Studies of visual discrimination learning decision making in rodents, particularly those using visual cues, can confound the effects of cue salience with reward value, making it difficult to determine which factor guides choice behavior. We addressed this issue using a two-alternative forced-choice task in which rats chose between visual cues associated with high or low sucrose rewards. After initial training with high (16 LEDs) and low (1 LED) luminance cues, we introduced a novel cue of intermediate luminance (4 LEDs) in a "luminance shift" test, keeping the associated reward values constant. We found that while rats maintained a preference for the higher-value option, the introduction of a perceptually more similar cue consistently reduced choice preference and eliminated latency differences compared to baseline. Using drift diffusion modeling, we determined that the luminance shifts primarily caused a reduction in the drift rate (the speed of evidence accumulation), reflecting increased difficulty in cue discrimination. This finding suggests that the visual salience of the options dictates the efficiency of evidence accumulation in value-based decisions. Furthermore, this effect on drift rate is notable because it shows a dissociation from our previous work (Palmer et al., 2024), where prefrontal cortex inactivation specifically affected the decision threshold but not the drift rate. Our results demonstrate that the relative salience of visual stimuli influences deliberation, with low-level perceptual features shaping the computational dynamics of value-based choice. Our findings clarify the distinct contributions of sensory input and prefrontal function in the decision process.
The Hidden Power of Normalization: Exponential Capacity Control in Deep Neural Networks
arXiv:2511.00958v1 Announce Type: cross Abstract: Normalization methods are fundamental components of modern deep neural networks (DNNs). Empirically, they are known to stabilize optimization dynamics and