AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Interventions in Model Behavior Change
Interventions in machine learning models can involve three key methods: knowledge editing which updates the model's represented knowledge through techniques like LoRa; unlearning where the model forgets previous information by training it to avoid generating certain outcomes using gradient ascent; and model compression which includes weight pruning and quantization methods to reduce unnecessary weights and lower resolution representation of the model.