Tensor: multidimensional array carrying model inputs/weights/activations.
Operational intensity: arithmetic work per byte moved, core roofline axis.
Quantization: reduced numeric precision for efficient inference deployment.
Dataflow: mapping strategy for data reuse in accelerator execution.
Surrogate model: faster approximation of expensive EDA analyses.
Calibration: alignment between predicted confidence and observed accuracy.
Drift: statistical shift between training and production data/behavior.