CONCEPT
Epigenetic Landscape
Waddington's 1957 image of a hillside grooved with branching valleys—each valley a developmental fate, the topography encoding which paths are easy and which are forbidden—now borrowed, name and all, by machine-learning researchers to describe the space a neural network descends during training.
The epigenetic landscape was drawn by
Conrad Waddington in 1957 to solve the central mystery of developmental biology: how does a single fertilized egg, carrying one genome, produce hundreds of different cell types with reliability, generation after generation, despite the noise and variation of the biological world? His answer was the landscape itself—a hillside of branching valleys, each valley a cell fate, the slopes and ridges encoding which paths are developmentally possible and which are forbidden. A region of tissue at the top of the slope is released and rolls downward; at each branch point it must commit to one channel or another; once committed, it is canalized toward the resting place at the bottom. The topography is not static fate but a dynamic product of the genome: Waddington paired the landscape with an image of its underside, a tangle of guy-ropes running from the valleys down to a field of pegs—the genes—whose collective