AI Course
Transformed from Logseq page
AI Course
-
考试内容
-
1.第一章:概述(掌握基本常识)(1.1~1.6) 2.第二章:搜索求解问题(2.1基本思路重点掌握,2.22.32.4了解当中内容) 3.第三章:搜索策略,重点掌握(3.1盲目搜索和3.2启发式搜索重点掌握)(3.2.13.2.2,重点掌握爬山法)(3.3重点掌握,随 机搜索?) 4.第四章:图搜索(4.1重点掌握,涉及A算法和A*,4.2重点掌握4.2.1
4.2.4) 5.第五章:重点掌握博奔搜索(minmax,alphabeta,montecarlo) 6.第六章:遗传算法和免疫算法(6.1~6.10重点掌握,6.8一般了解即可,6.116.14重点掌握,6.15一般性了解) 7.第七章:群智能算法(7.2.1~7.2.3重点掌握,7.3一般了解) 第八九章跳过 知识表示方法&推理方法 10.第十章:基本知识表示方法(10.110.310.410.5,以及状态空间法,问题归约法,知识图谱) 11.第十一章:推理(11.111.2..的归结原理以及应用) 12.第十二章:不确定性推理(12.212.3,12.2证据理论,12.3主观bayes重点掌握(nmd),以及概率推理&可信度,但是 模糊推理不考??) 机器学习,人工神经网络,深度学习和强化学习 14.第十四章:机器学习(机器学习方法&机器学习算法,后者不在教科书范围内,但是ppt上有,例如svm,回归,决策树等 算法) 15.第十五章:人工神经网络(15.115.5要非常熟悉,15.6一般了解即可)(至少了解两个:15.7.1感知机,15.7.2BP算法 ??.深度学习&强化学习:至少掌握CNN,RNN,强化学习网络的基本工作原理,掌握基本常识
-
-
Search
-
Reasoning
-
知识表示 {{embed (((635f695e-82d9-44a9-b5f5-25fe3f28f030)) }}
-
-
Uncertain Reasoning
-
主观Bayes方法
-
Given
-
-
分类讨论:
- others: 由上述三个点线性插值
-
-
可信度方法
-
-
Dempster-Shafer理论
-
基本概率分配函数 置信函数 置信区间 其中不怀疑A的度量为
-
证据的组合
-
-
模糊推理
-
-
ML
-
classify by strategy
-
rote: memorize straight forward
-
instruction: induct generic rules from typical instances
-
explanation(EBL):
-
ML by analogy:
-
-
-
ANN
-
event camera事件相机vs场相机、脉冲神经网络
-
交感神经:Your** sympathetic nervous system controls your “fight-or-flight” response.** Danger or stress activates your sympathetic nervous system, which can cause several things to happen in your body.
-
小脑模型关节控制CMAC、互联前向网络(同层互联)、广泛互联网络(玻尔兹曼机)
-
有监督:广义Delta规则、LVQ算法
-
无监督:按相似特征分组聚集(如分组方差),Kohonen算法、Carpenter-Grossberg
-
-
RL
-
MDP马尔可夫过程 马尔可夫性质:下一个状态的条件概率分布仅取决于当前状态 bellman方程:关注当前reward与未来reward
-
Q-Learning
- initialize
- while not end
- explore or exploit(decide by random)
- if explore, choose max Q-value
- update new Q-value
-
特点、对比搜索
-
训练过程?无初始全局预估
-
Linked References (6)
ICLR: International Conference on Learning Representatives, originated from end-to-end idea #AI Course
Linear Algebra suggested course Gilbert Strang@MIT ↗ #AI Course
-
ML
-
cross table
-
meta-param search
-
K-fold validation
-
cross validation
-
-
Knowledge Representation
-
first-order logic
-
multi-var predicate=>can be expressed by bi-var predicate and currying
-
-
semantic web
-
general links
-
instance, isa
-
generic, a-kind-of(AKO)
-
compose, part-of
-
attribute, others
-
-
multi-var semantic web: convert to bi-var semantic web
-
predicates are in “and” relation with each other by default, or relations are marked by dotted box
-
蕴含关系ANTE(antecedent)/CONSE(consequence)
-
量化
-
e.g. draws/2022-11-07-16-20-00.excalidraw 武汉大学是一所具有百年历史的综合性大学
-
-
-
Knowledge Representation
-
knowledge
-
relation among replicable info
-
data->info->knowledge->intelligence
-
facts, rules, control, meta
-
-
hints
-
easy to aggregate, search, infer
-
-
state space
-
state def and state trans
-
-
framework for KR
-