🤖

ML 工程师面试

模型 · 训练 · 推理 · ML 系统

6 科目 MLE / Applied Scientist / AI EngineerAdvanced

原创复习卡。ML 技术演进极快，模型架构以最新论文为准。

考试结构 · 点击右侧按钮生成该科目卡组

#01

模型基础

DNN / CNN / RNN / Transformer 核心

backpropagation intuitionvanishing / exploding gradientsbatch / layer normattention mechanism core ideaTransformer key-value-query+2

#02

训练工程

optimizer · scheduler · 分布式训练

Adam vs SGD trade-offslearning rate schedulingdata parallel vs model parallelgradient accumulationmixed precision training+2

#03

推理 & 部署

延迟 · 吞吐 · 量化 · 缓存

quantization (int8 / fp16)knowledge distillationmodel serving (TF Serving / Triton)batching vs streaming inferenceKV cache for LLMs+2

#04

特征工程 & 数据

特征 store · 漂移 · 标签泄露

categorical encoding methodsfeature scaling when and whymissing value handlinglabel leakage detectionfeature store concepts+2

#05

ML 系统设计

推荐 · 搜索 · 反欺诈 · 图像分类

news feed rankingcandidate generation vs rankingembedding-based retrievalfraud / abuse detection pipelineimage / video classification at scale+2

#06

行为题

权衡 · 模糊 · 影响

STAR frameworkmodel that didn't work as expectedconvincing stakeholders to change metricnegotiating compute / data resourcesexplaining ML uncertainty to leadership