IMPLEMENTING CONTINUAL LEARNING SCENARIOS: REVIEW OF A ML TECHNQIUE
Ayşe Rüya Arslan · International Journal of Applied Science and Research · 2022
One of the biggest traps "AI" (artificial intelligence) projects fall into is demanding an entire solution fit into the ML (machine learning) paradigm. One of the continual learning techniques in ML to overcome this challenge is the so called 'Learning to Prompt for Continual Learning (L2P)' that can be applied to practical continual learning scenarios without known task identity or boundaries. L2P uses a single frozen backbone model and learns a prompt pool to conditionally instruct the model. After discussing main categories of recent continual learning algorithms, this paper provides an overview of LSP by discussing its layers.