Aligning adversarial perturbations with the near-null singular directions of intermediate linear layers in transformer VLMs yields stronger attacks than existing feature- and output-space methods.
Robust Learning with Jacobian Regularization
8 Pith papers cite this work. Polarity classification is still indexing.
8
Pith papers citing it
abstract
Design of reliable systems must guarantee stability against input perturbations. In machine learning, such guarantee entails preventing overfitting and ensuring robustness of models against corruption of input data. In order to maximize stability, we analyze and develop a computationally efficient implementation of Jacobian regularization that increases classification margins of neural networks. The stabilizing effect of the Jacobian regularizer leads to significant improvements in robustness, as measured against both random and adversarial input perturbations, without severely degrading generalization properties on clean data.
years
2026 8representative citing papers
Polynomial representations yield an effective-degree simplicity metric that predicts generalization across tasks and serves as a differentiable regularizer improving performance in classification and RL.
Strain in the velocity Jacobian exponentially amplifies integration errors in flow matching while vorticity contributes linearly; strain-weighted Jacobian regularization reduces error up to 2.7x at NFE=5 and improves CIFAR-10 FID by 14% at NFE=10.
Upper generalization bounds for neural oscillators scale polynomially with MLP size and time length, avoiding the curse of parametric complexity, with numerical validation on a Bouc-Wen nonlinear system.
A simple black-box robustness measure is introduced that bounds MSE under input perturbations with high probability, supported by experiments on real datasets and new robustness curves.
DREG achieves highest overall and clean-regime accuracy among regularizers in a large factorial study, ranking second in noise robustness and performing best under GELU with advantages in low-data regimes.
ILLUME+ is a scalable explainability method that generates stable multi-form explanations for drug response predictions, recovering known biology and enabling new hypothesis generation.
citing papers explorer
-
On Adversarial Vulnerability of Vision-Language Models through the Lens of Intermediate Spectral Subspaces
Aligning adversarial perturbations with the near-null singular directions of intermediate linear layers in transformer VLMs yields stronger attacks than existing feature- and output-space methods.
-
Quantifying and Optimizing Simplicity via Polynomial Representations
Polynomial representations yield an effective-degree simplicity metric that predicts generalization across tasks and serves as a differentiable regularizer improving performance in classification and RL.
-
On the Role of Strain and Vorticity in Numerical Integration Error for Flow Matching
Strain in the velocity Jacobian exponentially amplifies integration errors in flow matching while vorticity contributes linearly; strain-weighted Jacobian regularization reduces error up to 2.7x at NFE=5 and improves CIFAR-10 FID by 14% at NFE=10.
-
Upper Generalization Bounds for Neural Oscillators
Upper generalization bounds for neural oscillators scale polynomially with MLP size and time length, avoiding the curse of parametric complexity, with numerical validation on a Bouc-Wen nonlinear system.
-
Robustness of neural networks to random noise perturbations of their inputs
A simple black-box robustness measure is introduced that bounds MSE under input perturbations with high probability, supported by experiments on real datasets and new robustness curves.
-
DREG: A Layer-Wise Jacobian Regularization as a General-Purpose Penalty
DREG achieves highest overall and clean-regime accuracy among regularizers in a large factorial study, ranking second in noise robustness and performing best under GELU with advantages in low-data regimes.
-
Explainable AI for Cancer Drug Response Prediction: Beyond Univariate Feature Attributions
ILLUME+ is a scalable explainability method that generates stable multi-form explanations for drug response predictions, recovering known biology and enabling new hypothesis generation.
- Supervised Learning Has a Geometric Blind Spot