Computational challenges & opportunities P. Perona California - PowerPoint PPT Presentation

Computational challenges & opportunities P. Perona California Institute of Technology 4 December 2014

Press alarm?

Time is money ER worse error rate better response time RT

$ feeding frienzy picking apples t

Tuned neurons orientation and frequency [Graf and Movshon 2012] [Hubel & Wiesel]

Principles of model selection • Optimal observer • Occam’s razor

Wald’s sequential analysis I 2 I 1 0 t 2 t k t 1 t [ [ [ [ (0 , T ) = I 0 t 1 I 1 t 2 · · · Drift log R ( I k ) = log P ( n = 0 | C = 2 , t ∈ I k ) P ( n = 0 | C = 1 , t ∈ I k ) N = log Π N i =1 P ( n = 0 | λ i 2 ∆ t k ) X ( λ i 1 − λ i 1 ∆ t k ) = ∆ t k 2 ) Π N i =1 P ( n = 0 | λ i i =1 δ t → 0 log P ( n = 1 | λ i ( k ) δ t ) 2 log R ( t k ) = lim Action potential P ( n = 1 | λ i ( k ) δ t k ) 1 jumps δ t ) 1 e − λ i ( k ) δ t → 0 log ( λ i ( k ) δ t = log λ i ( k ) δ t 2 2 2 = lim δ t ) 1 e − λ i ( k ) λ i ( k ) ( λ i ( k ) 1 1 1

Model T 1 z’ Bow T 0 S S S decision S S S hypercolumn: N = 32 L L λ max ... ... ∆ θ = 25 0 λ min = 1

Predictions vs data

Subjects’ parameters

Technology

Summary (1) • Perception in the context of behavior • Delay vs error • Account for action potentials • Ideas for technology?

What is behavior? Marla B. Sokolowski Nature Reviews Genetics 2, 879-892 (2001)

From images to actions eat t tracking classification t 1 y mate drink t 2 x Video stream Trajectory Ethogram dynamics 10Mb/s 10kb/s 1b/s x ( t ) I ( x, y, t ) ⇤ ( t 1 , b 1 ) . . . ( t N , b N )

[Branson et al. Nature Methods, Jun. ’09]

Phenotyping [Dankert et al., Nature Methods, 2009]

Ethograms [Dankert et al., Nature Methods, April 2009]

Unsupervised analysis Center 1 Center 2 Center 3 Center 4 Center 5 Center 6 Center 7 Center 8 Center 9 Center 10 Center 11 Center 12 Center 13 Center 14 Center 15 Center 16 Center 17 Center 18 Center 19 Center 20 Center 21 Center 22 Center 23 Center 24 Center 25 Center 26 Center 27 Center 28 Center 29 Center 30 Center 31 Center 32 Center 33 Center 34 Center 35 Center 36 Center 37 Center 38 Center 39 Center 40 Center 41 Center 42 Center 43 Center 44 Center 45 Center 46 Center 47 Center 48 Center 49 Center 50 Center 51 Center 52 Center 53 Center 54 Center 55 Center 56 Center 57 Center 58 Center 59 Center 60 Center 61 Center 62 Center 63 Center 64 23

Behaviors that stand out Dendrogram of distances mean distance from i Percentile distance from i 2 1.6 1.8 1.8 1.5 1.6 1.6 1.4 1.4 1.3 1.4 1.2 1.2 dist dist 1 1.1 1.2 0.8 1 1 0.6 0.9 0.4 0.8 0.8 0.2 0.7 0 1 5 2 6 4 3 1 2 3 4 5 6 1 2 3 4 5 6 i i

Summary (2) • Genes -> Brains -> Behavior • Blue-collar engineering work • Unsupervised discovery

Grand challenges • Genes -> Brains -> Behavior • Theory of behavior (info, control) • Spike-based computation