Invited Speakers

• Paola Goatin (Inria Sophia-Antipolis, France)

• Edouard Grave (Kyutai, France)

• Huseyin Topaloglu (Cornell University, USA)

• Enrique Zuazua (FAU Erlangen, Germany)

Paola Goatin

Université Côte d’Azur, Inria, CNRS, LJAD, Sophia Antipolis, France, paola.goatin@inria.fr

Traffic Flow Models for Current and Future Mobility Challenges

The mobility paradigms could undergo a significant transformation in the near future, as new technologies enable extended data collection and Vehicle-to-Vehicle (V2V) or Vehicle-to-Infrastructure (V2I) communication. This will offer novel means to control and optimize traffic flow. Within this context, mathematical models play an important role, allowing for the design and evaluation of new management approaches. In this talk, I will present applications to road traffic regulation using Connected Automated Vehicles (CAVs) [1] or dynamic routing [2].
Our results are based on the development of specific macroscopic models accounting for the interacting dynamics of the different classes of users. Numerical experiments show that controlling a small fraction of users is, in general, sufficient to consistently improve the global system performance.

References
[1] C. Daini, M. L. Delle Monache, P. Goatin and A. Ferrara, Traffic control via fleets of
connected and automated vehicles, submitted. hal:04366870.
[2] A. Festa, P. Goatin and F. Vicini, Navigation system based routing strategies in traffic
flows on networks, J. Optim. Theory Appl., 198 (2023), 930-957.

Edouard Grave

Kyutai, France

Moshi: a real-time spoken dialogue model

Current systems for spoken dialogue rely on pipelines of independent components, namely voice activity detection, speech recognition, textual dialogue and text-to-speech. Such frameworks cannot emulate the experience of real conversations. First, their complexity induces a latency of several seconds between interactions. Second, text being the intermediate modality for dialogue, non-linguistic information that modifies meaning—such as emotion or non-speech sounds—is lost in the interaction. Finally, they rely on a segmentation into speaker turns, which does not take into account overlapping speech, interruptions and interjections.

Moshi solves these independent issues altogether by casting spoken dialogue as speech-to-speech generation. Starting from a text language model, Moshi generates speech as tokens from the quantizer of a neural audio codec, and separately models its own speech and that of the user into parallel streams. This allows for the removal of explicit speaker turns, and the modeling of arbitrary conversational dynamics. We extend the hierarchical semantic-to-acoustic token generation of previous work, by predicting time-aligned text tokens as a prefix to audio tokens. Our resulting model is the first real-time full-duplex spoken large language model, with a latency of around 200 ms in practice.

Huseyin Topaloglu

Incorporating Discrete Choice Models into Revenue Management Decisions

Over the last couple of decades, there has been enormous progress in using discrete choice models to understand how customers choose and substitute among products and incorporating this understanding into operational models to decide which assortment of products to offer to customers or what prices to charge. We owe some of this progress to increase in the computational power so that we can build and solve more detailed operational models, but perhaps, most of this progress is due to the fact that online sales channels started providing fine-grained data on how customers browse the products. In this talk, we will go over fundamental discrete choice models that have been used in building operational assortment optimization and pricing models, overview the main algorithmic approaches that have been developed to solve the operational models, and identify research prospects. The focus will be on both static models that make one-shot assortment optimization or pricing decisions, as well as dynamic models that explicitly capture the evolution of demand and inventories over time.

Enrique Zuazua

enrique.zuazua@fau.de
https://dcn.nat.fau.eu/enrique-zuazua/

[1] Friedrich Alexander Universität Erlangen Nürnberg - Alexander von Humboldt Professorship, Germany
[2] Fundación Deusto, Bilbao
[3] Universidad Autónoma de Madrid

Control and Machine Learning

Systems control, or cybernetics—a term first coined by Ampère and later popularized by Norbert Wiener—refers to the science of control and communication in animals and machines. The pursuit of this field dates back to antiquity, driven by the desire to create machines that autonomously perform human tasks, thereby enhancing freedom and efficiency.

The objectives of control systems closely parallel those of modern Artificial Intelligence (AI), illustrating both the profound unity within Mathematics and its extraordinary capacity to describe natural phenomena and drive technological innovation.

In this lecture, we will explore the connections between these mathematical disciplines and their broader implications. We will also discuss our recent work addressing two fundamental questions: Why does Machine Learning perform so effectively? And how can data-driven insights be integrated into the classical applied mathematics framework, particularly in the context of Partial Differential Equations (PDE) and numerical methods?