机器学习原理
What is machine learning and how does it differ from traditional programming?
原题What are different types of machine learning techniques?
原题What is the difference between supervised and unsupervised learning?
原题What is semi-supervised learning?
原题What are stages of building machine learning models?
原题Can you explain the bias-variance trade-off in machine learning?
原题What is overfitting and how do you prevent it?
原题Why and how do you split data into train, test, and validation set?
原题What is cross-validation and why is it important?
原题Can you explain the concept of regularization and its types (L1, L2, etc.)?
原题How Do You Handle Missing or Corrupted Data in a Dataset
原题What is a decision tree and how does it work?
原题Can you explain logistic regression?
原题Can you explain the K-Nearest Neighbors (KNN) algorithm?
原题Compare K-means and KNN algorithms.
原题Explain decision-tree based algorithms (random forest, GBDT)
原题What is gradient descent and how does it work?
原题Can you explain the support vector machine (SVM) algorithm? what is Kernel SVM?
原题Can you explain neural networks and how they work?
原题What is deep learning and how does it differ from traditional machine learning?
原题Can you explain the backpropagation algorithm and its role in training neural networks?
原题What is a convolutional neural network (CNN) and how does it work?
原题What is transfer learning and how is it used in practice?
原题Walk through a transformer block. Why divide attention scores by √dₖ?
原题What is the KV cache and why does it matter for inference? How does its memory scale?
原题Compare MHA, MQA, and GQA. Why did Llama adopt GQA?
原题What is RoPE and why are rotary embeddings preferred over learned positional embeddings?
原题When would you choose RAG vs fine-tuning vs long-context?
原题Compare SFT, DPO, GRPO, and RLVR — data requirements, what extra models each needs, and when to use each.
原题Why does GRPO drop the critic network, and how does it estimate advantage?
原题What is LoRA / QLoRA and when would you use PEFT over full fine-tuning?
原题Explain quantization (INT8/INT4, FP8) and KV-cache quantization. What's the accuracy/latency tradeoff?
原题How does speculative decoding speed up generation?
原题How do you evaluate an LLM / RAG system? What is the RAG triad and LLM-as-judge?
原题How does a VLM combine a vision encoder with an LLM? (e.g., LLaVA / PaliGemma)
原题What is a VLA model? Contrast discrete action tokens (RT-2, OpenVLA) vs flow-matching action heads (π0).
原题Diffusion vs autoregressive generation — how do they differ and when is each used (image/video/audio/text)?
原题What is classifier-free guidance?
What is flow matching / rectified flow?
原题完整原题
What is classifier-free guidance? What is flow matching / rectified flow?
What is a Mixture-of-Experts model? Distinguish active vs total parameters.
原题how do you choose among these? (imbalanced datasets)
原题“these”承接上级提纲中的 Accuracy、Precision、Recall/Sensitivity、Specificity 和 F-score；场景限定为类别不平衡的数据集。
k-fold cross validation (what's a good k value?)
原题how to apply dropout to LSTM?
原题
暂无题目