TY  - RPRT
TI  - A Light Bilevel Refinement Aligns Self-Supervised Representations for Stronger Task-Specific Learning
AU  - Gustav Wagner Zakarias
AU  - Zheng-Hua Tan
PY  - 2026
UR  - https://arxiv.org/abs/2609.33424
ID  - 2609.33424
ER  -