Brief

Researchers have developed Adaptive Signal Resuscitation (ASR), a novel training-free method to repair sparse vision networks after pruning. ASR addresses the accuracy collapse seen in high-sparsity models by applying corrections at a channel-wise granularity, unlike previous layer-wise approaches. This technique estimates and stabilizes variance-matching corrections for each output channel, significantly improving performance in high-sparsity scenarios. For instance, ASR recovered 55.6% top-1 accuracy on ResNet-50 at 90% sparsity on CIFAR-10, a substantial improvement over existing methods. AI

Researchers have characterized the fault response of the Intel Neural Compute Stick 2 (NCS2) when subjected to electromagnetic fault injection. Their experiments revealed four distinct outcome classes, including silent data corruption and persistent degradation of accuracy, which can occur in a significant percentage of trials at specific hotspots. Notably, these faults can persist until the model is reloaded and can even be triggered on an idle device, indicating that standard integrity checks are insufficient for safety-critical edge applications. AI

Researchers have developed MDS-DETR, a novel object detection model that improves upon the DEtection TRansformer (DETR) architecture. MDS-DETR addresses DETR's slow convergence and low recall issues by integrating both one-to-one and one-to-many label assignment within a single decoder. This is achieved through a Masked Duplicate Suppressor (MDS) that filters redundant predictions, leading to more efficient and accurate object detection. AI

A new research paper explores the effectiveness of transfer learning for industrial visual inspection tasks. The study compares DINOv3, a self-supervised model, against traditional ImageNet pretraining for RGB and X-ray defect detection. Results indicate DINOv3 offers benefits after full fine-tuning on RGB data, but ImageNet pretraining remains superior for X-ray applications. AI

Researchers have developed a Hybrid Quantum-Classical Neural Network (HQNN) architecture to improve the classification of blood cells in medical images. This approach combines a ResNet-50 backbone with a variational quantum circuit, demonstrating superior performance compared to purely classical models. Experiments showed a 3.7% improvement in macro F1-score on one dataset and a slight increase in F1-score on a more challenging 8-class scenario. The HQNN model also proved robust when tested on actual IBM quantum hardware, indicating practical potential for medical imaging tasks. AI

Researchers have published a study comparing how different learning rules in artificial neural networks align with visual processing in both humans and macaques. The study found that early visual cortex alignment was conserved across species, with artificial neural networks showing higher correlation with macaque electrophysiology data than with human fMRI data. However, at higher visual areas like the IT cortex, the alignment rankings of learning rules diverged significantly between species, suggesting that model capacity and training data play a larger role than the specific learning rule in these areas. AI