Reinforcement Learning for Cyber Security: Unmanned Aerial Vehicle Dynamic Routing via Multi-Armed Bandit
Dr. Xiaohua Xu - Thursday, February 27, 2:30 – 3:30pm, D-204
Security threats are evolving and getting more hidden and complicated. Detecting malicious security threats and attacks have becomes an enormous burdens to our world. We should proactively prevent and early detect security vulnerabilities and threats rather than patch security holes afterwards. This talk focuses on the unmanned aerial vehicle (UAV) dynamic routing scenario where there are multiple possible malicious targets. Each target has an information state which is not observed if the target is not visited by some UAV vehicle. The risk predications are obtained depending on the states observed and the information state is updated. Once the risk predications have been collected, the states of the targets evolve. The goal for the vehicles is to minimize the effect caused by malicious targets. We consider the UAV dynamic routing under the Multi-armed Bandit (MAB) paradigm and get constant approximation policies. The project deviates from the recent prevailing single-hop model, which cannot support multiple resource constraints simultaneously, or complex constraints such as interference constraints. Instead, the project proposes a robust multi-hop paradigm for the problem. In the new paradigm, the project proposes an integrated and cohesive approach to address target state dynamics subject to coexistence of different types of resource constraints. Furthermore, the new approach generates a family of algorithms, which significantly outperforms the solutions via merging of single-hop solutions. The project takes the proposed algorithms as representative to promote the study of approximation algorithm for UAV dynamic routing to upgrade the prevailing heuristics in the literature. Through theoretical and simulation studies, the proposed research is able to minimize the risk caused by malicious targets. The project advances the techniques such as semi-infinite programming, potential function analysis.
Machine Learning in Cancer Diagnosis and Personalized Medicine
Dr. Abedalrhman Alkhateeb - Tuesday, February 25, 2:30 – 3:30pm, D-204
Cancer is caused by genetic changes that lead to abnormal cell growth in tissues. These cells can potentially invade different parts of the body which is known as tumor cancer, and leads to fatalities. Detecting the genetic changes in the tumor cells plays a key role in the diagnosis, prognosis, and treatment of the disease. Based on the genetic changes and other genomic variations such as DNA mutations and alterations, the genes transcribe into different mRNAs then translates into different proteins for different variants. Machine learning techniques are utilized to build prediction models that can identify a set of biomarkers that separate specific cancer stage or subtype from the rest. These models are based on next-generation sequencing (NGS) data and pharmacogenomics databases.
The common questions in the field of personalized medicine are how i) to prescribe specific therapeutics best suited for an individual based on oncology, where the therapeutics may consist of one or a combination of drugs, surgery, hormonotherapy, radiation, and clinical trials. ii) to estimate the required amount of the drug (the dosage). iii) to find a less invasive cancer test instead of biopsy. This presentation focuses on the basic models, and how to build a multi-disciplinary research team in personalized medicine for cancers. The advancement in the computational resources and NGS technology made this practice financially feasible for the patients, increased the survival ratio, and improved the quality of life.
Building Reliable and Trustworthy AI in the Cognitive Era
Dr. Guan Yue Hong - Friday, February 21, 2:30 – 3:30 pm, D-132
Artificial intelligence (AI) presents tremendous opportunities today in areas like smart manufacturing, personalized healthcare, autonomous transportation, and effective education. However, if an AI system starts behaving erratically or failing unexpectedly, we lose trust. Even if AI works as designed, but in a manner that does not align with human expectations, we still tend to distrust it.
Trust must be earned, and must therefore be purposely built into AI systems for society to adopt and embrace them. Considering AI as software that runs on some architecture, Dr. Hong will discuss her research and practical experience in building reliable software and explore areas of opportunities to instill trust into AI. The path toward achieving trustworthy AI is multifaceted. It is about developing AI technology that would be reliable and safe to use. It is about trust that the technology would be unbiased against age, gender, and any disadvantaged groups. Trust in AI is about understandable and transparent outcomes. It is also about trust that new technology would be used to augment our intelligence and multiply our cognitive power rather than replace us.
Dr. Hong will share her research activities and explore a few research directions in an effort to build reliability and trust into our increasing complex AI systems.
Doctoral Oral Examination: High Performance and Machine Learning Algorithms for Brain fMRI Data
Candidate: Taban Eslami
For the degree of: Doctor of Philosophy
Department: Computer Science
High Performance and Machine Learning Algorithms for Brain fMRI Data
Committee: Dr. Fahad Saeed, Chair, Dr. Alvis Fong, Dr. Kevin Lee, Dr. Ajay Gupta
Friday, February 21, 2020, 1 to 3 p.m. C122 Parkview Campus
Brain disorders are very difficult to diagnose for reasons such as overlapping nature of symptoms, individual differences in brain structure, lack of medical tests and unknown causes of some disorders. The current psychiatric diagnostic process is based on behavioral observation and may be prone to misdiagnosis
Noninvasive brain imaging technologies such as Magnetic Resonance Imaging (MRI) and functional Magnetic Resonance Imaging (fMRI) make the process of understanding the structure and function of the brain easier. Quantitative analysis of brain imaging data using machine learning and data mining techniques can be advantageous not only to increase the accuracy of brain disorder diagnosis but also to unravel unknown facts about the complex function of the brain. Research studies have shown functional connectivities of brain contain discriminative patterns that are widely used in a variety of studies such as fMRI classification.
In this study, we designed machine learning and deep learning models for diagnosing brain disorders such as ADHD and ASD using fMRI data. In order to reduce the risk of overfitting in deep learning methods, we proposed a data augmentation approach for generating artificial samples from available data. Our models are able to improve the accuracy of classifying healthy samples from patients up to 28% comparing to state-of-the-art solutions.
Analysis of fMRI data considering a huge number of voxels (smallest addressable element of fMRI data) is very time-consuming. One example is computing pairwise functional connections between voxels using measures like Pearson's correlation. To tackle this issue, we designed two GPU based frameworks based on matrix multiplication for computing pairwise correlations that deliver around 3 times speedup against state-of-the-art GPU based methods. We expanded these frameworks to compute dynamic functional connectivity which involves computing multiple sets of pairwise correlations each associated with specific time windows in original time series followed by designing two methodologies for reducing the space requirements of pairwise correlations.
Ph.D. Dissertation Defense: Formal Verification on Software Testing
By Jialiang Chang on January 28th, 2020. D-201, 10:00 am - 12:00 pm.
Dynamic Analysis on Concurrent Programs Based on Scheduling Control and Input Selection
By Hao Li on September 25, 2018
Mandatory Integrity Control for Arbitrary Structured Data
By Wassnaa Al-Mawee on August 10th, 2018