Bayan Bruss, the VP of AI Foundations at Capital One, collaborates with academic researchers to apply cutting-edge AI in finance. He delves into the complexities of out-of-distribution (OOD) detection, emphasizing its significance for safety in AI applications. The discussion also covers the challenges of integrating AI research into corporate environments, the need for model explainability, and the cautious adoption of generative AI. Bruss highlights innovations and limitations in current AI models, underlining the importance of real-world testing to ensure robustness and reliability.
Read more
AI Summary
Highlights
AI Chapters
Episode notes
auto_awesome
Podcast summary created with Snipd AI
Quick takeaways
Out-of-distribution detection is a critical challenge in AI, requiring models to adapt to unseen data to avoid incorrect predictions.
Capital One emphasizes the importance of AI explainability in high-stakes fields like banking to ensure stakeholder trust in algorithmic decisions.
Deep dives
Understanding Out of Distribution Detection
Out of distribution (OOD) detection involves identifying data that does not conform to the distribution that a machine learning model was trained on. This situation arises when a model encounters unseen data that is different from its training examples, which can lead to incorrect predictions. For instance, a classifier trained to differentiate between cats and dogs may struggle when faced with entirely new animal species, such as parakeets. The challenge lies in creating models robust enough to handle such variations and potentially adapting or detecting failures in real-time.
The Journey of AI Applications in Financial Services
Bayan Bruce discusses his journey with Capital One, where applied AI research plays a crucial role in enhancing financial services. The company has made significant strides in integrating machine learning across various operations, from fraud detection to customer service. The migration to the cloud has facilitated the use of advanced models that rely on extensive customer data, enabling Capital One to meet consumer needs effectively. The iterative process of moving from academic research to practical applications highlights the importance of contextual understanding within the business environment.
Challenges of Explainability and Trust in AI Models
The podcast emphasizes the critical issue of explainability in AI systems, especially within high-stakes fields like banking. Trustworthiness is essential, as stakeholders require confidence that algorithms are making decisions based on correct and relevant factors. Capital One has prioritized research to improve model explainability, addressing concerns about how AI learns and applies its knowledge in real-world scenarios. The need to ensure that models do not rely on spurious correlations, which can lead to flawed decision-making, remains a challenge in AI development.
Navigating the Future of AI and Human Interaction
The discussion touches on the role of assistive technologies in various industries as a stepping stone toward full autonomy. The gradual integration of AI, leveraging human oversight, allows for a better understanding of failure modes while building confidence in AI systems. This iterative learning process, comparable to advancements in autonomous vehicles, highlights the complexities involved in deploying AI in real-world settings. As AI technologies evolve, the focus will be on refining models to better interact with human users and adapt to new situations seamlessly.
A major challenge in applied AI is out-of-distribution detection, or OOD, which is the task of detecting instances that do not belong to the distribution the classifier has been trained on. OOD data is often referred to as “unseen” data, as the model has not encountered it during training.
Bayan Bruss is the VP of AI Foundations at Capital One and in this role he works with academic researchers to translate the latest research to address fundamental problems in financial services. Bayan joins the show with Sean Falconer to talk about OOD, the importance of bringing AI research to real world applications, and more.
Full Disclosure: This episode is sponsored by Capital One
Sean’s been an academic, startup founder, and Googler. He has published works covering a wide range of topics from information visualization to quantum computing. Currently, Sean is Head of Marketing and Developer Relations at Skyflow and host of the podcast Partially Redacted, a podcast about privacy and security engineering. You can connect with Sean on Twitter @seanfalconer.