adaptiveOptimalTraining

This package includes the matlab code for the paper:

• Ji Hyun Bak, Jung Yoon Choi, Athena Akrami, Ilana Witten, Jonathan Pillow. (2016) Adaptive optimal training of animal behavior. Advances in Neural Information Processing Systems 29. [link]

Sample scripts:

getSimDat.m: generates a simulated behavior dataset (run this first!) - surrogate for a real animal behavior dataset.

AOT_script_estWgt.m: script for analyzing past observations,

• first with the random-walk prior only (hyperparameter sigma)
  • corresponds to Fig 2 in paper
• then with added learning component as drift (hyperparameter alpha)
  • corresponds to Fig 3 in paper

AOT_script_training.m: script for simulated active/passive training - corresponds to Fig 4 and S2 in paper

Core functions:

• funs_MNLogistic.m: (this is a script) contains basic operations for (multinomial) logistic model usually called at the beginning of each core function

• getMAP_RWprior.m: does the MAP estimate for the weights with the random walk prior

  • getLP_MNLogistic_RWprior (core external subfunction)
  • negLogPost_MNLRW (wrapper for getLP)

• getSimRat_active.m: runs a simulated active training experiment

  • calls getPolGrad_discrimTask.m

• getPolGrad_discrimTask.m: calculates the policy gradient and the higher gradients, taylored for the specific task / model structure