Tom Joy

adversarial learning (1) uncertainty quantification (1) object detection (1) direct preference optimization (1) preference alignment (1) preference optimization (1) autonomous driving (1) confidence calibration (1) out-of-distribution detection (1) mechanistic interpretability (1) safety fine-tuning (1) domain shift (1) temperature scaling (1) adversarial input (1) mlp weight transformation (1) jailbreak defense (1) expected calibration error (1) neural network calibration (1) large language model (1) neural network (1)