophisticated deception. A technological paradox is emerging. Generative AI, a tool known for its creation ability, may hold the key to exposing and enabling fraud. Fueled by ever-evolving technology, fraudsters constantly devise new ways to exploit the system. Now, Generative AI enters the arena – a tool with the potential to both illuminate and obfuscate. This article delves into the heart of this technological paradox, exploring how the same AI that can mimic human creativity can also unveil hidden fraud patterns. We'll explore the blurred lines between creation and detection, examining the promises and the perils of Large Language Models (LLMs) as they reshape the landscape of FinCrime. Show less

The rise of big data has spurred the development of online incremental machine learning, which is essential for real-time decision-making and automation in various industries. These algorithms adapt to new data as it arrives, efficiently managing continuous updates and concept drift in data streams. Key to these systems is balancing exploration with exploitation, which is crucial for handling non-stationary data. Despite challenges like concept drift and data volatility, online incremental… Show more

The rise of big data has spurred the development of online incremental machine learning, which is essential for real-time decision-making and automation in various industries. These algorithms adapt to new data as it arrives, efficiently managing continuous updates and concept drift in data streams. Key to these systems is balancing exploration with exploitation, which is crucial for handling non-stationary data. Despite challenges like concept drift and data volatility, online incremental learning has proven invaluable in applications ranging from fraud detection to predictive maintenance, driving forward research and the integration of real-time, automated technologies. Show less

While AI-based systems were primarily used to gain operational efficiencies in financial crime, model fairness and data bias occur when a system is skewed for or against certain groups or categories in the data. Typically, this stems from erroneous or unrepresentative data being fed into a machine learning model. Biased AI systems represent a serious threat, particularly when reputations are on the line. In fraud detection, biased data and predictive models can associate last names from other… Show more

While AI-based systems were primarily used to gain operational efficiencies in financial crime, model fairness and data bias occur when a system is skewed for or against certain groups or categories in the data. Typically, this stems from erroneous or unrepresentative data being fed into a machine learning model. Biased AI systems represent a serious threat, particularly when reputations are on the line. In fraud detection, biased data and predictive models can associate last names from other cultures with fraudulent accounts, or falsely decrease risk within population segments for certain types of financial activities. Show less

Exploration and Exploitation Tradeoff is a vital component within a learning systems which goals are to detect and prevent fraud. For a predictive machine learning model to be adequately adapt to evolving data by balancing between the exploration (update its decision border) and exploiting (do not update its decision border). Balancing between those actions ensures adjustable and flexible fraud detection over long period without significant degradation in performance.

The article summarizes the plethora of Uncertainty Quantification (UQ) methods using Bayesian techniques, shows issues and gaps in the literature, suggests further directions, and epitomizes AI-based systems within the Financial Crime domain.

This article introduces quantum artificial intelligence (QAI) and focuses on fundamental principles of QAI, and its various applications within financial crime services-based AI systems.

Widespread use of Artificial intelligence (AI) technologies in financial institutions and financial crime risk management systems have encouraged debate around the ethical challenges and risks AI-based technology pose. AI technologies have a significant impact on the development of humanity, and so they have raised fundamental questions about what we should do with these systems, what the systems themselves should do, what risks are potentially involved, and how we can control these.

Feature selection usually can lead to better learning performance, higher learning accuracy, lower computational cost, and better model interpretability. This article focuses on the Feature selection process and provides a comprehensive and structured overview of feature selection types, methodologies, and techniques from data and algorithm perspectives.

This article is about one of the most fascinating and challenging sub-domains in Computational Learning Theory — Online Machine Learning. Recently, online learning and incremental learning gained attention especially in the context of big data and learning from data streams, conflicting with the traditional assumption of complete data availability.

Wireless Body Area Networks (WBANs) are widely used for collecting and monitoring patients' vital healthcare parameters, such as breathing, heart function, and muscle activity. A serious flaw of WBANs is their vulnerability to various security issues, one of which is the physical tampering of the sensors. Transmission of invalid data by a damaged or compromised sensor may lead to an incorrect diagnosis, improper treatment, and undesirable results. In this paper, we analyze blood glucose-level… Show more

Wireless Body Area Networks (WBANs) are widely used for collecting and monitoring patients' vital healthcare parameters, such as breathing, heart function, and muscle activity. A serious flaw of WBANs is their vulnerability to various security issues, one of which is the physical tampering of the sensors. Transmission of invalid data by a damaged or compromised sensor may lead to an incorrect diagnosis, improper treatment, and undesirable results. In this paper, we analyze blood glucose-level sensors and propose a machine learning algorithm that detects intentional and inadvertent data modification intrusions for this type of sensor. Show less