Living in a volatile and unpredictable world, we rely on sensory evidence that is noisy, ambiguous, and of variable quality. Optimal inference under these conditions involves continuously accumulating and discarding statistical evidence for and against alternative world states [2,3]. Suitable neural representations (‘population codes’) can reduce this inference problem to summing or subtracting neural activity over time [3,4]. We believe that multi-stable visual perception may reveal the dynamical circuit underlying continuous inference. Having characterized multi-stable phenomena in some detail [1], we can now reconstruct the underlying dynamics in terms of a hierarchical interaction between small populations of bistable assemblies. Remarkably, many aspects of this reconstructed dynamics and its operating regime — Poisson-like variance, exponential coupling, drift-dominance, differential threshold, winner-take-all decision, decision-triggered-adaptation — satisfy the requirements for optimal continuous inference [2-4]. Given generic sensory inputs, the reconstructed dynamical system accumulates and evaluates noisy evidence to make nearly optimal categorical choices. In addition, the reconstructed dynamics of bistable assemblies, which combines a graded representation of evidence with a categorical representation of choice, potentially offers a plethora of neurophysiological predictions [5,6].
References:
[1] Cao, Pastukhov, Mattia, Braun (2016) Collective activity of many bistable assemblies reproduces characteristic dynamics of multistable perception. J. Neurosci., 36: 6957-72.
[2] Bogacz, Brown, Moehlis, Holmes, Cohen (2006). Psychological Review, 113: 700-765.
[3] Veliz-Cuba, Kilpatrick, Josic (2016). SIAM Review, 58: 264-289.
[3] Ma, Beck, Pouget (2008) Current Opinion in Neurobiology, 18: 217-222.
[4] Pouget, Beck, Ma, Latham (2013). Nat. Neurosci., 16: 1170-1178.
[5] Latimer, Yates, Meister, Huk, Pillow (2015) Science 349, 184-7
[6] Engel, Steinmetz, Gieselmann, Thiele, Moore, Boahen (2016). Science 354, 1140-4.
