Deep Learning

Deep learning predicts structural vibrations faster than traditional methods, reducing noise in airplanes, cars, and buildings, as shown by a new benchmark and frequency-query operator network.

Learn infinitesimal generators of stochastic diffusion processes efficiently via a novel energy-based risk functional, overcoming the unbounded nature of the generator and providing learning bounds in…

Learn data symmetries directly from data with flexible weight-sharing using learnable doubly stochastic tensors!

Learn macroscopic dynamics efficiently using only partial microscopic force computations! This novel method leverages sparsity assumptions and stochastic estimation for accurate, cost-effective modeli…

Learn continuous symmetries from data without pre-defined groups using Neural ODEs and a novel validity score to improve model generalization and efficiency.

Boost deep learning generalization with Learning from Teaching (LOT)! LOT trains auxiliary ‘student’ models to imitate a primary ’teacher’ model, improving the teacher’s ability to capture generalizab…

OPCR, a novel one-step spatio-temporal imputation method, surpasses existing iterative approaches by directly propagating limited observations to the global context, achieving superior accuracy and ef…

Neural networks learn efficiently from higher-order correlations, exceeding the capabilities of random features, as demonstrated through hypothesis tests and novel theoretical analysis in high-dimensi…

Layer-Adaptive STate pruning (LAST) optimizes deep state space models by efficiently reducing state dimensions, improving performance and scalability without retraining.

Large stepsize GD on non-homogeneous neural networks shows monotonic risk reduction after an initial oscillating phase, demonstrating implicit bias and optimization gains.

Novel intermediate variables regularization boosts knowledge graph completion!

Boosting Out-of-Distribution Detection with Kernel PCA!

IsoNet++ iteratively refines subgraph matching via early interaction GNNs and node-pair partner interactions, significantly boosting graph retrieval accuracy.

Unlocking efficient non-negative function approximation: This paper introduces inverse M-kernels, enabling flexible, linear universal approximators for one-dimensional inputs.

Spectral Attention boosts long-range dependency capture in time series forecasting, achieving state-of-the-art results across various models and datasets.

Deep learning framework integrating GNNs and neural ODEs precisely estimates non-reciprocal two-body interactions in mixed-species collective motion, accurately replicating both individual and collect…

Deep Equilibrium Models (DEQs) infused into DFT Hamiltonian prediction achieves self-consistency, accelerating large-scale materials simulations.

Calibrated Bayesian inference achieved via novel diffusion models uniquely mapping high-dimensional data to lower-dimensional Gaussian distributions.

Researchers reveal how the training dynamics of transformer models behave at infinite width, depth, and head count, providing key insights for scaling up these models.

Researchers developed a method using variational sequential Monte Carlo to fit stochastic low-rank recurrent neural networks to neural data, enabling efficient analysis and generation of realistic neu…