Where does stacking collide?

In reflection seismology, stacking is the summation of many seismic traces that record the same subsurface point from different source-receiver offsets; because the reflected signal is coherent across those traces while the noise is not, adding them cancels random noise and sharpens the image. Mayne's 1962 Common reflection point horizontal data stacking techniques introduced common-reflection-point stacking, still the backbone of seismic imaging. The referent is a noise-reduction operation on redundant geophysical recordings, unrelated to aromatic chemistry or ensemble learning save for the word.[1]

Unit

The signal-to-noise gain from summing redundant traces at a common point

Recovery means

Not applicable

Measured by

Judged by the signal-to-noise improvement in the stacked section versus single traces

In chemistry, stacking (π-stacking) is the attractive noncovalent interaction between aromatic rings that leads them to arrange in parallel face-to-face or offset geometries, an organising force in DNA base pairs, protein folds, and molecular crystals. Hunter and Sanders' 1990 The nature of π-π interactions gave the electrostatic model that explains why the rings favour an offset rather than a perfectly eclipsed stack. The referent is a geometric, energetic relationship between molecules, a wholly different construct from summing seismic traces or combining predictive models.[2]

Unit

The interaction energy and geometry between stacked aromatic rings

Recovery means

Not applicable

Measured by

Quantified by interaction energies and ring-offset geometries from spectroscopy and computation

In machine learning, stacking (stacked generalization) is an ensemble method that trains a higher-level meta-learner to combine the predictions of several different base models, learning how much to trust each one rather than weighting them equally. Wolpert's 1992 Stacked generalization introduced the scheme as a way to reduce the generalization error of any single learner. The referent is a layered model-combination procedure, sharing only its name with seismic trace summation and aromatic-ring interactions.[3]

Unit

The meta-learner's combined output over several base models

Recovery means

Not applicable

Measured by

Measured by the stacked ensemble's generalization error against the base models it combines