Readout-level mechanisms for interior distinguishability optima

Determine whether readout-level mechanisms can produce interior optima in distinguishability, as suggested by partial-coherence results, despite the impossibility of such optima at the level of photonic kernel geometry.

Background

The paper proves that, for the two-photon partial-distinguishability model, the kernel-induced model complexity is convex in visibility. Consequently, interior optima in distinguishability cannot arise from the geometric kernel structure itself and must instead originate from other components of the learning pipeline.

The authors identify readout-level mechanisms as a possible explanation, motivated by partial-coherence results reported in prior work. Establishing whether such mechanisms can genuinely generate interior performance optima remains unresolved.

References

whether readout-level mechanisms can produce them, as suggested by the partial-coherence results of ref., is an open question.

— When Does Interference Help Learning? Kernel Geometry as a Pre-Experimental Test for Photonic Reservoir Computing  (2610.11360 - Kar et al., 8 Oct 2026) in Discussion, paragraph beginning “Partial distinguishability, usually treated as an imperfection…”