|
Younes Ben Mazziane
About Me
I am a postdoctoral researcher at the University of Avignon under the supervision of Francesco De Pellegrini
and Eitan Altman.
I obtained my Ph.D. in Computer Science from Université Côte d’Azur and Inria Sophia Antipolis in 2024 under the supervision of
Sara Alouf and Giovanni Neglia.
My research focuses on the theoretical analysis and performance evaluation of networking algorithms.
The objective is to understand their behavior
and tune their parameters to the characteristics of the workload. To this end, I use mathematical tools such as Markov processes, stochastic approximation,
online optimization, and game theory.
More details are available in my CV.
News
- (Nov 17, 2025) Paper accepted in IEEE/ACM Transactions on Networking
- (June 12, 2025) Presented a paper at SIGMETRICS
Research highlights
Tractable analysis of a variant of the balls-and-bins problem with power of choice,
in which all minimum-load bins among the selected ones are incremented in the event of ties.
Asymptotic analysis of a multi-branch asymmetric random-walk Markov chain.
Efficient performance evaluation and parameter tuning of Count-Min Sketch
with Conservative Updates and Elastic Sketch, popular streaming algorithms, under different workload characteristics.
Streaming algorithms provide approximate statistics, such as item frequencies,
using limited memory in environments with very high data arrival rates.
No-regret guarantees for Follow-the-Perturbed-Leader in online linear optimization
over non-convex decision sets under partial observability. This result extends previous no-regret guarantees
for label-efficient forecasters with Hedge in the experts problem [IEEE/ACM TIT'05].
Caching algorithms that, for any request sequence, asymptotically perform at least as well as the optimal static policy,
even under partial observability, while requiring only O(1) expected amortized update time.
Partial observability is particularly relevant in femto-caching systems,
where a base station (BS) jointly manages the contents of multiple edge caches,
and full visibility of all requests at the BS would require continuous communication with the caches.
Convergence of no-regret learning and best-response dynamics in repeated
bandwidth-sharing/proportional allocation games to the Nash equilibrium of the stage game.
Supports proportional allocation auctions as a practical mechanism for fair and efficient
bandwidth sharing, for example in network slicing.
Mean-field-type approximations for adaptations of the classical Least Recently Used
policy to similarity caching under stationary stochastic request processes.
Efficient performance evaluation and parameter tuning of LRU adaptations to similarity caching;
a generalization of the classical caching problem by allowing requests to be served approximately by similar cached items.
Similarity caching is particularly relevant in applications such as recommendation systems.
|
