Deep Learning

Post-hoc model transformations can reverse performance trends, prompting a reevaluation of model selection strategies and suggesting a new ‘post-hoc selection’ method for improved model development.

PHYSS-GP: a novel physics-informed state-space Gaussian process model for efficient spatio-temporal data modeling, outperforming existing methods in predictive accuracy and computational speed.

TREAT: a novel framework boosting dynamical systems modeling accuracy by enforcing Time-Reversal Symmetry (TRS) via a regularization term. High-precision modeling is achieved across diverse systems, …

Physically consistent multi-task learning bridges heterogeneous molecular data by directly leveraging physical laws to improve predictions, enhancing accuracy beyond the limitations of individual data…

TPGN, a novel framework for long-range time series forecasting, uses Parallel Gated Networks (PGN) to efficiently capture long-term dependencies, achieving state-of-the-art results on multiple dataset…

Persistent Test-Time Adaptation (PeTTA) prevents AI model collapse in recurring scenarios by dynamically adjusting the adaptation strategy based on divergence from the initial model, ensuring long-ter…

pcaGAN boosts posterior-sampling cGANs by using principal component regularization, achieving faster, more accurate results in various imaging tasks.

HOAM learns reduced models of population dynamics for complex systems, enabling fast predictions across various physics parameters, outperforming state-of-the-art techniques.

PageRank Bandits (PRB) revolutionizes link prediction by framing it as a sequential decision-making problem, thus enabling the system to adapt to evolving data. Combining contextual bandits with PageR…

PACE, a novel neural operator, achieves unprecedented accuracy and speed in optical field simulation for complex photonic devices, surpassing existing methods by significantly reducing errors and boos…

P2C2Net: A physics-encoded neural network efficiently predicts complex spatiotemporal dynamics using coarse grids and limited training data, achieving state-of-the-art results.

Boost OOD detection accuracy with diverseMix: a novel method enhancing auxiliary outlier diversity, provably improving generalization and achieving state-of-the-art results.

Ordered Momentum (OrMo) significantly boosts asynchronous stochastic gradient descent (ASGD) convergence by cleverly incorporating momentum, resolving prior convergence issues. This novel approach is…

Vector-valued spectral learning algorithms finally get rigorous theoretical backing, showing optimal learning rates and resolving the saturation effect puzzle.

This paper introduces open-book neural algorithmic reasoning, a novel framework that significantly enhances neural reasoning capabilities by allowing networks to access and utilize all training instan…

ORI: a novel online relational inference framework efficiently identifies hidden interaction graphs in evolving multi-agent systems using streaming data and real-time adaptation.

Truncated kernel methods consistently outperform standard methods by eliminating the saturation effect, offering faster learning rates and enhanced theoretical guarantees.

Giant leap in molecular GNNs! MolGPS, a new foundation model, achieves state-of-the-art performance on molecular property prediction by leveraging massive datasets and demonstrating the scalability o…

Boosting certified robustness of machine learning models by 3-4% using generated data from diffusion models!

Fractal dimension, while showing promise, fails to consistently predict neural network generalization due to hyperparameter influence and adversarial initializations; prompting further research.