Institution profile

EPFL

Academic institutioneurope · ch
Official website
Research library1,600linked papers
Opportunities0open roles
Selected work

Representative Papers

Chance-Constrained Trajectory Planning With Multimodal Environmental Uncertainty

Mar 09, 2025IEEE Control Systems Letters

This work addresses the safety-critical trajectory planning problem for autonomous driving under multimodal uncertainty in obstacle behavior. Methodologically, it proposes a novel chance-constrained optimization framework based on Gaussian Mixture Models (GMMs), wherein GMMs are explicitly embedded into chance constraints for the first time. Tight concentration bounds are derived via finite-sample statistical inference to guarantee confidence levels, and Conditional Value-at-Risk (CVaR) is innovatively adopted as a risk-averse surrogate to quantify and control constraint violation risk. The resulting formulation is cast as a tractable Mixed-Integer Conic Program (MICO). Extensive experiments on standard trajectory prediction benchmarks and real-world autonomous driving datasets demonstrate that the method significantly improves trajectory safety and computational feasibility in complex uncertain environments, while maintaining theoretical rigor and engineering practicality.

21 citationsRead paper

Reviewriter: AI-Generated Instructions For Peer Review Writing

Jun 04, 2025Workshop on Innovative Use of NLP for Building Educational Applications

This study addresses the limited capacity of learners to produce effective peer reviews in German writing instruction. To this end, we developed the first AI-driven peer review guidance tool specifically designed for educational writing tasks. Methodologically, we propose an adaptive AI instruction generation paradigm: German-GPT2 is fine-tuned on authentic student peer review corpora and integrated with instruction-based prompt engineering and a human-AI collaborative evaluation mechanism. Validation employs dual-track assessment—quantitative metrics (BLEU/ROUGE) and expert human evaluation. Empirical results from a pilot with 14 German learners demonstrate statistically significant improvements in review normativity and reflective depth, alongside high technology acceptance. Our core contributions are twofold: (1) the first education-oriented adaptive instruction generation framework for peer review support, and (2) empirical validation of generative AI’s feasibility and pedagogical efficacy in authentic language teaching contexts.

14 citationsRead paper

An Anytime Algorithm for Good Arm Identification

Oct 16, 2023arXiv.org

This paper studies the Good-Arm Identification (GAI) problem in stochastic multi-armed bandits—identifying arms whose expected rewards exceed a given threshold at any time. To address the lack of unified solutions across fixed-budget and anytime recommendation settings, we propose APGAI, the first parameter-free, truly anytime adaptive sampling algorithm for GAI. Its key contributions are threefold: (i) it is the first GAI algorithm achieving both anytime validity and parameter independence; (ii) we theoretically prove that its adaptive sampling strategy significantly outperforms uniform sampling in detecting the “no-good-arm” scenario; and (iii) we derive tight anytime upper bounds on both error probability and sampling complexity. APGAI integrates confidence-interval estimation, an anytime stopping rule, stochastic bandit analysis under random time horizons, and an empirical Bayes–inspired decision heuristic. Extensive experiments on synthetic and real-world datasets validate its efficiency, robustness, and theoretical guarantees.

4 citations1 influentialRead paper

On the Inherent Anonymity of Gossiping

Aug 04, 2023International Symposium on Distributed Computing

This work addresses the lack of a general quantitative framework for source anonymity in gossip protocols. We establish, for the first time, tight fundamental limits on source anonymity for arbitrary graphs under ε-differential privacy. To achieve this, we introduce a novel reduction paradigm—“random walk with probabilistic extinction”—which enables rigorous differential privacy analysis of classic gossip protocols such as Cobra Walk. Our theoretical analysis precisely characterizes the intrinsic trade-off between dissemination latency and source anonymity. We prove that graphs with low connectivity cannot provide meaningful source anonymity, whereas Cobra Walk achieves nontrivial ε-differential privacy on graphs with controllable connectivity. The results are tight, scalable, and have been successfully applied to analyze real-world anonymity protocols—including Dandelion—providing foundational theoretical support for the design and evaluation of anonymous communication systems.

4 citations1 influentialRead paper

Selective Randomization Inference for Adaptive Experiments

May 11, 2024

In adaptive experiments, data-driven design adjustments invalidate conventional statistical inference; existing methods suffer from narrow applicability and strong assumptions. This paper proposes a selective randomization inference framework: it models the data-generating process via a directed acyclic graph (DAG) and implements conditional-on-selection inference within randomization tests. It is the first systematic integration of this principle into randomization-based inference—requiring neither i.i.d. nor parametric modeling assumptions, and accommodating arbitrary adaptive experimental designs. To address disconnected confidence intervals, we innovatively introduce a holdout-unit method. Theoretically and empirically, our approach strictly controls selective Type-I error and constructs valid confidence intervals for homogeneous treatment effects. It substantially outperforms conventional methods in both robustness and generality.

4 citationsRead paper
Recent publications

Latest Papers