Extended Data Fig. 5: The data are better captured by FINDR than by a variant of FINDR constrained to parametrize the dynamics described by previously proposed hypotheses.
From: Transitions in dynamical regime and neural mode during perceptual decisions
a, The constrained FINDR (cFINDR) model replaces the neural networks parametrizing F in FINDR with a linear combination of affine dynamics, specified by M and N, and bistable attractor dynamics specified by φ. The dynamics are furthermore constrained to be two-dimensional. b-g, cFINDR model can generate and infer dynamics described by previous hypotheses. b, Example bistable attractor dynamics generated from cFINDR. c, Example DDM line attractor dynamics generated from cFINDR. d, Example non-normal dynamics with a line attractor generated from cFINDR. e, cFINDR-inferred dynamics from a synthetic dataset generated using the bistable attractor dynamics in b. f, cFINDR-inferred dynamics from a synthetic dataset generated using the DDM line attractor dynamics generated in c. g, cFINDR-inferred dynamics from a synthetic generated using the non-normal dynamics with a line attractor in d.
