Download presentation
Presentation is loading. Please wait.
Published byTheodore Tate Modified over 9 years ago
1
Artificial Intelligence Final Project Text document Classification with new type Rule-based PLM Chang, Jung Woo Shin, Dong In Jung, Hyun Joon School of Computer Science and Engineering Seoul National University Presented by Jung Hyun Joon 2004. 12. 21
2
Artificial Intelligence Final Project Contents Introduction Architecture Wet design scheme Performance Evaluation Conclusion References
3
Artificial Intelligence Final Project Introduction Classification Problem –Decision Tree & Version space learning.. –Some shortcomings Not include all possible rule sets, only focus part.. Vulnerable to noisy data In this paper –Utilize massive parallelism of DNA computing –Define rules as a element with 1 / 0 / don’t care –Make noise-tolerant classification system
4
Artificial Intelligence Final Project Architecture Rule-based PLM –training and test Rule-based PLM 의 전체적인 구조
5
Artificial Intelligence Final Project Model structure Property of target data Target Data and Model Structure 1010 Can be involved in class A, B or C Document i Class Tag + n digit binary bit + history count in Training 1010A13 1010B22 1010C07
6
Artificial Intelligence Final Project Training … … 1010A Training query 0000A0 0000B 01010A01010B01010C0****C0 … … 0000A0 0000B 0 1010A1 1010B0 1010C0 ****C0 1010B … … 0000A0 0000B 0 1010A1 1010B1 1010C0 ****C0 0000B … … 0000A0 0000B 1 1010A1 1010B1 1010C0 ****C0 1010A … … 0000A0 0000B 1 1010A2 1010B1 1010C0 ****C0
7
Artificial Intelligence Final Project Test 0000A 0000B 1010A 1010B 1010C ****C … … 4 3 12 2 3 98 1010A Test query Class A – 12 / 17 Class B – 2 / 17 Class C – 3 / 17 Class A
8
Artificial Intelligence Final Project Wet-design Scheme Initial DNA strand 생성 과정
9
Artificial Intelligence Final Project Wet design Scheme training example set 생성 과정
10
Artificial Intelligence Final Project Wet design Scheme classification 과정
11
Artificial Intelligence Final Project Forward and Backward scheme to untrained query Comparison of the forward and backward model scheme
12
Artificial Intelligence Final Project Performance Evaluation.
13
Artificial Intelligence Final Project Performance Evaluation Average Classification Success RateCISI classification Success Rate
14
Artificial Intelligence Final Project Performance Evaluation CRAN Classification Success RateMED Classification Success Rate Cause of MED classification success rate 1.Preprocessing ( all zero term document delete ) 2.Sparse vector of term
15
Artificial Intelligence Final Project Conclusion Present new type rule-based PLM –Support the flexibility with don’t care property –Forward and backward search scheme to untrained query –Showing the similar performances compared with WEKA –Possibility of wet-design
16
Artificial Intelligence Final Project References Version Space Learning with DNA Molecules, Lim, H.-W. et al, LNCS, vol. 2568, pp. 143-155, 2003 DNA computing on surfaces, Liu et al., Nature, 2000 A Bayesian Algorithm for In Vitro Molecular Evolution of Pattern Classifiers, Zhang, B.-T. and Jang, H.-Y., Preliminary Proceedings of the Tenth International Meeting on DNA Computing, pp. 294-303, 2004 10 more papers and many web-sites
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.