Search papers, labs, and topics across Lattice.
This paper introduces the adjusted cup product neural layer, a novel neural primitive that integrates the cup product with an adjustment term derived from higher gauge theory, ensuring gauge invariance in its outputs. The key theoretical finding reveals that the output on a closed cycle is solely dependent on the adjustment coefficient, which, when set to zero, nullifies the output regardless of other parameters. This establishes the adjustment as the exclusive source of gauge-invariant signals and confirms its status as a nonzero quadratic form invariant under one and two gauge transformations.
The adjusted cup product neural layer reveals that gauge invariance in neural networks can hinge entirely on a single adjustment coefficient, fundamentally altering our understanding of signal generation in these systems.
Many important observables in physics and geometry are cup products of cochains. The adjusted cup product neural layer has been introduced in this paper. It is a neural primitive that hard wires the cup product with an adjustment term from higher gauge theory. This creates a readout that is gauge invariant by design. Their main theoretical result shows that on a closed cycle the output relies entirely on the adjustment coefficient. Setting this coefficient to zero removes the output completely regardless of other parameters. Thus the adjustment is the only source of gauge invariant signal. They prove this observable is a nonzero quadratic form and is exactly invariant under one and two gauge transformations.
