I am a researcher in Machine Learning. In February 2025, I received a PhD in Machine Learning from Télécom Paris university.

Consciousness is a mystery that lies at the very core of our experience. Yet, it has so far evaded our best efforts to define or measure it. Machine learning is a powerful tool to explore these elusive questions. It provides a way to formulate and quantitatively evaluate theories about the inner workings of our mind, the origin of creativity and freedom, the purposes that implicitly drive our thoughts, or the mechanisms that rule the brain – this intriguing interface between consciousness and the measurable material world.

During my PhD, I started to study these fascinating issues, and focused on language. Indeed, language shapes our mind: it articulates our thoughts and supports the crucial abilities of reasoning, planning and decision. A striking property of language is its discrete nature. It maps observations from our surrounding world, which is seemingly continuous, to a finite set of symbols. Our main contribution during this thesis was to leverage Multiple Choice Learning (MCL) to define this mapping, and study its mathematical properties.

The MCL framework is interesting in itself, and could be used in many areas of machine learning (unsupervised source separation, self-supervised learning, interactive data generation) to solve core issues such as task ambiguity or representation collapse. We are currently exploring these topics, which have various applications such as speech processing, brain signal processing, music editing, or environment tracking.

News

• I am currently looking for post-doctoral or research engineer positions in machine learning and statistics. Feel free to reach out if you wish to share an opportunity.

Selected publications

(2025) Multiple Choice Learning for Efficient Speech Separation with Many Speakers, D. Perera, F. Derrida, T. Mariotte, G. Richard and S. Essid. International Conference on Acoustics, Speech and Signal Processing (ICASSP).

(2024) Annealed Multiple Choice Learning: Overcoming limitations of Winner-takes-all with annealing, D. Perera, V. Letzelter, T. Mariotte, A. Cortés, M. Chen, S. Essid and G. Richard. Conference on Neural Information Processing Systems (NeurIPS).

(2024) Winner-takes-all learners are geometry-aware conditional density estimators, V. Letzelter, D. Perera, C. Rommel, M. Fontaine, S. Essid, G. Richard and P. Pérez. International Conference on Machine Learning (ICML).