NF
Nicolas Flammarion
cs.LGstat.MLmath.OCcs.AIcs.CVcs.CLcs.DScs.CRmath.STstat.TH
On Valency
published · living versionsW_sqbkcgs6·v1 · currentpublished
Controlling Sharpening in LLM Post-Training
with Simon Matrenok, Nicolas Menet, Andreas Krause
1 version
Preprints & journals
77 papers in the corpus · 2016–2026Learning Distributions from Samples and Qualities2610.17956v1 · Marc Jourdan, Nicolas Flammarion2026 · 0 citationsarXiv
Two-Timescale Fine-tuning Provably Learns New Features for Two-Layer ReLU Networks2609.34667v1 · Etienne Boursier, Nicolas Flammarion2026 · 0 citationsarXiv
Statistical Benefits of Fine-Tuning from Pretrained Initialization in Diagonal Linear Networks2609.34756v1 · Alexandre Decleves, Etienne Boursier, Nicolas Flammarion2026 · 0 citationsarXiv
Induction Heads Interpolate N-Grams2607.02800v1 · Francesco D'Angelo, Oguz Kaan Yuksel, Swathi Shree Narashiman et al.2026 · 0 citationsarXiv
Space-Efficient Language Generation in the Limit2606.25777v1 · Nicolas Flammarion, Chirag Pabbaraju, Hristo Papazov et al.2026 · 0 citationsarXiv
Learning What to Forget: Improving LLM Unlearning via Learned Token-Level Importance2606.06320v1 · Gizem Yuce, Giorgos Nikolaou, Nicolas Flammarion2026 · 0 citationsarXiv
(How) Learning Rates Regulate Catastrophic Overtraining2604.13627v2 · Mark Rofin, Aditya Varre, Nicolas Flammarion2026 · 0 citationsarXiv
Transformers Learn Latent Mixture Models In-Context via Mirror Descent2604.10848v1 · Francesco D'Angelo, Nicolas Flammarion2026 · 0 citationsarXiv
Gradient Flow Polarizes Softmax Outputs towards Low-Entropy Solutions2603.06248v1 · Aditya Varre, Mark Rofin, Nicolas Flammarion2026 · 0 citationsarXiv
Incremental Learning of Sparse Attention Patterns in Transformers2602.19143v2 · Oguz Kaan Yuksel, Rodrigo Alvarez Lucendo, Nicolas Flammarion2026 · 0 citationsProceedings of the 43rd International Conference on Machine Learning (ICML 2026)
On the Adversarial Robustness of Discrete Image Tokenizers2602.18252v1 · Rishika Bhagwatkar, Irina Rish, Nicolas Flammarion et al.2026 · 0 citationsarXiv
On the Out-of-Distribution Generalization of Reasoning in Multimodal LLMs for Simple Visual Planning Tasks2602.15460v1 · Yannic Neuhaus, Nicolas Flammarion, Matthias Hein et al.2026 · 0 citationsarXiv
HalluHard: A Hard Multi-Turn Hallucination Benchmark2602.01031v1 · Dongyang Fan, Sebastien Delsad, Nicolas Flammarion et al.2026 · 0 citationsarXiv
Exact Learning of Arithmetic with Differentiable Agents2511.22751v1 · Hristo Papazov, Francesco D'Angelo, Nicolas Flammarion2025 · 0 citationsarXiv
Selective Induction Heads: How Transformers Select Causal Structures In Context2509.08184v1 · Francesco D'Angelo, Francesco Croce, Nicolas Flammarion2025 · 0 citationsarXiv
Learning In-context n-grams with Transformers: Sub-n-grams Are Near-stationary Points2508.12837v2 · Aditya Varre, Gizem Yuce, Nicolas Flammarion2025 · 0 citationsarXiv
Learning Algorithms in the Limit2506.15543v2 · Hristo Papazov, Nicolas Flammarion2025 · 0 citationsarXiv
OS-Harm: A Benchmark for Measuring Safety of Computer Use Agents2506.14866v2 · Thomas Kuntz, Agatha Duzan, Hao Zhao et al.2025 · 5 citationsarXiv
FuseLIP: Multimodal Embeddings via Early Fusion of Discrete Tokens2506.03096v2 · Christian Schlarmann, Francesco Croce, Nicolas Flammarion et al.2025 · 0 citationsarXiv
Chain-of-Frames: Advancing Video Understanding in Multimodal LLMs via Frame-Aware Reasoning2506.00318v2 · Sara Ghazanfari, Francesco Croce, Nicolas Flammarion et al.2025 · 0 citationsarXiv
Learning Parametric Distributions from Samples and Preferences2505.23557v1 · Marc Jourdan, Gizem Yuce, Nicolas Flammarion2025 · 0 citationsarXiv
Towards Unified Benchmark and Models for Multi-Modal Perceptual Metrics2412.10594v1 · Sara Ghazanfari, Siddharth Garg, Nicolas Flammarion et al.2024 · 0 citationsarXiv
Long-Context Linear System Identification2410.05690v2 · Oguz Kaan Yuksel, Mathieu Even, Nicolas Flammarion2024 · 0 citationsO. K. Y\"uksel, M. Even, N. Flammarion, "Long-Context Linear System Identification", in The Thirteenth International Conference on Learning Representations, 2025
Simplicity bias and optimization threshold in two-layer ReLU networks2410.02348v2 · Etienne Boursier, Nicolas Flammarion2024 · 0 citationsarXiv
Does Refusal Training in LLMs Generalize to the Past Tense?2407.11969v4 · Maksym Andriushchenko, Nicolas Flammarion2024 · 8 citationsarXiv
Implicit Bias of Mirror Flow on Separable Data2406.12763v3 · Scott Pesme, Radu-Alexandru Dragomir, Nicolas Flammarion2024 · 0 citationsarXiv
Is In-Context Learning Sufficient for Instruction Following in LLMs?2405.19874v3 · Hao Zhao, Maksym Andriushchenko, Francesco Croce et al.2024 · 4 citationsarXiv
Competition Report: Finding Universal Jailbreak Backdoors in Aligned LLMs2404.14461v2 · Javier Rando, Francesco Croce, Krystof Mitka et al.2024 · 1 citationarXiv
Jailbreaking Leading Safety-Aligned LLMs with Simple Adaptive Attacks2404.02151v4 · Maksym Andriushchenko, Francesco Croce, Nicolas Flammarion2024 · 14 citationsarXiv
JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models2404.01318v5 · Patrick Chao, Edoardo Debenedetti, Alexander Robey et al.2024 · 114 citationsarXiv
Leveraging Continuous Time to Understand Momentum When Training Diagonal Linear Networks2403.05293v1 · Hristo Papazov, Scott Pesme, Nicolas Flammarion2024 · 0 citationsarXiv
Long Is More for Alignment: A Simple but Tough-to-Beat Baseline for Instruction Fine-Tuning2402.04833v2 · Hao Zhao, Maksym Andriushchenko, Francesco Croce et al.2024 · 2 citationsarXiv
Early alignment in two-layer networks training is a two-edged sword2401.10791v3 · Etienne Boursier, Nicolas Flammarion2024 · 0 citationsarXiv
Why Do We Need Weight Decay in Modern Deep Learning?2310.04415v2 · Francesco D'Angelo, Maksym Andriushchenko, Aditya Varre et al.2023 · 19 citationsarXiv
Transferable Adversarial Robustness for Categorical Data via Universal Robust Embeddings2306.04064v2 · Klim Kireev, Maksym Andriushchenko, Carmela Troncoso et al.2023 · 3 citationsarXiv
Sharpness-Aware Minimization Leads to Low-Rank Features2305.16292v2 · Maksym Andriushchenko, Dara Bahri, Hossein Mobahi et al.2023 · 5 citationsarXiv
Saddle-to-Saddle Dynamics in Diagonal Linear Networks2304.00488v2 · Scott Pesme, Nicolas Flammarion2023 · 3 citationsarXiv
Quantum Channel Certification with Incoherent Strategies2303.01188v2 · Omar Fawzi, Nicolas Flammarion, Aur'elien Garivier et al.2023 · 1 citationProceedings of Thirty Sixth Conference on Learning Theory, PMLR 195:1822-1884, 2023
First-order ANIL provably learns representations despite overparametrization2303.01335v3 · Oguz Kaan Yuksel, Etienne Boursier, Nicolas Flammarion2023 · 1 citationO. K. Y\"uksel, E. Boursier and N. Flammarion, "First-order ANIL provably learns representations despite overparametrisation", in The Twelfth International Conference on Learning Representations, 2024
Penalising the biases in norm regularisation enforces sparsity2303.01353v4 · Etienne Boursier, Nicolas Flammarion2023 · 1 citationarXiv
Linearization Algorithms for Fully Composite Optimization2302.12808v2 · Maria-Luiza Vladarean, Nikita Doikov, Martin Jaggi et al.2023 · 0 citationsarXiv
(S)GD over Diagonal Linear Networks: Implicit Regularisation, Large Stepsizes and Edge of Stability2302.08982v2 · Mathieu Even, Scott Pesme, Suriya Gunasekar et al.2023 · 2 citationsarXiv
A Modern Look at the Relationship between Sharpness and Generalization2302.07011v2 · Maksym Andriushchenko, Francesco Croce, Maximilian Muller et al.2023 · 4 citationsarXiv
SGD with Large Step Sizes Learns Sparse Features2210.05337v2 · Maksym Andriushchenko, Aditya Varre, Loucas Pillaud-Vivien et al.2022 · 6 citationsarXiv
Label noise (stochastic) gradient descent implicitly solves the Lasso for quadratic parametrisation2206.09841v1 · Loucas Pillaud-Vivien, Julien Reygner, Nicolas Flammarion2022 · 3 citationsarXiv
Towards Understanding Sharpness-Aware Minimization2206.06232v1 · Maksym Andriushchenko, Nicolas Flammarion2022 · 29 citationsarXiv
Gradient flow dynamics of shallow ReLU networks for square loss and orthogonal inputs2206.00939v3 · Etienne Boursier, Loucas Pillaud-Vivien, Nicolas Flammarion2022 · 4 citationsarXiv
Sequential algorithms for testing identity and closeness of distributions2205.06069v1 · Omar Fawzi, Nicolas Flammarion, Aur'elien Garivier et al.2022 · 1 citationarXiv
Accelerated SGD for Non-Strongly-Convex Least Squares2203.01744v1 · Aditya Varre, Nicolas Flammarion2022 · 0 citationsarXiv
ARIA: Adversarially Robust Image Attribution for Content Provenance2202.12860v1 · Maksym Andriushchenko, Xiaoyang Rebecca Li, Geoffrey Oxholm et al.2022 · 5 citationsarXiv
Career total: 97 works. 77 are in this corpus.Showing the 50 most recent.
Profile built from the corpus for this byline.
Author records are still filling in while the Hub is in alpha. If this is your page, you'll be able to claim it soon. Spot a mistake? Tell us.