CIFAR-100
PulseAugur coverage of CIFAR-100 — every cluster mentioning CIFAR-100 across labs, papers, and developer communities, ranked by signal.
- used by ResNet-18 90%
- instance of Class Incremental Learning 90%
- used by ScienceCast 70%
- used by Gotit.pub 70%
- used by Tiny-ImageNet 70%
- instance of Tiny-ImageNet 70%
- instance of federated learning 70%
- used by CatalyzeX 70%
- instance of ResNet-18 70%
- used by Imagenet 1k 70%
- instance of ScienceCast 70%
- used by ImageNet-100 70%
15 day(s) with sentiment data
-
Hybrid learning method boosts CNN training efficiency and accuracy
Researchers have developed a novel hybrid approach to accelerate the distributed training of convolutional neural networks (CNNs). This method combines a dual-batch learning scheme, which uses both large and small batch…
-
New method profiles Hessian mass in machine learning models
Researchers have developed a new method for evaluating block-diagonal curvature models in machine learning. This approach measures the squared off-block fraction of the parameter Hessian and analyzes the within-group cu…
-
New SSoG operator offers subquadratic cost for Vision Transformers
Researchers have developed a new spatial operator called SSoG (Separable sum-of-Gaussians) designed to replace the computationally expensive self-attention mechanism in Vision Transformers. This novel approach reduces t…
-
TrustMix method enhances pretrained models against distribution shifts
Researchers have introduced TrustMix, a novel post-hoc method designed to improve the robustness of pretrained models when faced with distribution shifts. This technique makes a model's extrapolative behavior explicit a…
-
New $S^3$-MoE framework enhances federated learning for LLMs
Researchers have developed a new framework called $S^3$-MoE for federated learning of large language models on heterogeneous edge devices. This server-side sparse assignment approach aims to improve expert relevance est…
-
New PAVE method accelerates DNN verification by 10x
Researchers have developed a new method called PAreto VErification (PAVE) to address the scalability challenges in formally verifying deep neural networks (DNNs). PAVE augments DNNs with early exits, allowing for the ve…
-
LLMs dynamically select aggregation rules in federated learning
Researchers have developed FedReason, a novel approach that utilizes large language models (LLMs) to dynamically select aggregation rules in federated learning. Instead of relying on fixed rules, FedReason employs an LL…
-
DISCERN method enhances incremental machine unlearning for transformers
Researchers have developed DISCERN, a novel method for incremental machine unlearning that addresses the challenge of sequentially removing data classes from a trained model while preserving remaining knowledge. DISCERN…
-
New IDSISA method improves machine unlearning efficiency and utility
Researchers have developed IDSISA, a novel method for efficient machine unlearning that aims to remove the influence of specific training data while maintaining model utility. Unlike full retraining, which is computatio…
-
New research identifies 'catastrophic overfitting' in AI training and proposes a fix
Researchers have identified a phenomenon called catastrophic overfitting (CO) in fast adversarial training (FAT) of neural networks. CO occurs when robustness to multi-step attacks rapidly degrades, even as the model co…
-
Federated learning research tackles client selection for efficiency · 2 papers
Two new research papers explore methods for improving federated learning efficiency by optimizing client selection. The first paper, FedCSGA, uses a genetic algorithm to jointly consider statistical and system heterogen…
-
Withdrawn arXiv paper proposed unified matrix-spectral framework for deep learning
A research paper, since withdrawn by its author Ronald Katende, proposed a unified matrix-spectral framework for analyzing stability and interpretability in deep learning models. The framework aimed to reveal spectral q…
-
DIFFNAS framework eliminates predictor need in neural architecture search
Researchers have developed DIFFNAS, a novel framework for Neural Architecture Search (NAS) that eliminates the need for performance predictors. By utilizing a discrete graph diffusion model and zero-cost evaluation metr…
-
MahaVar improves OOD detection using class-wise Mahalanobis distance variance
Researchers have introduced MahaVar, a novel method for detecting out-of-distribution (OOD) samples in deep neural networks. The technique leverages an empirical observation that in-distribution samples exhibit higher v…
-
New framework links neural network parameter motion to output changes
Researchers have introduced "Mode Dynamics," a new framework designed to understand how coordinated movements within a neural network's parameters relate to changes in its outputs. This framework decomposes parameter up…
-
New Spiking Neural Network Achieves Group Equivariance for Vision Tasks
Researchers have developed GC-LIF, a novel spiking neural network architecture that integrates group equivariance to improve performance on vision tasks. This new model, SGE-Net, addresses a key challenge in spiking neu…
-
AI models often sound confident even when wrong, a fluency vs. accuracy issue
A common failure mode in AI models is overconfidence, where they sound certain even when incorrect. This phenomenon, observed in models like ResNet-34 and across various architectures as noted by Guo et al. (2017), stem…
-
New RoCo-FL framework boosts federated learning robustness against noise and heterogeneity
Researchers have introduced RoCo-FL, a novel framework designed to enhance the robustness and convergence of Over-the-Air Federated Learning (OTA-FL) systems. This framework specifically addresses challenges posed by ch…
-
New AGAT method slashes adversarial training time by 70-80%
Researchers have developed a new method called Augmentation-Guided Adversarial Training (AGAT) to improve the efficiency of adversarial training for AI models. AGAT selectively applies adversarial supervision to only 20…
-
New Control Theory-Based Data Augmentation Algorithm Unveiled
Researchers have introduced ControlAugment (Ctrl-A), a novel data augmentation algorithm for image-vision tasks that leverages control theory principles. This method dynamically adjusts augmentation strength distributio…