Currently, Anthony is a 5th year PhD student at Northeastern University. His research focuses on building robust, context-aware dialogue systems with a focus on the machine learning theory behind these systems. He has designed theoretical ML frameworks and NLP algorithms published across *ACL venues including work on parsing, dialogue management, language generation, and bias mitigation. His theoretical contributions span broad topics in ML theory like Domain Generalization, PAC-Bayesian Domain Adaptation, & Multitask Learning with his work on novel PAC-Bayesian theories for multiclass neural-networks winning a best paper award at UAI 2022. He also has experience in practical deployment of NLP systems, leading an Alexa Prize TaskBot team to 3rd place overall in this international contest.
Previously, Anthony has worked on research projects in computer vision, medical imaging, and sports analytics.
Contact: sicilia [DOT] a [AT] northeastern [DOT] edu | GitHub
Good News! 📰 🗞 Awards and More
- 12.09.2023: Best Area Chair Award for Dialogue Track at EMNLP 2023
- 11.01.2023: Tutorial on Domain Adaptation for Generative Models at AACL 2023
- 10.03.2023: Lead team to 3rd place in the international Alexa Prize TaskBot Challenge 2 contest
- 09.18.2023: Interning at AI2 with the Mosaic Team
- 2.6.2023: AISTATS ’22 Top Reviewer Award
- 8.1.2022: Best Paper Award (UAI, ’22)
Current Research
LEATHER: A Theoretical Framework for Language Generation
We design algorithms and evaluation protocols for context-aware text generation. Applications include bias mitigation, human-like text generation/evaluation; and human-explainable evaluation.
Domain Adaptation Statistics for Quantifying Data Shift
We design theoretically grounded statistics for quantifying impact of data shift on ML algorithms. Applications include predicting performance changes in discourse parsers and other classifiers; and estimating data-efficiency of deep learning algorithms.
Robust Algorithm Design
We use tools from ML theory (e.g., PAC learning bounds) to design robust learning algorithms. Applications include management of non-cooperative dialogue, classification on unseen data, and bias-mitigation in text generation.
Collaborative and Inclusive Dialogue
We design multi-modal, task-oriented dialogue systems which enable flexible and human-like collaboration. Users with different preferences and capabilities can tailor their experience, so our system is more accessible to diverse groups.
Publications
Combining Discourse Coherence with Large Language Models for
More Inclusive, Equitable, and Robust Task-Oriented Dialogue
Joint International Conference on Computational Linguistics, Language Resources and Evaluation, Torino, Italy, May, 2024.
Kate Atwell, Mert Inan, Anthony Sicilia, Malihe Alikhani
Learning to Generate Equitable Text in Dialogue from Biased Training Data
Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics. Toronto, Ontario, July, 2023.
Anthony Sicilia, Malihe Alikhani
PAC-bayesian domain adaptation bounds for multiclass learners
(Best-Paper Award) Conference on Uncertainty in Artificial Intelligence, Eindhoven, Netherlands, August, 2022.
Anthony Sicilia, Katherine Atwell, Malihe Alikhani, Seong Jae Hwang
HumBEL: A Human-in-the-Loop Approach for Evaluating Demographic Factors of Language Models in Human-Machine Conversations
18th Conference of the European Chapter of the Association for Computational Linguistics, Malta, March, 2024
Anthony Sicilia, Jennifer C. Gates, Malihe Alikhani
Deal, or no deal (or who knows)? Forecasting Uncertainty in Conversations using Large Language Models
arXiv PrePrint, 2024
Anthony Sicilia, Hyunwoo Kim, Khyathi Raghavi Chandu, Malihe Alikhani, Jack Hessel
LEATHER: A Framework for Learning to Generate Human-like Text in Dialogue
Findings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Processing, November, 2022
Anthony Sicilia, Malihe Alikhani
Modeling non-cooperative dialogue: Theoretical and empirical insights
Transactions of the Association for Computational Linguistics, September, 2022
Anthony Sicilia, Tristan Maidment, Pat Healy, Malihe Alikhani
The change that matters in discourse parsing: Estimating the impact of domain shift on parser error
Findings of Annual Meeting of the Association for Computational Linguistics, Dublin, Ireland, May, 2022
Katherine Atwell, Anthony Sicilia, Seong Jae Hwang, Malihe Alikhani
Test-time fourier style calibration for domain generalization
The International Joint Conference on Arti- ficial Intelligence and The European Conference on Artificial Intelligence, Vienna, Austria, July, 2022
Xingchen Zhao, Chang Liu, Anthony Sicilia, Seong Jae Hwang, Yun Fu
Robust white matter hyperintensity segmentation on unseen domain
IEEE International Symposium on Biomedical Imaging, April, 2021
Xingchen Zhao, Anthony Sicilia, Davneet S Minhas, Erin E O’Connor, Howard J Aizenstein, William E Klunk, Dana L Tudorascu, Seong Jae Hwang
Deephoops: Evaluating micro-actions in basketball using deep feature representations of spatio-temporal data
ACM SIGKDD, Anchorage, AK, USA, August, 2019
Anthony Sicilia, Konstantinos Pelechrinis, Kirk Goldsberry
Domain adversarial neural networks for domain generalization: When it works and how to improve
(to appear) Machine Learning, Springer, 2023
Anthony Sicilia, Xingchen Zhao, Seong Jae Hwang
PAC bayesian performance guarantees for deep (stochastic) networks in medical imaging
Medical Image Computing and Computer Assisted Intervention, Strasbourg, FR, October, 2021
Anthony Sicilia, Xingchen Zhao, Anastasia Sosnovskikh, Seong Jae Hwang
Multi-domain learning by meta-learning: Taking optimal steps in multi-domain loss landscapes by inner-loop learning
IEEE International Symposium on Biomedical Imaging, April, 2021
Anthony Sicilia, Xingchen Zhao, Davneet S Minhas, Erin E O’Connor, Howard J Aizenstein, William E Klunk, Dana L Tudorascu, Seong Jae Hwang
PittGrub: a frustration-free system to reduce food waste by notifying hungry college students
ACM SIGKDD, London, UK, August, 2018
Mark Silvis, Anthony Sicilia, Alexandros Labrinidis
Anthony Sicilia 