A medical team wants an LLM that understands radiology terminology but cannot afford to retrain a model from scratch. Which approach is most appropriate?

Question

BlackTom AI · Accepted Answer

Transfer learning with fine-tuning leverages a pretrained language model’s knowledge and adapts it to a new domain with comparatively little compute and data, so fine-tuning is the appropriate approach.

1) Relevant knowledge points and concepts
- Pretrained LLMs have general linguistic and factual knowledge from large-scale unsupervised pretraining.  
- Fine-tuning (supervised or continued pretraining) adjusts a pretrained model’s weights on domain-specific labeled or unlabeled data.  
- Training from scratch or full domain pretraining is extremely compute- and data-intensive.  
- GANs are a generative modelling technique typically used for images and not the right paradigm for producing a domain-specific LLM.  
- RNNs are older architectures; training a new RNN-based language model from random initialization is inefficient and unlikely to match transformer LLM performance.

2) Step-by-step reasoning
- Constraint: team needs radiology terminology understanding but cannot afford full retraining.  
- Goal: add domain-specific knowledge efficiently.  
- Fine-tuning fits: start from a pretrained transformer LLM and adapt it with a radiology corpus or labeled examples, requiring far less compute than training from scratch.

3) Example calculation / intermediate steps (illustrative)
- Suppose you have 10,000 radiology notes, batch size 32, and you train 3 epochs. Number of updates = (10,000 / 32) * 3 ≈ 312.5 * 3 = 937.5 → about 938 gradient updates.  
- By contrast, pretraining from scratch typically needs billions of tokens and millions of updates (orders of magnitude more compute and cost).

4) Why the correct answer is correct
- Fine-tuning reuses learned parameters and requires only modest compute and data to specialize the model on radiology vocabulary, style, and tasks (e.g., report summarization, label extraction). It directly addresses the team’s budget constraint while delivering domain competence.

5) Why other options are incorrect
- Domain pretraining: often means large-scale continued pretraining on domain corpora which can still be expensive and closer to retraining than lightweight fine-tuning.  
- Training a GAN: GANs are not the appropriate architecture for building an LLM tuned for text understanding.  
- Training an RNN from random initialization: requires training from scratch, is computationally expensive, and yields inferior performance compared to modern transformer LLMs.

6) Conclusion
Fine-tuning a pretrained model is the best and most cost-effective choice. This matches the provided correct answer: Fine-tuning a pretrained model.

A medical team wants an LLM that understands radiology terminology but cannot afford to retrain a model from scratch. Which approach is most appropriate?单项选择题

类似问题

Question at position 10 What is transfer learning in neural networks? Using a pre-trained neural network to extract features from a new datasetUsing a pre-trained neural network as a starting point for training a new modelUsing a pre-trained neural network to generate new dataNone of the above

Transfer learning is invariably effective. eg. Irrespective of the amount of data, we can always rely on transfer learning.

When attempting to transfer learn for an image captioning task, we must use a source dataset for image captioning or visual question answering, since image classification by itself is not similar enough

Which of the following statements about transfer learning in CNNs are correct? (mark all that apply)

Which of the following statements is correct about pre-training and fine-tuning?

What is the key difference between pre-training and fine-tuning stages in transformer model development? Hint: Lec 19, Slide 50.

更多留学生实用工具

智能学习助手

风格化写作助手

论文查重助手

文献引用助手

课堂转译助手

课堂笔记助手

Quiz搜索助手

学校历年真题

智能刷题助手

智能匹配练习

希望你的学习变得更简单