Daniel Packer Email & Phone Number

New York City Metropolitan Area

As a Senior Machine Learning Model Risk Specialist at Upstart, I:Leverage advanced expertise in Machine Learning and AI to oversee and mitigate ethical, financial, and regulatory risks across Upstart's ML model inventory.Collaborate with stakeholders, regulators, and research scientists to ensure comprehensive testing, documentation, and validation of all models.Conduct in-depth explainability analyses to elucidate model decision-making processes, enhancing transparency and trust.Rigorously examine and challenge core modeling assumptions to ensure continued robustness and accuracy.Develop and implement risk management strategies to maintain compliance with evolving regulatory standards in the AI/ML space.Lead cross-functional initiatives to integrate responsible AI practices throughout the model lifecycle.

Columbus, Ohio, United States

I work on the Model Risk Management team. I provide statistical and mathematical analyses and stress-tests of data-scientific models used throughout the company.This work involves coordinating with many people and teams across the company, understanding their methodologies, and clearly communicating my own analyses. I work flexibly across a broad range of programming languages and presentation formats to make sure I am using the right "hammers" for the right "nails".

Columbus, Ohio, United States

My research focuses on building machine learning models that implement priors for efficiency, robustness, and generalization. I have contributed directly to the literature around group invariant/equivariant machine learning (https://arxiv.org/abs/2205.14039). I provided both theorems about efficient implementations of these algorithms, as well as numerical results that confirm the algorithms' efficacy. I have additionally been working on constructing algorithms to approximate geometric/topological measurements such as the Gromov-Wasserstein distance, a powerful tool for understanding the global *and* local structure of a data set.I also have contributed to the mathematical formalization library mathlib, part of the lean project to formalize much of modern mathematics. Formal verification is an exciting field, which I believe can add significant value to the Machine Learning world. Another side research project of mine is the use of neural networks (and other machine learning tools) such as ResNet to solve PDEs.

Dayton, Ohio, United States

In this role I developed machine learning and data analytics methods in order to provide interpretable classifications of EKG readings. I worked closely on a small team to develop our entire platform, incorporating many of MinedXAI's in-house methods into our models (along with our own novel ideas, of course!).The central problem that we solved in this role was how to transform multi-dimensional time-series data into a format where the underlying topology was available for analysis. We explored a variety of strategies for achieving various forms of time-shift invariance/equivariance, including novel approaches in the ML literature.

Doctor Of Philosophy - Phd, Computational And Applied Mathematics

Bachelor'S Degree, Mathematics And Physics

Education record

Mathematics

Activities and Societies: Budapest Semesters in Mathematics