EssayCN
激活函数：神经网络为什么需要非线性
从 XOR 与线性变换的边界出发，理解激活函数怎样改变表示空间，并比较 Sigmoid、Tanh、ReLU 与 GELU 的输出、梯度和使用位置。
周磊
Rejoice not in possessions,
grieve not over yourself
Work email
Academic email
Links
About me
MoreThe University of Tokyo GSII / III · M.S. Student Tokyo
I build AI-native applications, developer tools, and human-centered systems that connect research, interaction, and real-world implementation.
My current question: how can AI become part of human memory, work, research, and everyday thinking?
Currently working on
Research
Human-centered AI, personal systems, and interaction-aware data that can become useful product workflows.
Engineering
Full-stack prototypes, local AI workflows, and tools that move quickly from idea to real use.
Minerei / Kotolog
An AI-native personal system that brings together tasks, logs, and long-term memory.
Currently using
Not a complete stack list, but the tools I keep reaching for while connecting research, product work, and personal systems.
Tech stack
Latest Activities
Essays, notes, and devlogs in one stream
- EssayCN
不只是聊天：从 OpenAI API 看现代模型能做什么
从公开 API 的输入、推理、结构化输出、工具、状态、实时与异步接口出发，重新理解现代模型的能力，以及模型与平台之间的边界。
- EssayCN
从逻辑回归到神经元：机器究竟学到了什么
从规则、数据与参数的边界出发，沿人工神经元的历史走到一个可训练的 Sigmoid 神经元。
- EssayCN
从命令到界面
从批处理、分时系统、命令行、图形界面、Web 到移动设备，追踪人与计算机如何从提交任务走向连续操作。
- EssayCN
从操作到解释
当自然语言成为计算入口，人与系统的关系如何从调用预设功能，转向对表达、意图与共同任务的持续解释。
- EssayCN
交互的起源
从“交互”、interaction 与日语中的「交互」「相互作用」「インタラクション」出发，追溯交互如何从彼此作用生长为一门设计学科。