Machine Learning

Attackers can steal massive private data from federated learning diffusion models using multiple Trojans and an advanced attack, AdaSCP, which circumvents existing defenses.

Data-efficient neural operator learning is achieved via unsupervised pretraining and in-context learning, significantly reducing simulation costs and improving generalization.

Data-Free Backdoor Attacks (DFBA) injects undetectable backdoors into pre-trained classifiers without retraining or architectural changes, bypassing existing defenses.

Boosting semi-supervised learning, a new data augmentation method using diffusion models significantly improves model accuracy, especially with mismatched data distributions.

Federated data acquisition via experimental design (DAVED) achieves lower prediction error without labeled validation data, optimizing cost-effectively for test-set predictions in decentralized market…

DASH combats neural network training’s plasticity loss during warm-starting by selectively forgetting memorized noise while preserving features, improving accuracy and efficiency.

D2R2: A novel diffusion-based model for tabular few-shot learning, achieves state-of-the-art results by leveraging semantic knowledge and distance matching.

CoLA: Collaborative Lifelong Adaptation boosts test-time adaptation efficiency by sharing domain knowledge across multiple devices, achieving significant accuracy gains with minimal computational over…

CRONOS: Scaling convex neural network training to ImageNet!

Credal Deep Ensembles (CreDEs) improve uncertainty quantification in deep learning by predicting probability intervals, enhancing accuracy and calibration, particularly for out-of-distribution data.

Biologically inspired Counter-Current Learning (CCL) uses dual networks for deep learning, offering comparable performance to other biologically plausible algorithms while enhancing biological realism…

This paper accelerates second-order optimization in CNNs by 4.5x, using a novel tensor network representation that simplifies convolutions and reduces memory overhead.

Split Federated Learning (SFL) convergence is analyzed for heterogeneous data, achieving O(1/T) and O(1/√T) rates for strongly convex and general convex objectives respectively. The study also extend…

ControlSynth Neural ODEs (CSODEs) guarantee convergence in complex dynamical systems via tractable linear inequalities, improving neural ODE modeling.

Continuous Relaxation Annealing (CRA) significantly boosts unsupervised learning-based solvers for combinatorial optimization by dynamically shifting from continuous to discrete solutions, eliminating…

NeuroMOP, a novel neural principle, maximizes neural variability while ensuring reliable performance in recurrent neural networks, offering new insights into brain function and artificial intelligence…

This research introduces a hypothesis test and a contrastive dimension estimator to identify unique foreground information in contrastive datasets, advancing the field of dimension reduction.

CONTRAST efficiently adapts multiple source models to dynamic data distributions by optimally weighting models and selectively updating only the most relevant ones, achieving robust performance withou…

Koodos: a novel Koopman operator-driven framework that tackles Continuous Temporal Domain Generalization (CTDG) by modeling continuous data dynamics and learning model evolution across irregular time …

CutSSL: a novel framework for graph-based semi-supervised learning, surpasses state-of-the-art accuracy by solving a continuous nonconvex quadratic program that provably yields integer solutions, exce…