Martin Jaggi
Associate Professor
Leads the Machine Learning and Optimization Laboratory, working on distributed and collaborative training, optimization methods and open medical language models.
Martin Jaggi's lab at EPFL works on learning theory & optimization, signal processing & information theory and distributed, cloud & edge systems. 46 papers since 2023, 37 at top venues, mostly NeurIPS and ICLR. 12 PhD students. 13 alumni, most often to Google and Meta. Applications close .
At a glance
- papers, 2023–2026
- 12/ year
- at T1 venues
- 80%
- PhD students
- 12
- MS students
- 0
- alumni listed
- 13
Papers by yearfrom DBLP, 2023–2026
- 7 papers2026 · T1 4 · collecting
- 11 papers2025 · T1 9
- 15 papers2024 · T1 13
- 13 papers2023 · T1 11
T1 venuesOtherVenues: NeurIPS, ICLR, ICML, ACL, npj Digit. Medicine
Research focuspaper titles that use each topic's words
- Learning theory & optimization19 papers
- optimization5 papers
- Signal processing & information theory3 papers
- Distributed, cloud & edge systems1 paper
- optimization methods0 papers
Paper list
From DBLP.
20267 papers
- Trans. Mach. Learn. Res.A New First-Order Meta-Learning Algorithm with Convergence Guarantees (opens in a new tab)
- ACL (T1)Apertus: Democratizing Open and Compliant LLMs for Global Language Environments (opens in a new tab)
- ICLR (T1)Beyond URLs: Metadata Diversity and Position for Efficient LLM Pretraining (opens in a new tab)
- ICLR (T1)Gradient-Normalized Smoothness for Optimization with Approximate Hessians (opens in a new tab)
- Trans. Mach. Learn. Res.Leveraging the True Depth of LLMs (opens in a new tab)
- Trans. Mach. Learn. Res.Mitigating Unintended Memorization with LoRA in Federated Learning for LLMs (opens in a new tab)
- ICLR (T1)Weight Decay may matter more than µP for Learning Rate Transfer in Practice (opens in a new tab)
202511 papers
- ICLR (T1)Attention with Markov: A Curious Case of Single-layer Transformers (opens in a new tab)
- ICLR (T1)CoTFormer: A Chain of Thought Driven Architecture with Budget-Adaptive Computation Cost at Inference (opens in a new tab)
- ICLR (T1)Effective Interplay between Sparsity and Quantization: From Theory to Practice (opens in a new tab)
- SwissTextEnhancing Multilingual LLM Pretraining with Model-Based Data Selection (opens in a new tab)
- NeurIPS (T1)GRAPE: Optimize Data Mixture for Group Robust Multi-target Adaptive Pretraining (opens in a new tab)
- AISTATS (T1)Improving Stochastic Cubic Newton with Momentum (opens in a new tab)
- ICLR (T1)Intrinsic User-Centric Interpretability through Global Mixture of Experts (opens in a new tab)
- ICML (T1)On-Device Collaborative Language Modeling via a Mixture of Generalists and Specialists (opens in a new tab)
- NeurIPS (T1)Towards Fully FP8 GEMM LLM Training at Scale (opens in a new tab)
- Trans. Mach. Learn. Res.Training Dynamics of the Cooldown Stage in Warmup-Stable-Decay Learning Rate Scheduler (opens in a new tab)
- NeurIPS (T1)URLs Help, Topics Guide: Understanding Metadata Utility in LLM Training (opens in a new tab)
202415 papers
- NeurIPS (T1)Analyzing & Reducing the Need for Learning Rate Warmup in GPT Training (opens in a new tab)
- NeurIPS (T1)CoBo: Collaborative Learning via Bilevel Optimization (opens in a new tab)
- NeurIPS (T1)DenseFormer: Enhancing Information Flow in Transformers via Depth Weighted Averaging (opens in a new tab)
- ICML (T1)DOGE: Domain Reweighting with Generalization Estimation (opens in a new tab)
- AAAI (T1)Ghost Noise for Regularizing Deep Neural Networks (opens in a new tab)
- ICML (T1)LASER: Linear Compression in Wireless Distributed Optimization (opens in a new tab)
- ICLR (T1)Layer-wise linear mode connectivity (opens in a new tab)
- npj Digit. MedicineMyThisYourThat for interpretable identification of systematic bias in federated learning for biomedical images (opens in a new tab)
- ICML (T1)On Convergence of Incremental Gradient for Non-convex Smooth Functions (opens in a new tab)
- NeurIPS (T1)QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMs (opens in a new tab)
- ICML (T1)Rotational Equilibrium: How Weight Decay Balances Learning Across Neural Networks (opens in a new tab)
- NeurIPS (T1)Scaling Laws and Compute-Optimal Training Beyond Fixed Training Durations (opens in a new tab)
202313 papers
- ICLR (T1)Agree to Disagree: Diversity through Disagreement for Better Transferability (opens in a new tab)
- JMLR (T1)Beyond Spectral Gap: The Role of the Topology in Decentralized Learning (opens in a new tab)
- NeurIPS (T1)Collaborative Learning via Prediction Consensus (opens in a new tab)
- npj Digit. MedicineDeepBreath - automated detection of respiratory pathology from lung auscultation in 572 pediatric outpatients across 5 countries (opens in a new tab)
- NeurIPS (T1)Fast Attention Over Long Sequences With Dynamic Sparse Flash Attention (opens in a new tab)
- COLT (T1)Linearization Algorithms for Fully Composite Optimization (opens in a new tab)
- NeurIPS (T1)MultiMoDN - Multimodal, Multi-Task, Interpretable Modular Networks (opens in a new tab)
- NeurIPS (T1)Multiplication-Free Transformer Training via Piecewise Affine Operations (opens in a new tab)
- Trans. Mach. Learn. Res.Provably Personalized and Robust Federated Learning (opens in a new tab)
- NeurIPS (T1)Random-Access Infinite Context Length for Transformers (opens in a new tab)
- ICML (T1)Second-Order Optimization with Lazy Hessians (opens in a new tab)
- ACL (T1)SIMSUM: Document-level Text Simplification via Simultaneous Summarization (opens in a new tab)
Where alumni went13 listed, 50% to industry
Industry 4Academia 4
Most often to
- Meta
- Microsoft
Also to
- Shanghai University of Finance and Economics
- University of Zurich. Her new group is hiring!
- USC
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
- CNZhejiang
Jiawei Chen
- Recommendation, search & web
- GBImperial
Francesco Leofante
- Trustworthy & safe ML
- CNPeking
Hong Mei
- Graph learning
- USStanford
Christopher Ré
- Generative & foundation models
- USUIUC
Kate Donahue
- Trustworthy & safe ML
- USCMU
Ameet Talwalkar
- Human-AI collaboration & trust
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- DBLP
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
EPFLcloses