Machine Learning Speaker :Chia-Shing Huang Advisor :Dr. Kai-Wei Ke 2016/01/14 1
Outline Machine learning Decision tree Artificial neural Network Conclusion 2
Machine Learning Definition Field of study that gives computers the ability to learn without being explicitly programmed - Arthur SamuelArthur Samuel A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E - Tom M. MitchellTom M. Mitchell 3
Simple Learning Flow 4
Method Supervised learning Unsupervised learning Semi-supervised learning Reinforcement learning 5
Decision Tree What time is it? Has homework? Has date? N YYN Play game or not? < 19:00 >19:00 false true false true A decision tree is a flowchart-like structure in which each internal node represents a "test" on an, each branch represents the outcome of the test and each leaf node represents a class label. The paths from root to leaf represents classification rules. 6
Classification and Regression Tree(CART) Number of branches = 2 (binary tree) Base hypothesis = optimal constant Binary/multiclass classification(0/1 error) : majority of {yn} (result) Regression(squared error) : average of {y n } (result) Termination criteria = until forced to terminate All y n the same All x n the same 7
decision stumps h(x) Data rate in total data 8
Simple Data Set One more example Let’s play online ! 9
Artificial Neural Network (ANN) Definition: Artificial neural networks (ANNs) are a family of models inspired by biological neural networks and are used to estimate or approximate functions that can depend on a large number of inputs and are generally unknown. 10
Single Neuron XnXn X1X1 X2X2 X3X3 X0X0 SUM Transform Function F output w0w0 w1w1 w2w2 w3w3...wn...wn X i = nonlinear information (input) W i = weight of data features Perceptron Algorithm 11
Hidden layer 12
XnXn X1X1 X2X w1w1 w2w2 g2g2 g1g1 +1 X 0 =
w1w1 w2w2 w3w3 wnwn XnXn X1X1 X2X2 X3X3 b g2g2 g1g1 gngn G a1a1 a2a2 a3a3...an...an Fee dforward Network Feedback Network How to get optimization? Use Gradient descent 14
Example :DDoS attack detection Distributed Denial of Service(DDos) attack: is an attempt to make an online service unavailable by overwhelming it with traffic from multiple sources. SYN flood UDP Flood ICMP Flood LAND attack 15
Example :DDoS attack detection(con’t) Training data CPU idle rate Memory usage Network packets inflows Network packet outflows Current number of system process Ideal target (normal =0 /attack = 1) 16
i j 17
18 Schematic Simulation Environment
19 Simulation Environment Hardware Standard
Artificial Neural Network Preferences Input = 5 neurons CPU idle rate Memory usage Network packets inflows Network packet outflows Current number of system process Hidden layer = 10 neurons Output = 1 neuron (true or false) Weight & threshold = random (0~1) 20
21
22
23
24
Conclusion - Decision tree Pros: Human-explainable, widely used in business/medical data analysis Simple Efficient in prediction and training Cons: Heuristic: mostly little theoretical explanations Confusing to beginners 25
Conclusion - Artificial Neural Network Pros: good to model the non-linear data with large number of input features Robustness & fault-tolerance Strong adaptability Cons: So many answers that can’t identify which is the best answer. are prone to overfitting requires greater computational resources 26
Reference df df NGBmW50rrmFTqf NGBmW50rrmFTqf gradient_4.html gradient_4.html Ashraf, J. and Latif, S., “Handling intrusion and DDoS attacks in Software Defined Networks using machine learning techniques” in National Software Engineering Conference (NSEC), 2014,pp 紀宏宜 、 張偉德 、 陳志榮, “ 應用類神經網路於阻斷式服務攻擊之預測 ” 網際網 路技術學刊, pp , 9: [ 民 97.04] 27
Thank you for listening Happy winter vacation & happy new year 28
w1w1 w2w2 w3w3 wnwn w0w0 XnXn X1X1 X2X2 X3X3 X0X0 g2g2 g1g1 g0g0 gngn G a0a0 a1a1 a2a2 a3a3...an...an 29
30