Volkan Cevher
Full Professor
Leads the Laboratory for Information and Inference Systems, working on optimization theory, reinforcement learning and robust deep learning.
Volkan Cevher's lab at EPFL works on learning theory & optimization, reinforcement learning and efficient training & inference. 87 papers since 2023, 71 at top venues, mostly ICML and ICLR. 7 PhD students. 25 alumni, most often to Google DeepMind. Applications close .
At a glance
- papers, 2023–2026
- 23/ year
- at T1 venues
- 82%
- PhD students
- 7
- MS students
- 0
- alumni listed
- 25
Papers by yearfrom the lab homepage, 2023–2026
- 11 papers2026 · T1 5 · collecting
- 32 papers2025 · T1 25
- 21 papers2024 · T1 21
- 23 papers2023 · T1 20
T1 venuesOtherVenues: ICML, ICLR, NeurIPS, TMLR, JMLR
Research focuspaper titles that use each topic's words
- Learning theory & optimization25 papers
- Efficient training & inference23 papers
- Reinforcement learning21 papers
- robust deep0 papers
- micro engineering0 papers
Paper list
From the lab homepage.
202611 papers
- IEEE Signal Processing MagazineAligning Large Language Models With Human Feedback: Mathematical foundations and algorithm design [Special Issue on the Mathematics of Deep Learning] (opens in a new tab)
- International Conference on Web and Internet EconomicsConvergence to Equilibrium of No-Regret Dynamics in Congestion Games (opens in a new tab)
- TMLR (T1)Diffusion-based Cumulative Adversarial Purification for Vision Language Models
- TMLR (T1)ESLM: Risk-Averse Selective Language Modeling with Hierarchical Batch Selection
- ALIFE Artificial Life ConferenceEvolving Many Worlds: Towards Open-Ended Discovery in Petri Dish NCA via Population-Based Training (opens in a new tab)
- IEEE Journal of Solid-State CircuitsLogSync: A Low-Power Logarithmic SoC for Cross-Regional Synchrony Extraction and Adaptive Psychiatric State Classification (opens in a new tab)
- ICML (T1)Multi-agent imitation learning with function approximation: Linear Markov games and beyond
- ICML (T1)On the Role of Batch Size in Stochastic Conditional Gradient Methods
- IEEE Signal Processing MagazineSequence Modeling Architectures: Foundations [Special Issue on the Mathematics of Deep Learning] (opens in a new tab)
- ICML (T1)Spatial Priors via Space Filling Curves for Small and Limited Data Vision Transformers
- IEEE Signal Processing MagazineTraining Neural Networks at Any Scale: An exposition (opens in a new tab)
202532 papers
- ICML (T1)A Simple Model of Inference Scaling Laws
- ICML (T1)Accelerating Spectral Clustering under Fairness Constraints
- ACM / IMS Journal of Data ScienceAdaptive Bilevel Optimization (opens in a new tab)
- ICLR (T1)Addressing Label Shift In Distributed Learning Via Entropy Regularization
- ICLR (T1)Adversarial Training For Defense Against Label Poisoning Attacks
- NeurIPS (T1)Ascent Fails to Forget
- ICML (T1)Best of Both Worlds: Regret Minimization versus Minimax Play
- ICLR (T1)Certified Robustness Under Bounded Levenshtein Distance
- ICML (T1)CHAMELEON: A Flexible Data-mixing Framework for Language Model Pretraining and Finetuning
- ICML (T1)Continuous-Time Analysis of Heavy Ball Momentum in Min-Max Games
- Small MethodsDeep‐Learning‐Assisted SICM for Enhanced Real‐Time Imaging of Nanoscale Biological Dynamics (opens in a new tab)
- 26 Interspeech ConferenceDeepFilterGAN: A Full-band Real-time Speech Enhancement System with GAN-based Stochastic Regeneration (opens in a new tab)
202421 papers
- ICLR (T1)Efficient Continual Finite-Sum Minimization
- ICLR (T1)Efficient local linearity regularization to overcome catastrophic overfitting
- ICLR (T1)Generalization of Scaled Deep ResNets in the Mean-Field Regime
- ICML (T1)Going beyond Compositions, DDPMs Can Produce Zero-Shot Interpolations
- ICLR (T1)High-Dimensional Kernel Methods under Covariate Shift: Data-Dependent Implicit Regularization
- ICML (T1)Imitation Learning in Discounted Linear MDPs without exploration assumptions
- ICML (T1)Improving SAM Requires Rethinking its Optimization Formulation
- ICML (T1)Learning to Remove Cuts in Integer Linear Programming
- JMLR (T1)Learning with Norm Constrained, Over-parameterized, Two-layer Neural Networks
- NeurIPS (T1)Membership Inference Attacks against Large Vision-Language Models
- TMLR (T1)Mixed Nash for Robust Federated Learning
- ICLR (T1)Multilinear Operator Networks
202323 papers
- Mathematical ProgrammingA unified stochastic approximation framework for learning in games (opens in a new tab)
- —Adversarial Training Should Be Cast As a Non-Zero-Sum Game
- NeurIPS (T1)Alternation makes the adversary weaker in two-player games
- ICML (T1)Benign Overfitting in Deep Neural Networks under Lazy Training
- ICLR (T1)DiGress: Discrete Denoising diffusion for graph generation
- ICLR (T1)Distributed Extra-Gradient With Optimal Complexity And Communication Guarantees
- NeurIPS (T1)Efficient Online Clustering with Moving Costs
- Pattern RecognitionEnd-to-end kernel learning via generative random Fourier features (opens in a new tab)
- TMLR (T1)Federated Learning under Covariate Shifts with Generalization Guarantees
- ICLR (T1)Finding Actual Descent Directions For Adversarial Training
- NeurIPS (T1)Initialization Matters: Privacy-Utility Analysis of Overparameterized Neural Networks
- TPAMI (T1)Linear Complexity Self-Attention With 3rd Order Polynomials (opens in a new tab)
Where alumni went25 listed, 33% to industry
Industry 1Academia 2
Most often to
- Google DeepMind
Also to
- ETH Zurich
Apply to EPFLFall 2027
EDIC Computer and Communication Sciences PhD
- Round
- Round 1
- Opens
- —
- GRE
- optional
- English
- TOEFL 100 / IELTS 7.0
- Fee
- none
Confirmed on the program page.
Similar labsnearest by research summary
- GBImperial
Alessandra Russo
- Reasoning & neuro-symbolic AI
- CNPeking
Liwei Wang
- Learning theory & optimization
- GBCambridge
Richard E. Turner
- Probabilistic & causal inference
- USUIUC
Nan Jiang
- Learning theory & optimization
- USMichigan
Walter Dempsey
- Learning theory & optimization
- USMIT
Joshua Tenenbaum
- Computational neuroscience & cognition
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
EPFLcloses