17.txt

This complete matlab for neural network

【 在 guitar (新宠酱牛肉) 的大作中提到: 】
: There are a number of algorithms on outlier-detection, e.g., a SIGMOD2001 pape
: r, But I don't know if they are good or not for your purpose. ;)
: 【 在 ihappy 的大作中提到: 】


finalgas (数据挖掘*为你而累) 于Wed Dec 11 09:33:15 2002提到：

    当然了，老板没有明说，也没有限定一定用什么方法。。
    老板的意思是开始的时候先看广一些，拓宽思路......
    因此想遍寻现在做这方面比较好的算法， 这样才好站在巨人
的肩膀上呀。^_^   当然最重要的一点就是：避免重复工作。。
    ^_^  现在比较多的是决策树和算法。的确改装不容易。
    在聚类分析的方法里提到了索引技术，以及Guha证明了的有效取
样技术。我想应该都可以用在海量数据挖掘里。
    至于关联，时序等，目前我没有找到海量数据分析相关的。^_^
    ^_^

【 在 fpzh (fpzh) 的大作中提到: 】
: 你要做哪方面挖掘？关联，分类，聚类，还是其他？
: 其他的算法当然有，决策树其实有不少问题的，比如过学习（Overfitting）。决策树的
: 属性选择和剪枝策略各自都有十几种了，正向GzLi以前说的，大的改进是不太好做了。
: 自己感觉顶多是在大规模数据处理方面，在现有的SLIQ等算法的基础上，能否再有所改
: 进？
: 当前分类和回归比较好的理论是SVM，但关键是在海量数据处理方面再作出大的改进，比
: 如可以处理几十万甚至几百万的数据（估计不太容易，谁能提出新的训练算法就厉害了
: ，呵呵，所以说工程上还是用成熟的算法比较保险）
: 【 在 finalgas (数据挖掘*为你而累) 的大作中提到: 】
: (以下引言省略 ... ...)


finalgas (数据挖掘*为你而累) 于Wed Dec 11 09:41:06 2002提到：

     Yes,It does.
     :)   Maybe what guitar said can answer the
latter question.  ^_^

【 在 ihappy (seagullx~壮起鼠胆把猫打倒) 的大作中提到: 】
: Does IDS mean Intrusion Detection System?
: I never did DM and IDS, but I think that IDS faces a skew-distribution prob..
: i.e. in IDS, normal operation data are huge, however, attacks are extremely 
: rare. Is there any good algorithm to deal with this?
: 【 在 finalgas (数据挖掘*为你而累) 的大作中提到: 】
: (以下引言省略 ... ...)


ihappy (seagullx~壮起鼠胆把猫打倒) 于Wed Dec 11 09:44:45 2002提到：

ok, it's intrusion detection system, not intelligence decision support.

Now I have a few more words.

I do not do anything in this field. But I heard a talk of a faculty member
at my college, he is doing IDS, and has got 3 Best paper awards of best paper
runnerup award of SIGKDD. In his talk, he said that there is no effective way
to deal with the problem I mentioned. And the talk was given after SIGMOD2001.

Maybe I am wrong,:-) I am not doing IDS. 

【 在 finalgas (数据挖掘*为你而累) 的大作中提到: 】
:      Yes,It does.
:      :)   Maybe what guitar said can answer the
: latter question.  ^_^
: 【 在 ihappy (seagullx~壮起鼠胆把猫打倒) 的大作中提到: 】


ihappy (seagullx~壮起鼠胆把猫打倒) 于Wed Dec 11 09:46:28 2002提到：

and i have a look back into guitar's poster. 

What I am saying are basically data preparation/data acquisition, not outlier 
detection.

Outlier detection is another problem.

【 在 ihappy (seagullx~壮起鼠胆把猫打倒) 的大作中提到: 】
: ok, it's intrusion detection system, not intelligence decision support.
: Now I have a few more words.
: I do not do anything in this field. But I heard a talk of a faculty member
: at my college, he is doing IDS, and has got 3 Best paper awards of best paper
: runnerup award of SIGKDD. In his talk, he said that there is no effective way
: to deal with the problem I mentioned. And the talk was given after SIGMOD2001.
: Maybe I am wrong,:-) I am not doing IDS. 
: 【 在 finalgas (数据挖掘*为你而累) 的大作中提到: 】


finalgas (数据挖掘*为你而累) 于Wed Dec 11 09:48:26 2002提到：

    Maybe.
    First , Thanks for you opinion.
    But in some case the attack data is almost the same
as the normal data except one or two attributes. Then, Is the
algorithms on outlier-detection can work well .

【 在 guitar (新宠酱牛肉) 的大作中提到: 】
: There are a number of algorithms on outlier-detection, e.g., a SIGMOD2001 pape
: r, But I don't know if they are good or not for your purpose. ;)
: 【 在 ihappy 的大作中提到: 】


finalgas (数据挖掘*为你而累) 于Wed Dec 11 10:02:03 2002提到：

    Yes,
    Just like all of things in our world,we must find a
balance on this question,especially in apllication.
    If we use outlier-detection , there will be many
attack data who disguise well  we can not detect. and
I don't know It does well in studying adding-knowledge
or not. But if  we use other way , the cost will be
more high.
    用中文说：也就是一个误报与漏报的平衡。 ^_^

【 在 ihappy (seagullx~壮起鼠胆把猫打倒) 的大作中提到: 】
: ok, it's intrusion detection system, not intelligence decision support.
: Now I have a few more words.
: I do not do anything in this field. But I heard a talk of a faculty member
: at my college, he is doing IDS, and has got 3 Best paper awards of best paper
: runnerup award of SIGKDD. In his talk, he said that there is no effective way
: to deal with the problem I mentioned. And the talk was given after SIGMOD2001.
: Maybe I am wrong,:-) I am not doing IDS. 
: 【 在 finalgas (数据挖掘*为你而累) 的大作中提到: 】


juggernaut (万念俱灰) 于Wed Dec 11 11:54:28 2002提到：

偶觉的没有这么简单吧，要是这样那检测岂不是非常容易？应该还是模式上不一样吧。
【 在 finalgas (数据挖掘*为你而累) 的大作中提到: 】
:     Maybe.
:     First , Thanks for you opinion.
:     But in some case the attack data is almost the same
: as the normal data except one or two attributes. Then, Is the
: algorithms on outlier-detection can work well .
: 【 在 guitar (新宠酱牛肉) 的大作中提到: 】


minerboy (miner) 于Wed Dec 11 13:51:46 2002)
提到：

我感兴趣，可以一起合作一下吗？

【 在 helloboy 的大作中提到: 】
: IDS---Intelligent Desicion Support
: 我想，通过算法，在实际工程中进行检验，可以提高理论的实际应用性，对于将来找..
: 有用。在实践中不要忘记用理论来充实，在研究理论时不要脱离实际。
: 数据挖掘包括很多方面，基本的有分类、聚类、关联规则等。
: 每方面都有很多算法，在分类上包括了范例推理、决策树、神经网络、bayes等。各