Antonio Barbalau

Machine Learning Researcher @ Bitdefender

Scientific Committee Member @ AI Olympiad, Romania

Member @ European Laboratory for Learning and Intelligent Systems (ELLIS)

Google Scholar Profile

Academic Experience

Computer Science PhD

University of Bucharest2019 - 2024

Thesis on “Novel Approaches for Weakly Supervised and Self-Supervised Learning” featuring an oral presentation at NeurIPS (1% acceptance rate), a CVPR, a CVIU, an ECML-PKDD, an EMNLP and a NeurIPS datasets track paper, reaching 500 citations before the final defense. Additionally, I have received the Best Romanian AI PhD Thesis Award in 2025.

Scientific Committee Member

AI Olympiad, Romania2024 - Present

Engaging in the training, evaluation and selection of the IOAI and IAIO Romanian teams since the very first year these olympiads took shape.

Teaching Assistant

University of Bucharest2017 - 2022

Most notably: Machine Learning and Deep Learning for the AI Master and Artificial Intelligence for the undergraduate level.

Research Assistant

University of Bucharest2020 - 2022

Involved in research projects covering various topics including authorship identification and video anomaly detection. Those projects ran in partnership with various companies and universities including Bitdefender, Tremend, BRD and the University of Central Florida.

Short-Term Scholar

University of Central FloridaMay 2021 - September 2021

Involved in research at the UCF Center for Research on Computer Vision on multi-object tracking, mainly focusing on self-supervised learning for multi-object tracking.

Reviewer

ICML(2026), ICLR(2025), NeurIPS(2025), ECAI(2025), EEML(2025), ICPR(2022, 2020).

Industry Experience

Senior Machine Learning Researcher

Bitdefender2024 - Present

Developing novel approaches for domain generalization and bias detection and prevention. Currently focusing on improving the generalization capabilities of fine-tuned foundation models.

Head of AI

Vatis Tech2023 - 2024

Research and development for complete Automatic Speech Recognition systems: multi-lingual transcription, word-level timestamps, speaker diarization, real-time inference (stream processing). Most notably: developed a transcription system capable of transcribing audios featuring multiple languages and real-time language switches, proposed and developed a novel end-to-end speaker diarization deep learning model capable of real-time performance.

Machine Learning Research and Development

Sparktech Software2023 - 2024

Worked on an Automated Vouchering Tool project aimed at extracting data from invoices and invoice-like documents in order to be compared to General Ledgers. My contribution included research and development on machine sub-projects such as: signature detection and matching, logo detection and matching, page deskewing and handwritten number extraction. I have also coordinated the research and development for an Explainable AI platform hosted on the cloud for which I have proposed an exemplar synthetization framework.