A new arXiv preprint (2608.24568v1) proves that progressive grow-and-optimize training biases deep neural networks toward flatter loss-landscape minima via a volume effect from frozen constraints, but…
Researchers studying Focal Calibration Loss (FCL) found that while it preserves the Bayes decision rule for all gamma and lambda, it is generally not proper for gamma > 0, causing the Bayes-optimal pr…
Researchers propose Gated Adaptive Prompting (GAP-Prompt), a novel method for continual learning that introduces instance-level adaptability to prompt-based approaches, achieving state-of-the-art perf…
Researchers introduce Mixture of Channel Experts (MoCE), a structured sparse channel-mixing layer that replaces pointwise projections in convolutional networks, achieving a 16.7% reduction in MACs and…
Researchers introduced SAGE, an uncertainty-modulated surrogate-gradient mechanism for Transformer-based spiking neural networks (SNNs), which adapts the surrogate-gradient slope during training using…
A new arXiv preprint (2608.00078v1) presents a mobile deployment pipeline that uses QLoRA fine-tuning of an LLM to generate convolutional neural networks, with INT8 TensorFlow Lite export and on-devic…
Researchers propose a loss-aware feature-map pruning framework for convolutional neural networks that uses multi-armed bandits (UCB1 and Thompson Sampling) to identify and remove redundant feature map…
A new arXiv preprint (2607.20516v1) reports that scaling closed-loop LLM-based channel-configuration search to 250 candidate networks per cycle yields a positive linear accuracy trend (slope 9.87e-4, …
A study from arXiv reveals that a benchmark leak caused an OOD detector's AUROC to drop to 0.326, well below chance, because the designated OOD class was actually in the training set. After deleting t…
Researchers have introduced SechKAN, a Kolmogorov-Arnold Network architecture using hyperbolic secant functions, which achieves superior performance over MLPs and other KAN variants on benchmark datas…
RepTran, a new AI repair method targeting Transformer feed-forward networks, achieved a 74.7% average repair rate across 18 fault benchmarks from CIFAR-100 and Tiny-ImageNet, outperforming random sele…
Researchers introduced Meta Neural Cellular Automata (MetaNCA), a framework that learns local rules to self-organize the weights of artificial neural networks without backpropagation. The method gener…
Researchers introduced Pre-Warm, a zero-training-cost method for data-conditioned initialization of the first convolutional layer in CNNs. The technique clusters patches from a single training batch t…
Researchers introduced UtVAA, an ultra-tiny Vision Transformer architecture with a novel Affix Attention block for mobile image classification. The smallest variant has 204.67K parameters and 53.95M F…
Researchers at arXiv have identified that Batch Normalization causes gradient skew in dynamic sparse training (DST) methods, leading to slower convergence compared to dense neural network training. Th…