Re: data mining, computer security

[Vitaly Osipov <vitaly.osipov () gmail com>]

This is a cool-sounding idea - data mining hostile data. I don't think there
has been much research in this area - all well-known algorithms assume a
mountain of inert data that they try extracting knowledge from. At the same
time, most of these algorithms have well-researched problems and it is
usually possible to come up with a data set that will confuse any given
algorithm.

It will be really interesting to see some research on how a small
contamination can be introduced in various data set to render a (one or
several) data mining algorithm inefficient. Then, how to counteract this
contamination by using multiple algorithms, all kinds of aggregation,
bagging and so on. Pity I am not a postdoc...

On Thu, May 6, 2010 at 3:50 PM, Josh Saxe <joshsaxe () yahoo com> wrote:

> Hi --
>
> Here's a thought I thought might get an interesting response on this list.
>
> I've been thinking: cyber-security, narrowly constructed is focused on
> protecting IT resources from being compromised or disabled.  Perhaps search
> and classification problems are also an arena for cybersecurity, goals being
> obfuscating patterns in a dataset, or finding them.
>
> Anyone who wants to avoid classification by a facial recognition algorithm
> should know how that algorithm works, and use that knowledge to evade
> detection (is there a shape of fake mustache that optimally throws off a
> given face classifier? =).  On the other hand the designers of the facial
> recognition algorithms need to learn from the 'evaders' and respond with
> more sophisticated algorithms.  This isn't dissimilar from the cat and mouse
> game played between exploit developers and application security whitehats.
>  And as the size of the world's databases, and the 'instrumentation' of
> human activity grows, I wonder if this search / evasion problem won't become
> more and more important.
>
> There are definitely a lot of arenas where this is important.  Credit card
> fraudsters can strengthen their positions by understanding the anomaly
> detection algorithms used by the credit card companies and modifying their
> purchasing behaviors to elide their detection.  Terrorists building training
> camps need to understand image analysis algorithms that government agencies
> run on satellite photos.  Of course, spammers have already been playing this
> game with text classification algorithms, 'link-spammers' have played this
> game with web-graph analysis algorithms, etc, so this isn't a new idea, but
> one that I think is likely to become more important.
>
> Josh
>
>
> _______________________________________________
> Dailydave mailing list
> Dailydave () lists immunitysec com
> http://lists.immunitysec.com/mailman/listinfo/dailydave
_______________________________________________
Dailydave mailing list
Dailydave () lists immunitysec com
http://lists.immunitysec.com/mailman/listinfo/dailydave