Subordinating conjunctions

Cause or  Effect	Concession	Condition	Comparison  or Contrast	Purpose	Space or Time
as  because  since	though  although  even though  even if	if  since  unless  when  whenever	while  as  rather than	in order that  so that	before  since  once  after  while  when  until

Adverbial conjunctions

Addition	Emphasis	Comparison or Contrast	Cause or Effect	Time
in addition furthermore moreover further	in fact indeed	however nevertheless nonetheless otherwise in contrast in comparison	as a result consequently hence therefore thus	finally meanwhile next

PREPOSITIONS

about along behind beyond during inside off over to up

above among below but except into on since toward upon

across around beneath by for like onto through under with

after at beside despite from near out throughout underneath within

against before between down in of outside till until without

梯度下降 for NN (1)

梯度下降的 6 种变形：

• Back propagation
• delta-bar-delta (adaptive learning rate)
• steepest descent
• QuickProp
• Gauss-Newton
• Levenberg-Marquardt

梯度下降 for NN (2)

1. find the slope of the error surface
2. how much to go down the error surface
3. how the weights should be adjusted after each iteration
4. when to stop training and what governs the decision
5. whether the solution is suboptimal
6. how good the final model is and how to assess its goodness of fit

《群体智能与仿生计算》阅读笔记（1）

偶然从图书馆看到这本书。还没有深入读，先把讲授的主要算法列出：

• 演化算法 Evolutionary Algorithm
• 人工免疫算法 Artificial Immune Algorithm
• 混合遗传算法、遗传局部优化 Memetic Algorithm
• 粒子群算法 Particle Swarm Optimization
• 混合蛙跳算法 Shuffled Frog Leaping Algorithm
• 猫群算法 Cat Swarm Optimization
• 细菌觅食算法 Bacterial Foraging Optimization
• 蚁群算法 Ant Colony Algorithm
• 人工蜂群算法 Artificial Bee Colony Algorithm

GPU存储器结构的演变

《大规模并行编程实战》 12.2节 读书笔记

• 大型虚拟和物理地址空间
• 统一的设备地址空间
• 可配置的缓存和暂存空间
• 提高原子操作的速度
• 提高全局存储器的访问速度

Reference

Gelado, I. et al, An asymmetric distributed shared memory model for heterogeneous parallel systems.