Natural selection subsumes and unites multiple theories of perceptual compression

An ideal model of perceptual processing would be congruent with experimentally-observed perceptual performance and be theoretically generalizable. In the extant models with experimental predictive power, the most-invoked signal processing paradigms are rate-distortion theory, efficient coding, and Bayesian inference. A theoretical reconciliation of these alternatives has yet to emerge. Here I endeavor to constrain the model space by clarifying which processing paradigms align with natural selection objectives. I find that signal processing and fitness objectives often deviate; for example, in a finite game where risk-avoidant bet-hedging beats maximization of average performance. When these objectives do align, the specific signal processing performance metric that aligns with fitness over the most general set of circumstances is the minimization of utility-weighted signal distortion. Together, these results reveal that selection favors perceptual system adaptations to (at least) three types of noise; environmental, encoding, and decoding; and establish utility-weighted distortion minimization as the optimal adaptation to the second.