AI Post Transformers Visual Companion

Reverse-Mode Differentiation Across AD and Neural Nets

A visual map of Werbos’s main claim: reverse-mode AD, backpropagation, hand-coded adjoints, and adjoint circuits are mostly the same computational trick wearing different disciplinary uniforms. This page emphasizes mechanism, historical fragmentation, and why derivative infrastructure changed what people could practically build.

Source Paper PDF Interactive Viz URL Podcast Home 2004 review paper Reverse mode as infrastructure AD · neural nets · control · circuits
Core Thesis Landscape
Shared reverse-pass logic Community-specific framing Ambition beyond evidence
Episode Tension Meter

Hover the gauges. The strongest support sits with gradient mechanics and neural-net practice; the weakest support sits with broader claims about intelligent control and brain-like local circuits.

References

Compact source list for the visual themes on this page.