Tag: artificial neural networks
Browse our exclusive articles!
VFA: Boost Flash Attention with Global Max Pre-computation
Discover VFA, a novel method enhancing Flash Attention efficiency using global maximum pre-computation to reduce vector operation bottlenecks.
GF-Score: Fair, Certified Robustness Evaluation per Class
Discover GF-Score, an attack-free framework for certified class-wise robustness evaluation with fairness metrics to ensure balanced neural network security...
SpanKey: Secure Neural Network Access via Dynamic Key Conditioning
Discover SpanKey, a lightweight dynamic key conditioning method for efficient neural network access control without encrypting weights.
Socrates Loss: Enhancing Confidence Calibration & Classification
Discover Socrates Loss, a unified approach improving confidence calibration and classification stability in deep neural networks for reliable AI.
Layer-wise Insights for Efficient Supervised Fine-Tuning
Discover how selective layer tuning improves supervised fine-tuning, reducing forgetting and boosting model performance by 10%+ on GSM8K.