Sampel Project proposal by chidin


									1. Research problem

As mentioned in the background, wireless networks are particularly
vulnerable to traffic-analysis attacks that allow observers to obtain
knowledge about communications occurring over the network, without
necessarily being able to obtain the content of the communication. The
known methods for countering such attacks often degrade the network’s
performance, sometimes to an extent that disables certain network
applications. The proposed project will investigate improved methods for
defeating certain types of traffic-analysis attacks. Such methods, if found to
deliver satisfactory performance and security, could potentially be
implemented in both wired and wireless network routers, or in low-level
networking code for operating systems. The project is technically complex
due to the large number of potentially informative statistical properties in
network traffic, as well the many types of attacks that can be conducted.

2. Research methodology

The research methodology will be based on machine learning (ML)
techniques, which are computer algorithms that can automatically generate
decision-making behavior from a set of data. One example of a problem that
is often solved using machine learning techniques is a classification task,
where an agent is presented with a set of data items and asked to classify
each item into one of several categories. The traffic classification problem
considered by the work mentioned in the background document is a task of
this type. Instead of pre-programming the classification rule into the agent,
we allow the agent to automatically generate the rule from a set of already-
categorized data items, which serve as examples for the agent to learn from.
Another common machine learning problem is sequential decision making,
where an agent is periodically asked to decide among several actions, with
each action incurring a numerical “reward” or “cost” and modifying the state
of the agent’s environment.

We will use ML techniques to identify relevant properties and patterns
among the different characteristics of network traffic (packet sizes, inter-
arrival times, etc.), as well as to relate these to relevant characteristics of the
communication itself (application being used, etc.). The information found
by these techniques is then used by a “traffic modification” agent that alters
the traffic to strengthen it against analysis attacks, while still maintaining an
acceptable level of performance. This agent will likely use learning
techniques, such as reinforcement learning, in making its decisions.

3. Novelty of the proposed approach

This approach is novel in two main ways. First, the defense system is
“informed” in that it attempts to predict possible analysis attacks and take a
situation-specific action against them. Second, the defense system is
“intelligent” in that it deliberately tries to control the tradeoff between
performance and security.

4. Nominee’s role in the project

(To be completed)
5. Project participants and their roles

(To be completed)

6. Physical resources

(To be completed)

To top