Machine Learning

SPEAR, a novel algorithm, precisely reconstructs entire data batches from gradients in federated learning, defying previous limitations and enhancing privacy risk assessment.

SpeAr: A novel spectral approach significantly improves zero-shot node classification by using inherent graph structure to reduce prediction bias and effectively identifying unseen node classes.

Spatio-Spectral GNNs synergistically combine spatial and spectral graph filters for efficient, global information propagation, overcoming limitations of existing methods.

SparseLinUCB: First sparse regret bounds for adversarial action sets with unknown sparsity, achieving superior performance over existing methods!

SµPar stabilizes sparse neural network training, slashing tuning costs and boosting performance, especially at high sparsity levels, via a novel parameterization technique.

A new learnable prior for compressive sensing solves the inverse problem using only a few corrupted data samples, enabling sparse signal recovery without ground-truth information and uncertainty quant…

SPARKLE: A single-loop primal-dual framework unifies decentralized bilevel optimization, enabling flexible heterogeneity-correction and mixed update strategies for improved convergence.

SpaFL: A communication-efficient federated learning framework that optimizes sparse model structures with low computational overhead by using trainable thresholds to prune model parameters.

Sourcerer: A novel sample-based method for maximum entropy source distribution estimation, resolving ill-posedness while maintaining simulation accuracy.

SDEPO, a new natural policy gradient algorithm, efficiently solves zero-sum Markov games with continuous state spaces, achieving near-optimal convergence independent of state space cardinality.

SHORE: a novel two-stage framework efficiently solves sparse & high-dimensional output regression, boosting interpretability and scalability.

RC-PPO: Reinforcement learning solves minimum-cost reach-avoid problems with up to 57% lower costs!

Soft ascent-descent (SoftAD) improves test accuracy and generalization by softening the flooding method, offering competitive accuracy with reduced loss and model complexity.

Stochastic gradient bandit algorithms now guaranteed to globally converge, using ANY constant learning rate!

SLowcal-SGD, a new local update method for distributed learning, provably outperforms Minibatch-SGD and Local-SGD in heterogeneous settings by using a slow querying technique, mitigating bias from loc…

SleeperNets: A universal backdoor attack against RL agents, achieving 100% success rate across diverse environments while preserving benign performance.

Skill-aware Mutual Information optimization enhances RL agent generalization across diverse tasks by distinguishing context embeddings based on skills, leading to improved zero-shot performance and ro…

SkiLD, a novel unsupervised skill discovery method, uses state factorization and a new objective function to learn skills inducing diverse interactions between state factors, outperforming existing me…

This work presents a sharper analysis of sketching for distributed deep learning, eliminating the problematic dependence on ambient dimension in convergence analysis and proving ambient dimension-inde…

SLU: a novel, low-memory uncertainty score for neural networks, achieves logarithmic memory scaling with model parameters, providing well-calibrated uncertainties and outperforming existing methods.