Adaptive stimulus selection (aka "active learning") for 1D and 2D tuning curves for closed-loop experiments in Matlab.

Description: Selects optimal stimuli from a 1D or 2D grid according to infomax (maximizing mutual information between response and model parameters) or uncertainty sampling (stimulus for which tuning curve has maximal uncertainty), for Poisson neurons with tuning curves modeled as either:

1. a parametric function (demo1).
2. a nonlinearly transformed Gaussian process (demo2 and demo3).

Relevant publications:

• Pillow & Park (2016). Adaptive Bayesian methods for closed-loop neurophysiology. [link]

• Park, Weller, Horwitz, & Pillow (2014). Bayesian active learning of neural firing rate maps with transformed Gaussian process priors. Neural Computation, Neural Computation 2014 [link]

• Download: zipped archive activelearningTCs-master.zip
• Clone: clone the repository from github: git clone https://github.com/pillowlab/activelearningTCs.git

• Launch matlab and cd into the directory containing the code (e.g. cd code/activelearningTCs/).

• Examine the demo scripts for annotated example analyses of simulated datasets:

  • demo1_parametricTC_1Dinfomax - demo for parametric 1D tuning curve with infomax stim selection
  • demo2_gpTC_1D.m - demo for non-parametric 1D tuning curve under transformed GP prior
  • demo3_gpTC_2D.m - demo for non-parametric 2D tuning curve under transformed GP prior