Systematic review finds attention explanations are rarely tested for faithfulness
A systematic review of 168 primary studies found that 57.14% did not examine the performance–faithfulness relationship, 41.07% reported synergy and ...
Tag: convolutional neural networks
AI Models Predict Fresh Food Supply Chain Resilience
A new study in Scientific Reports introduces a hybrid framework using variable fuzzy sets and convolutional neural networks to assess ...
AI searches for a better, lightweight network to predict epileptic seizures
A neural architecture search framework called CMS-NAS automatically evolves hybrid convolutional, Mamba and spiking network designs for lightweight, accurate EEG-based ...
More Detail in AI Prompts Can Backfire When Chatbots Build Safety Models
A controlled study of four frontier chatbots building helmet-detection models finds that the most detailed prompts can trigger catastrophic training ...
Deep Learning Is Teaching Machines to Read the Brain’s Electrical Whispers
A systematic review maps how deep learning architectures are transforming EEG analysis for neurological disease diagnosis, while warning that data ...
AI on a Tiny Chip: Deep Learning Spots Treacher Collins Syndrome at the Edge
Researchers have built a deep learning framework that detects Treacher Collins syndrome from facial images on an NVIDIA Jetson Orin ...
When Neural Networks Learn Like Brains: Locality Beats Weight Sharing in Hebbian Vision Models
A new computational neuroscience study shows that biologically constrained Hebbian learning can build useful visual representations without exact weight sharing, ...
AI Ensemble Reads Brain Scans With 98.5% Accuracy and Shows Its Work
Researchers in India have built a unified deep learning framework that classifies brain tumors in MRI scans with 98.52 percent ...
AI Reads Brain Scans Twice: Dual-Task Deep Learning Spots Tumors and Names Them in Milliseconds
Researchers have built a modified dual-task DenseNet121 that classifies brain tumors from MRI scans with over 99 percent accuracy while ...
AI Pipeline Pinpoints Hidden Brain Regions in MRI Scans
A new AI framework called SliceMap uses lightweight two-dimensional neural networks to efficiently localize subtle discriminative regions in brain MRI, ...
When Accurate Tracking Isn’t Enough: New Score Ranks AI Motion Systems by Biomarker Reliability
A new reliability-centred framework shows that frame-wise tracking accuracy alone fails to capture how faithfully AI pose estimation preserves downstream ...
Deep Learning Is Transforming How the World Spots Sick Plants, Landmark Review Finds
A sweeping review of 1,081 studies from 2014 to 2024 shows convolutional neural networks rapidly becoming the world's leading tool ...
AI Listens for Coughs After Lung Surgery With Near-Perfect Accuracy
Researchers in Wuhan developed a two-stage deep learning system that detects and classifies postoperative coughs from audio recordings of lung ...
Pixels Become Cuboids: Turning RGB Values Into 3D Blocks Helps AI See and Draw
Researchers have reimagined pixels as three-dimensional cuboids whose sizes are set by RGB values, using the empty space between them ...
Old-School Convolutional Networks Strike Back in Long-Range Time Series Forecasting
A new convolutional architecture called MRTFNet combines multi-resolution temporal embeddings with frequency-domain modeling to outperform Transformer-based models on most long-term ...
Smart Federated Learning Brings Energy-Sipping AI to Battery-Powered IoT Devices
Researchers at Jerash University have developed an energy-aware federated deep learning framework that cuts per-round energy use by about 43 ...
How the Brain Unscrambles a Visually Warped Alphabet
A new computational study shows that human vision handles orientation-redundant distortions better than orientation noise, revealing how spatial imprecision in ...
Hidden Motion Maps Reveal a Secret Code in the Brain’s Visual Cortex
A computational study of cat primary visual cortex shows that motion direction and temporal frequency are encoded in subtle, spatially ...
Deep Learning Models Hit 98% Accuracy in Mammogram Breast Cancer Detection
A new benchmark study shows that data augmentation lifts pre-trained deep learning models to up to 98.06% accuracy in classifying ...
Hybrid Optimizer Sharpens Deep Learning’s Eye for Skin Cancer
Researchers in India report that a hybrid ADAMW plus SGD optimizer with embedded feature selection boosts deep learning accuracy for ...
