Shelvia Wongso

I am a Research Fellow at Nanyang Technological University working with Jeremie Houssineau on the theoretical foundations of uncertainty quantification for large language models (LLMs). I obtained my PhD from the National University of Singapore in 2025 under Mehul Motani. In my thesis, I investigated how information-theoretic measures can be used to analyze deep neural networks. My research areas are generalization and uncertainty quantification in deep neural networks. Currently, I am also looking towards specializing in AI safety, and in particular in the area of misalignment and hallucination of LLMs.

Contact: shelvia.wongso@ntu.edu.sg

News

May 1, 2025	My paper “Pointwise Information Measures as Confidence Estimators in Deep Neural Networks: A Comparative Study” is accepted at ICML 2025.
Nov 22, 2024	I passed my thesis defense. The title of my thesis is “Generalization and Trustworthiness in Deep Learning Through the Lens of Information-Theoretic Measures”.
May 7, 2024	My abstract is accepted at Recent Results Poster Session of the ISIT 2024.
Apr 14, 2024	My paper is accepted at ISIT 2024 Workshop on Information-Theoretic Methods for Trustworthy Machine Learning.

Latest Posts

Selected Publications

ICMLPoster

Pointwise Information Measures as Confidence Estimators in Deep Neural Networks: A Comparative Study

Shelvia Wongso, Rohan Ghosh, and Mehul Motani

In International Conference on Machine Learning, (ICML), 2025

Bib

@inproceedings{wongso2025pointwise,
  abbr2 = {Poster},
  author = {Wongso, Shelvia and Ghosh, Rohan and Motani, Mehul},
  title = {Pointwise Information Measures as Confidence Estimators in Deep Neural Networks: A Comparative Study},
  booktitle = {International Conference on Machine Learning, (ICML)},
  year = {2025},
}

ISITOral

Pointwise sliced mutual information for neural network explainability

Shelvia Wongso, Rohan Ghosh, and Mehul Motani

In IEEE International Symposium on Information Theory, (ISIT), 2023

Bib

@inproceedings{wongso2023pointwise,
  abbr2 = {Oral},
  author = {Wongso, Shelvia and Ghosh, Rohan and Motani, Mehul},
  title = {Pointwise sliced mutual information for neural network explainability},
  booktitle = {{IEEE} International Symposium on Information Theory, (ISIT)},
  year = {2023},
}

AISTATSOral

Using sliced mutual information to study memorization and generalization in deep neural networks

Shelvia Wongso, Rohan Ghosh, and Mehul Motani

In International Conference on Artificial Intelligence and Statistics (AISTATS), 2023

Oral Presentation at AISTATS (Top 1.9% of Submitted Papers).

Bib

@inproceedings{wongso2023using,
  abbr2 = {Oral},
  title = {Using sliced mutual information to study memorization and generalization in deep neural networks},
  author = {Wongso, Shelvia and Ghosh, Rohan and Motani, Mehul},
  booktitle = {International Conference on Artificial Intelligence and Statistics (AISTATS)},
  year = {2023},
  note = {Oral Presentation at AISTATS (Top 1.9% of Submitted Papers).},
}

ICLRPoster

Rethinking symbolic regression: Morphology and adaptability in the context of evolutionary algorithms

Fong Kei Sen, Shelvia Wongso, and Mehul Motani

In International Conference on Learning Representations (ICLR), 2022

Bib

@inproceedings{fong2022using,
  abbr2 = {Poster},
  title = {Rethinking symbolic regression: Morphology and adaptability in the context of evolutionary algorithms},
  author = {Sen, Fong Kei and Wongso, Shelvia and Motani, Mehul},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year = {2022},
}

ISITOral

Understanding deep neural networks using sliced mutual information

Shelvia Wongso, Rohan Ghosh, and Mehul Motani

In IEEE International Symposium on Information Theory, (ISIT), 2022

Bib

@inproceedings{wongso2022understanding,
  abbr2 = {Oral},
  author = {Wongso, Shelvia and Ghosh, Rohan and Motani, Mehul},
  title = {Understanding deep neural networks using sliced mutual information},
  booktitle = {{IEEE} International Symposium on Information Theory, (ISIT)},
  year = {2022},
}