Dynamic quantization creates side channels allowing partial or full recovery of other users' batched data in at least four popular ML frameworks.
Canonical reference
Title resolution pending
Canonical reference. 100% of citing Pith papers cite this work as background.
citation-role summary
citation-polarity summary
years
2026 6roles
background 2polarities
background 2representative citing papers
LAVA is a layered audio-visual watermarking system using cross-modal fusion and calibration-aware alignment to achieve robust deepfake tamper detection and localization under compression and asynchrony.
SIPS decomposes stochastic interpolant dynamics into predictive drift and generative denoising to combine arbitrary pretrained predictors with a degradation-agnostic clean-speech prior for better speech enhancement and separation.
The Bayesian Neural Kalman Filter uses a trained BNN to predict UAV states and uncertainties, then applies a Kalman update to outperform standard EKF and UKF on synthetic data under high noise and low sampling rates.
WhisperPipe delivers 89 ms median latency and 48% lower peak GPU memory than standard Whisper while keeping word error rate within 2% of the offline model.
Multi-task autoencoders with outlier detection and federated SVDD loss filter noisy samples in non-IID federated learning, yielding accuracy gains up to 7% on CIFAR-10.
citing papers explorer
-
Quantamination: Dynamic Quantization Leaks Your Data Across the Batch
Dynamic quantization creates side channels allowing partial or full recovery of other users' batched data in at least four popular ML frameworks.
-
LAVA: Layered Audio-Visual Anti-tampering Watermarking for Robust Deepfake Detection and Localization
LAVA is a layered audio-visual watermarking system using cross-modal fusion and calibration-aware alignment to achieve robust deepfake tamper detection and localization under compression and asynchrony.
-
Predictive-Generative Drift Decomposition for Speech Enhancement and Separation
SIPS decomposes stochastic interpolant dynamics into predictive drift and generative denoising to combine arbitrary pretrained predictors with a degradation-agnostic clean-speech prior for better speech enhancement and separation.
-
Neural Aided Kalman Filtering for UAV State Estimation in Degraded Sensing Environments
The Bayesian Neural Kalman Filter uses a trained BNN to predict UAV states and uncertainties, then applies a Kalman update to outperform standard EKF and UKF on synthetic data under high noise and low sampling rates.
-
WhisperPipe: A Resource-Efficient Streaming Architecture for Real-Time Automatic Speech Recognition
WhisperPipe delivers 89 ms median latency and 48% lower peak GPU memory than standard Whisper while keeping word error rate within 2% of the offline model.
-
Sample Selection Using Multi-Task Autoencoders in Federated Learning with Non-IID Data
Multi-task autoencoders with outlier detection and federated SVDD loss filter noisy samples in non-IID federated learning, yielding accuracy gains up to 7% on CIFAR-10.