Nicolas Flammarion

20

Papers

378

Total Citations

Papers (20)

Jailbreaking Leading Safety-Aligned LLMs with Simple Adaptive Attacks

Learning In-context $n$-grams with Transformers: Sub-$n$-grams Are Near-Stationary Points

Long Is More for Alignment: A Simple but Tough-to-Beat Baseline for Instruction Fine-Tuning

Square Attack: a query-efficient black-box adversarial attack via random search

Gen-Oja: Simple & Efficient Algorithm for Streaming Generalized Eigenvector Computation

Escaping from saddle points on Riemannian manifolds

Online Robust Regression via SGD on the l1 loss

Understanding and Improving Fast Adversarial Training

Sequential Algorithms for Testing Closeness of Distributions

Last iterate convergence of SGD for Least-Squares in the Interpolation regime.

Continuized Accelerations of Deterministic and Stochastic Gradient Descents, and of Gossip Algorithms

Implicit Bias of SGD for Diagonal Linear Networks: a Provable Benefit of Stochasticity

Gradient flow dynamics of shallow ReLU networks for square loss and orthogonal inputs

Saddle-to-Saddle Dynamics in Diagonal Linear Networks

Transferable Adversarial Robustness for Categorical Data via Universal Robust Embeddings

(S)GD over Diagonal Linear Networks: Implicit bias, Large Stepsizes and Edge of Stability

Sharpness-Aware Minimization Leads to Low-Rank Features

Penalising the biases in norm regularisation enforces sparsity

On the spectral bias of two-layer linear networks

On the Theory of Variance Reduction for Stochastic Gradient Monte Carlo