Search papers, labs, and topics across Lattice.
2
0
6
2
Unlearning one demographic group in CLIP doesn't eliminate bias, it just shifts it, primarily along gender lines, revealing a concerning gender-dominant structure in the model's embedding space.
Topological data analysis reveals that structurally stable and compact soft prompts lead to better downstream performance, enabling a new loss function (TSLoss) that improves convergence and tuning performance.
