Rats: 14 Long Evans rats. Controlled water schedule → water reward & punishment noise.
Training: Automated training protocol
Stimulus duration: 0.5 - 2 sec.
Sessions: At least 50 trials. Percent correct above 75% and fixation-drop rate below 25%. Average of 120 min on a single-day. Anti-biasing algorithm generates trials with correct answer on opposite side of rat’s favored side.
Claims from paper
Aim of study
probe whether rodents can optimally discount evidence by adapting the timescale over which they accumulate it.
Claimed results
Optimal timescale for evidence discounting depends on both:
environment volatility
noise in sensory processing
Rats accumulate(almost) optimally, if both variables above are considered.
Rats adapt their integration timescale to the volatility of the environment(short → long → short).
Model makes quantitative predictions about timing of changes of mind.
Overall, paper establishes a quantitative behavioral framework to study adaptive evidence accumulation.
Optimal evidence accumulation model
Called‘nonlinear theory’ in the paper
dtda=δR,t−δL,t−κ2hsinh(κa)
a is the posterior-odds ratio
δR,t,δL,t are the right and left auditory click trains(sum of delta functions)
h is the hazard rate, or volatility of the environment
κ is the click reliability parameter. It indicates how much evidence a single click provides.
Effect of sensory noise on click reliability, discount rate and changes of mind timing(Fig 2 in paper)
Linear model
Called ‘linear discounting agent’ in the paper
dtda=δR,t−δL,t−λ⋅a
λ is the discounting rate
1/λ is the integration timescale
Discounting rate λ varies with sensory noise n(panel G below)
Behavioral λ is estimated by fitting an exponential, aebt, to the reverse kernel curve
Task
Claims from paper
Aim of study
Claimed results
Optimal evidence accumulation model
Linear model