Machine learning · Language · Applied AI

Umanga Bista उमङ्ग बिष्ट

I research and build machine learning systems, from understanding language to making AI useful in practice.

I'm an applied scientist based in Melbourne, Australia, originally from Lamjung, Nepal. My work brings together natural language processing, representation learning, retrieval, and the evaluation of language models.

About my work

I'm a Senior Applied Scientist at Oracle Health & AI, working on language models and AI systems for enterprise and clinical applications. My experience spans research, experimentation, and production engineering, including earlier work at ANU and LogPoint.

I’m particularly interested in model intelligence, efficient inference, autonomous experimentation, and building capable models under real-world system constraints.

I completed my PhD in Machine Learning at the Australian National University (2021), researching Comparative Summarization of Document Collections, advised by Lexing Xie, Aditya Menon, and Alexander Mathews. I also hold a bachelor's degree in Computer Engineering from the Institute of Engineering, Pulchowk, Nepal.

Selected research

My doctoral research explored how to select information that represents a collection of documents while highlighting what is new or different.

• SupMMD: A Sentence Importance Model for Extractive Summarization using Maximum Mean Discrepancy
  Umanga Bista, Alexander Mathews, Aditya Menon, Lexing Xie. Findings of EMNLP, 2020.
  Paper · Code

• Comparative Document Summarisation via Classification
  Umanga Bista, Alexander Mathews, Minjeong Shin, Aditya Krishna Menon, Lexing Xie. AAAI, 2019.
  Preprint · Code & dataset

• Visualizing Graph Differences from Social Media Streams
  Minjeong Shin, Dongwoo Kim, Jae Hee Lee, Umanga Bista, Lexing Xie. WSDM, 2019, demo.
  Publication

• Knowledge Guided Attention and Inference for Describing Images Containing Unseen Objects
  Aditya Mogadala, Umanga Bista, Lexing Xie, Achim Rettinger. ESWC, 2018.

All publications on Google Scholar ↗

Selected patents & applications

• Contextual Query Rewriting — co-inventor; US patent application, published 2025.
• Training data generation to facilitate fine-tuning embedding models — co-inventor; US Patent 12,423,530, granted 2025.

Teaching & service

At ANU, I tutored Document Analysis, Introduction to Machine Learning, and Programming for Data Scientists (2017–2020). I have reviewed for The Web Conference and the Asian Conference on Machine Learning.