Nicholas Klein Email & Phone Number

Los Angeles, Ca, Us

Atlanta, Georgia, United States

Los Angeles, California, United States

* Devised and tuned several neural models (an MLP, an LSTM, and a Transformer-encoder model) to predict hourly change in emissions from 3 years of time-series data with a maximum R^2 value of .88.* Restricted the latent space to be interpretable via bottlenecking and regularization.

Los Angeles, California, United States

* Generated explainable representations for 26 million Wikidata entities that improved the accuracy of a neural net trained to perform entity linking in tables from .67 to .86, outperforming all baseline graph embeddings while uniquely explaining their behavior.* Constructed three random-walk embedding models for Wikidata using a Skip-gram model, each focusing on a different sense of similarity: homophily, numeric attribute, and structural. On the above entity linking task, these gave accuracies of .86, .71, and .67.

Madison, Wisconsin Area

* Led the first major effort to provide support for cross time zone software usage by updating a piece of temporal clinical data to consider the time zone in which it is captured, fixing issues with searches that filter by date/time.* Conducted performance research to design a novel indexing strategy for this data to ensure scalability.* Wrote a library of APIs for accessing this data in a maintainable manner. They are now used across my team and related teams.

Chicago, Il

* Composed a distributed system of Google Cloud instances for data-processing jobs. The system is fault tolerant, enabling the use of cheaper but less reliable instances.* Selected and tuned an automated approach for classifying the usability of ECG data. Addressed imbalanced data by class-weighting and using a balanced performance metric (MCC).

Ridgefield, Connecticut

Master Of Science - Ms, Computer Science (Artificial Intelligence), 4.0/4.0

Bachelor'S Degree, Computer Science