Cheaper attention mechanism gives AI models calibrated confidence scores

Original: Random Feature Gaussian Process Attention: Linear-Time Probabilistic Attention with Calibrated Uncertainty

Source: arxiv.org ↗