About

In December 2020, I graduated with a Masters in Data Science from Harvard’s Institute for Applied Computational Sciences. Before that, I graduated first in my class with a BS in Computer Science and Mathematics at the University of Edinburgh. I am passionate about solving real-world problems with machine learning and statistics.

Research

I conducted machine learning research at Harvard’s Data to Actionable Knowledge (DtAK) lab, focusing on uncertainty estimation for deep models. Specifically, I am interested in building Bayesian deep learning models which provide reliable and useful uncertainty estimates, an essential property for downstream applications in high-risk domains.

Recent Work

Théo Guénais, Dimitris Vamvourellis, Yaniv Yacoby, Finale Doshi-Velez, Weiwei Pan, BaCOUn: Bayesian Classifier with OOD Uncertainty, ICML Workshop on Uncertainty and Robustness in Deep Learning, 2020.

Dimitris Vamvourellis, Mate Attila Toth, Dhruv Desai, Dhagash Mehta, Stefano Pasquali, Learning Mutual Fund Categorization using Natural Language Processing, 3rd ACM International Conference on AI in Finance, 2022.

Work Experience

Currently, I am working as Quantitative Researcher at BlackRock’s Investment AI Group, developing machine learning and NLP solutions to aid investment processes. Previously, I worked as a data scientist for intelligencia.ai, building Natural Language Processing models to uncover trends and emerging areas of innovation in clinical research. Prior to that, I worked in the fintech industry for BlackRock, where I developed subject matter expertise on the quantitative capabilities of the Aladdin platform like stress testing, portfolio optimization and portfolio risk modelling tools.

Recent Projects

• Developed an LLM (application) for creating hypothetical debates between two (CAMEL) agents. The app can be used to rehearse a conversation you are interested in between two given roles (e.g. negotiate with your manager) and find potentially useful arguments and counterarguments on a given topic.

• Predicting flu rates across US states (github repo).
  • Built recurrent neural networks to forecast flu rates across US for multiple weeks into the future.
  • Used a hierarchical Bayesian model to analyze the impact of various socioeconomic and environmental factors on flu rates across different states.
• Predicting public opinion in the 2020 Democratic Primaries using Twitter #tweetpoll.
  • Developed recurrent neural networks to perform sentiment analysis on two million relevant tweets.
  • Created a Bayesian model to infer the popularity of each candidate based on the sentiment of tweets referring to this candidate.
• Mining Biomedical Text: Transfer Learning to the Rescue
  • Used BioBERT to mine relations between genes and diseases from massive medical literature, constructing a knowledge graph from unstructured text.

Contact

• email: dimvamvourellis@gmail.com
• linkedin