Neural Circuits Trained with Standard Reinforcement Learning Can Accumulate Probabilistic Information during Decision...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Neural circuits accumulate evidence for decision-making.
• Current models require dedicated synaptic plasticity rules for learning log-likelihood ratios.
• Log-likelihood ratio accumulation is crucial for choices between options.

Purpose of the Study:

• To investigate if standard reinforcement learning rules can estimate log-likelihood ratios.
• To propose a model that avoids novel plasticity rules for probabilistic inference.
• To demonstrate that reinforcement learning mechanisms can support efficient decision-making.

Main Methods:

• Developed a computational model linking expected rewards to log-likelihood ratios.
• Utilized standard reinforcement learning principles for model training.
• Simulated the model on tasks involving probabilistic cue accumulation.

Main Results:

• Log-likelihood ratios are approximately linearly proportional to expected action rewards.
• A reinforcement learning-based model successfully estimated log-likelihood ratios from experience.
• Simulations replicated experimental behavioral and neural data.

Conclusions:

• Standard reinforcement learning mechanisms are sufficient for estimating log-likelihood ratios.
• The brain may not require dedicated plasticity rules for probabilistic inference.
• Reinforcement learning offers a simpler explanation for neural decision-making processes.