試験
通常
試験時間: 00:00問題時間: 00:00
問題一覧
問題 31 / 65
Which approach is suitable for quickly improving classification performance when very little data is available for learning new classes?
①Collect all the data and retrain from scratch.
②Fine-tune at scale through continual pretraining.
③Perform small-scale fine-tuning using LoRA or similar methods.
④Preserve existing parameters using regularization methods such as EWC.
⑤Prototype-based meta-learning (prototypical networks, etc.)