12 Time New Roman double spaced, And well authenticated resources and cite everything/n 1 Project Description You can choose one and only one of the following projects: P1. Detection of Deep Fake P2. Reinforcement Learning for Cyber Security Details on the projects are given as follows. 1.1 P1. Detection of Deep Fake As Al becomes more powerful in generating realistic speeches, images, text, videos, etc., a natural question is the following: is it possible to detect fake data that is generated by Al. Since essentially all such data are generated by an Al system that uses deep neural network, such fake media or data are commonly called "deep fake." If you select this project topic, you may choose one of the following subtopics: P1.1 Implement a deep fake detector: you implement a detector that can detect with certain reli- ability deep fake data. Note that it is trivial to create a deep fake detector that can detect 100% of the faked media the detector will always mark the media as fake. To be useful, the detector should have low false negative probability and low false positive rate. - You can pick one of the media type: text, speech, image, video, or other types of sensory data. You will need to collect data, implement the code, and test the performance, and submit the code and a short report (2-5 pages, double spacing), with citations to all used resources, e.g., reference papers and code repositories used. P1.2 Propose a scheme that can be used to detect deep fake: You propose a scheme that can be used to identify which media are real and which media are fake. You should explore academic papers, industry reports, and other credible sources to understand the various approaches and advancements in detecting deep fakes. You need to propose a new scheme or extend an existing scheme some novelty is necessary. You will need to submit a report of about 10 pages, double space. - 1 1.2 P2. Reinforcement Learning for Cyber Security P2.1 A simulation environment for cyber security: you implement a simulation environment that is compatible with the Farama Gymnasium package. Specifically, you need to implement the make, reset, and step functions. The render function can be a dummy function (empty). The environment should be usable to simulate at least one sequential attack scenario. You need to submit the code, as well as a short report of 2-5 pages, double spacing, with citations to all used resources, e.g., reference papers and code repositories used. P2.2 A reinforcement learning based penetration tester, or reinforcement learning based intrusion detector. You can use an existing simulated environment, and only need to implement the agent. You need to submit the code, as well as a short report of 2-5 pages, double spacing, with citations to all used resources, e.g., reference papers and code repositories used. 2 Evaluation Criteria The report will be evaluated based on the following criteria: • Function (50 points): for projects P1.1, P2.1, P2.2, whether the desired function is imple- mented. Note that this is not binary - partial credits will be given to unfinished work. For P1.2, how good the proposed solution is — simple, reliable, and secure systems are preferred. • Quality of Writing and Presentation (50 points): Clarity, coherence, and rigor in the writing and presentation of findings. 3 Expected Effort As these topics are open ended, there is no end point for perfection. You are not expected to spend more than 2 days on the final project, including writing the report. Keep in mind that any of these topics is a viable research topic and may lead to a eventual publication. Although publication of the results would be excellent, your work will not be evaluated using the same high standard as that for academic publications. However, you are encouraged to pursue quality within the time available. The report should be submitted to Blackboard in PDF format. The code should be submitted in source code format. 2/n1.1 P1. Detection of Deep Fake As Al becomes more powerful in generating realistic speeches, images, text, videos, etc., a natural question is the following: is it possible to detect fake data that is generated by Al. Since essentially all such data are generated by an Al system that uses deep neural network, such fake media or data are commonly called "deep fake." If you select this project topic, you may choose one of the following subtopics: P1.1 Implement a deep fake detector: you implement a detector that can detect with certain reli- ability deep fake data. Note that it is trivial to create a deep fake detector that can detect 100% of the faked media - the detector will always mark the media as fake. To be useful, the detector should have low false negative probability and low false positive rate. You can pick one of the media type: text, speech, image, video, or other types of sensory data. You will need to collect data, implement the code, and test the performance, and submit the code and a short report (2-5 pages, double spacing), with citations to all used resources, e.g., reference papers and code repositories used.

Fig: 1

Fig: 2