5 papers
Implicit Bias and Convergence of Matrix Stochastic Mirror Descent
Danil Akhtiamov, Reza Ghane, Omead Pooladzandi +1
We investigate Stochastic Mirror Descent (SMD) with matrix parameters and vector-valued predictions, a framework relevant to multi-class classification and matrix completion proble…
PUREVQ-GAN: Defending Data Poisoning Attacks through Vector-Quantized Bottlenecks
Alexander Branch, Omead Pooladzandi, Radin Khosraviani +3
We introduce PureVQ-GAN, a defense against data poisoning that forces backdoor triggers through a discrete bottleneck using Vector-Quantized VAE with GAN discriminator. By quantizi…
PureEBM: Universal Poison Purification via Mid-Run Dynamics of Energy-Based Models
Omead Pooladzandi, Jeffrey Jiang, Sunay Bhat +1
Data poisoning attacks pose a significant threat to the integrity of machine learning models by leading to misclassification of target distribution data by injecting adversarial ex…
PureGen: Universal Data Purification for Train-Time Poison Defense via Generative Model Dynamics
Sunay Bhat, Jeffrey Jiang, Omead Pooladzandi +2
Train-time data poisoning attacks threaten machine learning models by introducing adversarial examples during training, leading to misclassification. Current defense methods often…
Curvature-Informed SGD via General Purpose Lie-Group Preconditioners
Omead Pooladzandi, Xi-Lin Li
We present a novel approach to accelerate stochastic gradient descent (SGD) by utilizing curvature information obtained from Hessian-vector products or finite differences of parame…