Roundup: Our Research to Advance Machine Learning in Finance

At Capital One, we believe the power of technology can give customers greater protection, confidence, and control of their finances. Among the most impactful ways for us to achieve these goals are the responsible, human-centered use of real-time data and machine learning.

Our machine learning research program centers on exploring and elevating forward-leaning machine learning—the methods, applications, and techniques that will make banking simpler and safer. We’re advancing this research to inform how machine learning is developed and implemented across the banking industry in the years to come. Our research findings are also incorporated into the fabric of our machine learning ecosystem, helping our models become more powerful, adaptive, and well-managed.

Our research agenda explores areas critical to our business as well as machine learning theory—often alongside some of the nation’s top research universities. We’re open sourcing tools to make machine learning models more well-managed, repeatable, and searchable. We’re working to understand how a range of deep learning techniques can become more explainable and interpretable. We’re exploring novel applications of graph embeddings to uncover financial crime. And we’re leveraging neural networks to protect sensitive data.

Capital One Machine Learning Research Priorities:

• Explainable AI: Creating transparency and ensuring fairness through explaining ML models.
• Graph networks: Using ML on nodes and edges of financial networks to more accurately identify fraud.
• Anomaly Detection: Identifying changes in data to protect customers and adapt to fluctuating environments.
• Natural Language Processing: Teaching intelligent assistants to understand and generate natural language.
• Privacy and Data: Developing models and techniques for protecting sensitive customer data.
• Machine Learning at Scale: Building systems to scale the entire model building and deployment process.

Check out some of our most recent machine learning research publications and keep an eye out for more updates in 2022. We’ll be further exploring synthetic data generation and federated learning at scale to enhance privacy efforts; tabular data solutions to bolster our machine learning capabilities in areas like fraud detection; explainability methods including topical data analysis and model introspection; sequence modeling for credit risk, and much more.

2021

• Dynamic Customer Embeddings for Financial Service Applications, presented at the ICML Workshop on Representation Learning in Finance
• SAINT: Improved Neural Networks for Tabular Data via Row Attention and Contrastive Pre-Training
• Latent-CF: A Simple Baseline for Reverse Counterfactual Explanations, presented at the NeurIPs workshop on Fair AI in Finance
• MetaBalance: High-Performance Neural Networks for Class-Imbalanced Data
• Unsupervised Anomaly Detection with Adversarial Mirrored AutoEncoders
• Counterfactual Explanations via Latent Space Projection and Interpolation

2020

• Towards Ground Truth Explainability on Tabular Data, presented at the ICML GRL+ Workshop
• Navigating the Dynamics of Financial Embeddings over Time, presented at the ICML GRL+ Workshop
• Quantifying Challenges in the Application of Graph Representation Learning, published in IEEE
• Machine Learning for Temporal Data in Finance: Challenges and Opportunities, presented at KDD MLF
• Sensitive Data Detection with High-Throughput Neural Network Models for Financial Institutions, presented at the AAAI Workshop on Knowledge Discovery in Finance
• Melody: Generating and Visualizing Machine Learning Model Summary to Understand Data and Classifiers Together
• SUBPLEX: Towards a Better Understanding of Black Box Model Explanations at the Subpopulation Level
• Effects of Model Misspecification on Bayesian Bandits: Case Studies in UX Optimization

2019

• Graph Embeddings at Scale, presented at KDD MLG
• DeepTrax: Embedding Graphs of Financial Transactions, published in IEEE
• Global Explanations of Neural Networks: Mapping the Landscape of Predictions
• On the Interpretability and Evaluation of Graph Representation Learning, presented at the NeurIPs GRL Workshop

2018

• Towards Explainable Deep Learning for Credit Lending: A Case Study
• Mixed Membership Recurrent Neural Networks

If you’re interested in helping us research and build the technology that will drive the future of banking, learn more about our machine learning efforts and take a look at our open tech career opportunities.