Fig. 1: Latent-cause inference.
From: Using computational models of learning to advance cognitive behavioral therapy
After observing a CS (e.g., a light) followed by a US (e.g., a shock), the CS comes to predict the US. The prediction is mediated by a latent cause C1 that jointly “emits” both CS and US. In exposure therapy, when the CS is observed but not followed by a US, the prediction can be either weakened through prediction-error driven learning (Option 1: dashed line signifying lower probability of the latent cause emitting the US, as per the new learning) or a new latent cause C2 is created, which only emits the CS (Option 2). CS = conditioned stimulus, US = unconditioned stimulus.
