Senior research ideas
If you are looking for a senior research project, consider the following ideas, or talk to me to brainstorm new ones.
Pentesting/intrusion/exploitation plan recognition
Suppose Dr. Plante is teaching the computer security course and each student has his/her own virtual machine on the delenn server. Each student is attempting to find vulnerabilities in the VM’s services. Suppose further that Dr. Plante is able to monitor and record all network traffic to/from each student’s VM. Is it possible to recognize the kinds of pentesting or intrusions or exploitation techniques that the student is using? Can we understand how students find successful vulnerabilities or identify reasons they fail to do so? With such a recognition engine, we could build an automated tutoring system that guides students on the right path.
Some existing research exists for (separately) intrusion detection from network traffic signatures and plan recognition. By combining the two subfields of AI and computer security, we could produce a useful tool for computer security educators.
See the project idea below, “Automated configuration of computer security scenarios,” for a companion project.
American Sign Language tutor
Is it possible to build a computer-based sign language recognition system that uses only a computer or smartphone’s built-in camera? This project was the subject of earlier senior research. The next task in this project is to find a way to use deep learning or another technique to train an accurate recognizer using a set of training images. Assuming that works, a usable application could be built and possibly fashioned into a computer-based tutoring system for ASL learners. Current ASL tutoring systems require special gloves or other sensors. The benefit with the proposed system is that it would be low-cost and usable by anyone with a laptop or smartphone, though this constraint brings significant challenges in accurately recognizing signs.
Recommendations based on contextual proximity
This idea was contributed by Jeffrey Wray of Mobilozophy (firstname.lastname@example.org), CSCI alumnus from 2009. Using a grid of inexpensive Bluetooth LE beacons in a large space, can we identify and analyze the travel paths and other behaviors of people and make individualized recommendations? E.g., can we recommend products and services to customers wandering around a store or public market? Can we build an individualized tour guide for visitors in a museum by inferring their interests based on the exhibits they visit? Can we optimize a space for efficient crowd flow by detecting choke points and idling? This project combines several aspects of AI such as sensor-based tracking, behavior recognition, and recommendation engines.
Spoken programming language
Is it possible to design a useful programming language that can be spoken and heard? What kind of “syntax” would be appropriate? How can we keep the semantics constrained so that a programmer’s short term working memory is not overloaded with deeply nested logic or long functions or too many functions? Which programming paradigms (procedural, functional, logical, object-oriented, etc.) are most appropriate for “programming in your head”? How can we implement a speech recognition system that is sufficiently accurate for this task?
Any successful work in this project is surely publishable, since nothing like it exists. The closest work that I have found is speech recognition for programming Java (i.e., speaking Java code). A spoken programming language, on the other hand, may need to be unlike any other programming language since the code cannot be seen. All programming would have to be mental, so limits to short term working memory become very important.
Visual homoiconic programming language
Homoiconic languages are those in which the code is written in the same way as data, so code can be processed as data. One benefit of homoiconic languages is that code can be generated or transformed during compilation or execution. Examples of homoiconic languages are the various Lisp languages, Prolog, and Tcl.
Unrelated to homoiconic languages, visual languages allow programmers to write “code” with visual constructs instead of text. These constructs may take the form of colored shapes representing abstract art, as with the Piet language, or data flow models like Max.
To my knowledge, there is no language that is both homoiconic and visual. Is that even possible? What visual construct would represent a function or unit of execution? A square perhaps? That representation should also work for data as well as code, i.e., a square could also be input to a function (so, squares consume squares). The payoff of such a project is that you would develop something completely novel. Bonus points if it can also be made useful.
Automated configuration of computer security scenarios
The computer security course at Stetson, and others like it at many colleges, was designed by Dr. Plante to give students realistic scenarios for penetration testing, exploitation and exploit detection, etc. He and I have begun working on a declarative language for defining scenarios and corresponding tool that generates a host of virtual machines (e.g., one VM per student or group) with specific operating system, network, and application configurations. The configurations are determined by the predefined textual scenarios. Vagrant and Ansible are used to launch and configure the VMs.
According to my research, no such tool exists. There are various repositories of prebuilt VM images for various security course scenarios. What we hope to offer is a more flexible tool in which the scenarios are defined in a text file, and the VM images are built on-demand. With this flexibility, the scenarios could even be randomly generated (for exam situations). They can also be shared among the community of security educators as the scenarios will be defined by simple text files rather than shared as complete VM images.
We already have a start on the implementation. See this GitHub repository.
Please talk to me if you have other ideas. Consider using some of the following devices previously purchased for senior research:
- Fitbit Surge (fitness/health monitoring)
- iPod Touch (mobile apps)
- Kinect (interfaces, games)
- Muse headband (brainwave monitoring)
- 4-wheeled robot with arm gripper, two cameras, LIDAR (robotics)
The Math/CS department typically purchases more equipment upon request.