if_Guo
Search
Search
Dark mode
Light mode
Explorer
Popular Tags
#interpretability
#LLM
#optimization
#weight-geometry
#agentic-RL
#neural-network
#implicit-regularization
#mathematics
#model-compression
#reparameterization
Home
❯
Notes
Folder: Notes
22 items under this folder.
Jul 20, 2026
权重初始化为何要小
Jul 20, 2026
最优传输OT
Jul 20, 2026
扩散模型连续时间变分目标
Jul 20, 2026
反向传播教学详解
Jul 20, 2026
切面
Jul 20, 2026
学习率衰减与权重平均的联系
optimization
learning-rate-schedule
weight-averaging
weight-geometry
Jul 20, 2026
参数空间与计算图：把神经网络拉成一个向量，到底省略了什么？
neural-network
optimization
weight-geometry
Jul 20, 2026
LLM 内部坐标的排列不变性
interpretability
model-compression
reparameterization
Jul 20, 2026
Jacobian、Hessian 与 SVD：神经网络训练中的敏感度、曲率与低秩结构
mathematics
neural-network
optimization
weight-geometry
Jul 20, 2026
为什么学习率不宜过小：有限步长的隐式梯度正则
optimization
implicit-regularization
weight-geometry
Jul 20, 2026
Superposition（叠加）
interpretability
Jul 20, 2026
Residual Stream（残差流）
interpretability
Jul 20, 2026
Post-hoc vs Intrinsic 可解释性
interpretability
Jul 20, 2026
Polysemanticity（多义性）
interpretability
Jul 20, 2026
Linear Representation Hypothesis（线性表示假说）
interpretability
Jul 20, 2026
Induction Head（归纳头）
interpretability
Jul 20, 2026
Gumbel-Softmax
Jul 20, 2026
Features 与 Dictionary Learning（特征与字典学习）
interpretability
Jul 20, 2026
Faithfulness（忠实性）
interpretability
Jul 20, 2026
Circuits（电路）
interpretability
Jul 20, 2026
Causal Mediation（因果中介分析）
interpretability
Jul 20, 2026
Activation Patching（激活patching / Causal Tracing）
interpretability