Targeted feedback delivered to specific neurons during learning that drives parameter updates; MIT's 2026 finding shows these are more precise and localized than previously understood
