Bridging Critical Gaps in Convergent Learning: How Representational Alignment Evolves Across Layers, Training, and Distribution Shifts

Chaitanya Kapoor (University of California, San Diego) · Sudhanshu Srivastava (University of California, San Diego) · Meenakshi Khosla (UC San Diego)
alignment metricsarchitectural biasesconvergent learningdistribution shiftinternal representationslayer-pair comparisonslinear regressionorthogonal procrustesorthogonal transformationsout-of-distribution imagespermutation matchingrepresentational basisrepresentational convergenceshared input statisticstraining epochsvision models

Understanding convergent learning