Fisher-Guided Progressive Parameter Selection for Adaptive Fine-Tuning

Name: Fisher-Guided Progressive Parameter Selection for Adaptive Fine-Tuning
Uploaded: 2026-10-01T12:45:34.441318+00:00
Duration: 569 s
Description: The episode discusses a paper proposing Fisher-Guided Progressive Parameter Selection for Adaptive Fine-Tuning. The framework uses the temporal evolution of Fisher geometry to dynamically select which parameter groups to keep trainable during fine-tuning by freezing those whose influence has stabilized, leading to more principled and efficient adaptation strategies.

The episode discusses a paper proposing Fisher-Guided Progressive Parameter Selection for Adaptive Fine-Tuning. The framework uses the temporal evolution of Fisher geometry to dynamically select which parameter groups to keep trainable during fine-tuning by freezing those whose influence has stabilized, leading to more principled and efficient adaptation strategies.

Open the full episode