Intelligent computing system based on
pattern recognition and data mining algorithms
Abstract
application and technology
the development direction of the intelligent system
the basic of data mining
the advantages and disadvantages
expert system
intelligent system integrations
learning theory
Artificial intelligence
Machine learning
Data mining technology
statistical analysis
fuzzy logic
pattern recognition
Artificial neural networks
the structure of the general algorithm of data mining
Introduction
第二段
第三段
the dynamic of problem
don't need rely on gradient information
based on population probability search
methods
genetic algorithm
particle swarm algorithm
ant colony algorithm
differential evolution algorithm
cultural algorithm
don't need to sent the initial points
not fall into local extremum
Artificial intelligence
a branch of computer science
applied fields
the point of view of its functions
pattern recognition
neural network
expert system
a comprehensive fringe science
control theory
information technology
calculations
linguistics
...
the cognitive system
deductive reasoning
by symbolic processing
the theory of intelligent computing
mainly based on connectionism
fuzzy mathematics
iterated function system
developing directions
Artificial neural networks
genetic algorithms
evolutionary computation
artificial life
ecological computing
immune information processing
multi-agent systems
etc.
"Collaboration"
distributed artificial intelligence
main research
how to coordinate their knowledge, skills and planning
how to solve multi-objective problem with the effective way
to support collaborative work of large and complex intelligent system or computer design
Chapter 2
intelligent computing system
First paragraph
the several basic techniques
the trends
integration of intelligent technology and expert system
fuzzy logic and neural networks
Electric power system
decision trees
fuzzy systems
collective methods
cloud theory
statistical methods
self-organizing mining technology
the imitation biotechnology
features
self-organizing feature mapping network method
group intelligent
optimized
evolution strategy
evolution objects
unsupervised clustering method
group intelligence algorithm to cluster,
to reduce the knowledge of the data vector group randomly into a two-dimensional plane
solving complex dataset
Search engines
Video monitoring system
Knowledge processing methods
collaborative intelligent computing
main application
the diagnosis of power equipment fault
two core problem
information extraction
uncertainly of information increases
methods
neural network based fault diagnosis
fault diagnosis based on optimization technology
expert system based fault diagnosis
if the information is incomplete
prone to errors
characteristics
nonlinear characteristics
parallel processing ability
robustness
self-organizing learning
for small and medium power system failures is good
genetic algorithm
the problem get into an integer programming problem
understand the focal point of the user
customize web page
motion tracking technology
automatic video retrieval technology
moving object technology
pattern analysis technology
Chapter 3
First paragraph
Recognition behavior
specific things
abstract things
Pattern recognition system
consists
data processing
feature extraction and selection
classification decision making
data acquisition
transformation data from sensors to computer
eliminate the noise
filtering algorithms
effective features
be careful the curse of dimensionality
Pattern recognition method
neural networks
Artificial neural network based pattern recognition
Template matching pattern recognition
Four basic models of fuzzy pattern recognition
Proximity principle
Proximity principle/Synthetic fuzzy set
Maximum membership principle
Synthetic fuzzy set/Synthetic approach degree
A number of identification methods
not developed a unified and effective pattern
neural network
allow pattern to be noisy
EX: BP neural network
Table 2
purpose
develop general data analysis techniques
not rely on application domains
current work
figure out a combination of the specific problems
propose new methods of pattern recognition
it is very difficult to achieve in practice.
the computation of template matching is very large.
mainly used in
tracking of moving objects
detection of object position in images
based on support vector machines
solving
construct an optimal hyper-plane
to maximize the distance
hyper plane
different samples
pattern recognition
function estimation problems
Synergetic pattern recognition
a high-dimensional nonlinear problem
reduced to
a nonlinear equation with the same set of dimensions.
mainly to establish stochastic differential equations.
used to eliminate the stable modulus
to obtain the closed equation of order parameter.
the high-dimensional problem
reduced to
a low dimensional problem .
Compared with the traditional pattern recognition system
reduces the process of feature extraction and selection.
Data mining algorithms
Bayesian algorithm
Timing algorithm
the main methods
Clustering algorithm
Association algorithm
Neural Network algorithm
Regression algorithm
Basic concepts of association rules
A mining algorithm of association rules
Data set partitioning algorithm
Incremental updating algorithm
Parallel mining algorithm
the actual development often turns into formalization of domain knowledge and discovery tasks
Model discovery is a cyclic heuristic process
Decision tree
earliest model
ID3
bigger dataset
IBLE method
Neural network
MP model and the Hebb learning rule
propagation model and functional network
used in
prediction
pattern recognition
Concept tree method
Fuzzy set method
Visualization technology
a classification algorithm
requires a small amount of computation
for predictive
discover the relation between the input column and the predictable column
used for initial data detection
a regression algorithm
product sales in a prediction scheme
to predict continuous columns,
a segmentation algorithm
identifying exceptions
Scatter diagrams
describe how these items are grouped.
依照購客已買的東西預測未來會買的東西
在多層感知器網絡中
每個神經元接收一個或多個輸入並產生一個或多個相同的輸出。每個輸出是神經元輸入總和的簡單非線性函數。
輸入值從輸入層中的節點傳遞到隱藏層中的節點，最後傳遞到輸出層。
輸入圖層，可選的隱藏圖層和輸出圖層
a variant of the decision tree algorithm
completeness
mining frequent itemsets
maximal frequent itemsets
closed frequent itemsets
constrained frequent itemsets
association rule
single layer rule
multi-level associations rule
the types of values
Boolean association rules
quantitative association rules.
the types of mining rules
association rules
related rules mining
mining pattern types
sequential pattern mining
structural pattern mining
frequent itemsets mining
mining constraints
knowledge type constraints
data constraints,
interestingness constraints
rules constraints
the mining algorithms of association rules
In depth first algorithms
breadth first algorithm
data set partitioning algorithm
depth first algorithm
sampling algorithm
Incremental updating algorithm
and so on
hierarchical algorithm
AprioriTid
AprioriHy-brid algorithms
Apriori
DHP
uses hash table technology
no longer scanning the entire database
using the initial scan of the database
FP growth algorithm
OP algorithm
does not need to generate a large set of candidate items
Tree Projection algorithm
但其應用難點在於，在處理大型稀疏數據庫時，挖掘，處理和遞歸計算需要相當大的空間。
Partition算法
DIC算法
包括D.W.Cheung提出的FUP算法，IUA，PIUA，IUAR算法等
FUP算法是Apriori算法的改進，也是解決增量更新問題的經典算法。
FUP算法主要針對在最小支持度和最小置信度保持不變的情況下修改數據庫時如何生成更新數據庫的關聯規則。
它使用挖掘過程獲得的頻繁項集信息，避免重複計算頻繁項集支持數的時間成本，提高算法的效率。
Agrawal等人提出的CD，DD和CaD算法。
Park等人提出的PDM算法。
並行算法使用一組同時進程，交互和協調來完成給定問題的解決方案。
Cheung等人提出的DMA和FDM算法
但它的缺點是流量和候選頻繁項目集都相對較大。
CaD算法試圖通過劃分數據庫和候選集來減少處理器之間的數據依賴性，每個處理器可以獨立計算.
DD算法的缺點，然而由於低效率通信負載的移動解決方案的數據，處理器之間的大交互模式容易導致處理器是 處於空閒狀態，每個事務記錄基於多個哈希樹處理導致冗餘計算.
PDM算法類似 對於CD算法，所有處理器都具有相同的哈希表和候選集。
結論
智能係統集成主要包括人工智能，計算智能方法和其他智能技術。介紹了幾種智能係統集成的應用和技術，以及學習理論和專家系統的優缺點，以及神經網絡在智能係統中的應用。在求解智能計算時，不需要解決問題，或者不需要係統的梯度信息，因此可以獨立處理連續和離散問題。智能計算方法可以較好地解決不同優化問題的全局優化問題的最優解，並且可以利用簡單易懂的啟發式邏輯計算方法簡單地引入智能計算方法。詳細討論了數據挖掘中關聯規則的挖掘，並根據統計數據對一些常用的挖掘算法進行了分析，比較和總結。現有的改進算法無法滿足人們對採礦系統快速，及時響應的需求。因此，我們需要提高采礦過程的效率，並與用戶互動以產生視覺效果。同時，我們介紹了模式識別的各種技術和模型。實例表明，基於模式識別和數據挖掘算法的智能計算系統具有較高的效率和識別率。