Classical AI Models

导言

机器学习和人工智能模型算法,从一开始模仿神经元设计,到现在根据任务定制或者基于naive的思想构建(例如对抗思想、感受野、注意力机制)。模型的设计可以说是日新月异,截然不同。但是从高性能计算的角度来看,还是离不开求导操作、矩阵操作、激活函数计算这几点。剩下值得考虑的就是寻找现有或者未来模型构成计算操作的最大公约数,来对其进行特殊软硬件设计加速。或者只是对现有模型的适配加速工作。

Read more

HTML

导言

learning HTML is the foundation of building my hugo theme

Read more

Salary & Tax & Insurance

导言

毕竟都是要工作的人了,有些💴相关的事情还是要了解的。 计算税率有几个难点:

  1. 计算每月累计的,和全年计算方法的不同。
  2. 预交款和退税怎么算?6万是5000*12. 但是预交款不扣吗
Read more

When & How 4 team presentation page & knowledge database pool

导言

我原本是想开发一个实验室主页的模板的,一开始想着还可以带到公司里去。但是我突然想到 公司里还有 “嫡系” 这种事情。让我意识到实验室和公司的环境完全不一样,让机制的部署和落地变得更加复杂。

Read more

Data Link

导言

There are some new concepts in data link needed to learn

Read more

Data Structure Summary

导言

ASCYLIB (with OPTIK) is a concurrent data-structure library. It contains over 40 implementations of linked lists, hash tables, skip lists, binary search trees (BSTs), queues, priority queues, and stacks. ASCYLIB contains sequential, lock-based, and lock-free implementations for each data structure.

Read more

AddressTranslationForPIM

导言

Accumulation is key in all endeavors, and xxx is no exception.

 1. Motivation-Action-Result: The introduction of a blog post should serve as a 'prelude' where I explain the source of inspiration and the core **reason** behind my writing of the new post. It should also briefly **outline** the initial idea or solution to achieve a particular **goal** or address a specific problem.

 2. Knowledge Organization: Expanding, organizing, and summarizing the encountered information.
Read more