Deep Learning

Deep learning optimization gets a major upgrade with Modula, a new method that uses the modular norm to normalize weight updates, enabling learning rate transfer across network widths and depths, thus…

SARAD: A novel anomaly detection approach for multivariate time series leverages spatial information and association reduction patterns to achieve state-of-the-art performance.

SAND, a novel transformer network variant, smoothly imputes sparse and noisy functional data by leveraging self-attention on derivatives, outperforming existing methods.

ConFrag, a novel approach to noisy label regression, leverages contrastive fragmentation and neighborhood agreement to select clean samples, significantly outperforming state-of-the-art baselines on s…

SampDetox uses diffusion models to purify poisoned machine learning samples by strategically adding noise to eliminate backdoors without compromising data integrity.

S2HPruner bridges the discretization gap in neural network pruning via a novel soft-to-hard distillation framework, achieving superior performance across various benchmarks without fine-tuning.

Rough Transformers: A lightweight continuous-time sequence modeling approach using path signatures to significantly reduce computational costs, improving efficiency and accuracy, particularly for long…

This paper introduces robust group inference procedures for high-dimensional single index models, offering substantial efficiency gains for heavy-tailed errors and handling group testing effectively w…

RMLR: A novel framework extends multinomial logistic regression to diverse geometries, overcoming limitations of existing methods by requiring minimal geometric properties for broad applicability.

Revive Re-weighting in Imbalanced Learning by Density Ratio Estimation dynamically adjusts class weights during training using density ratio estimation, significantly improving model generalization, e…

Boosting time series forecasting accuracy, Retrieval-Augmented Diffusion Models (RATD) leverage relevant historical data to guide the diffusion process, overcoming limitations of existing models and d…

LoCalPFN: boosting in-context tabular learning via retrieval & fine-tuning!

Boosting spiking neural network accuracy by 4.05% on ImageNet and achieving state-of-the-art results on CIFAR10-DVS and N-Caltech101 through learnable initial membrane potential and refined training s…

Revolutionizing long-term time series forecasting, a new Fourier Basis Mapping method enhances accuracy by precisely interpreting frequency coefficients and considering time-frequency relationships, a…

Stable algorithm learning achieved by Deep Thinking networks with Lipschitz Constraints, ensuring convergence and better extrapolation to complex problems.

MRConv: Reparameterized multi-resolution convolutions efficiently model long sequences, improving performance across various data modalities.

Relational Concept Bottleneck Models (R-CBMs) merge interpretable CBMs with powerful GNNs for high-performing, explainable relational deep learning.

This paper introduces a novel framework for classification with rejection by learning density ratios between data and idealized distributions, improving model robustness and accuracy.

RAMDA: a new algorithm ensures efficient training of structured neural networks by achieving optimal structure and outstanding predictive performance.

REDUCR, a novel data downsampling method, significantly improves worst-class test accuracy in imbalanced datasets by using class priority reweighting, surpassing state-of-the-art methods by ~15%.