Past Projects & Papers

African American Vernacular English (AAVE) is an English dialect spoken by the vast majority of members of the African American community. Decades of American stigma have made it an academic obstacle for African students in US public education [1–4]. Students who speak a different English dialect from SAE must learn to work between their own dialect and SAE, the dialect taught in formal speaking and writing in the US. This skill is called code-switching and differs from translating only in that it is specific to dialects from a shared parent language. There are similarities between the challenges of learning a different language and learning a different dialect. It has been shown that at a national level over the past decade, an achievement gap exists between African American students and other students in reading comprehension at different grade levels[5]. Historically, AAVE has also been directly linked to this achievement gap and is still viewed negatively in academic settings. Baugh notes how AAVE can create barriers for African American students to succeed, especially in Reading and Language Arts, and attributes much of this gap to the negative attitudes towards AAVE and its users [6]. Within US public school system grades K-12, this view has contributed to this gap[4]. Therefore, to reduce this academic performance gap, there is a critical need to develop methods to minimize this communication barrier. In the absence of such methods, the academic and professional advancement of the African American community will continue to be hindered . Artificial intelligence and machine learning methods, similar to Neural machine translation (NMT), will be useful tools, as such methods have been successfully used to facilitate learning English in other contexts [7].

The long-term goals of this research are to reduce the academic performance gap in reading comprehension that exists between African American students and their non-African American counterparts. To achieve this goal, our specific objective is to build an automatic dialect-to-dialect translation model between AAVE and SAE , specifically designed for educational use. The central hypothesis is that implementing an inter-dialect translation model developed from an AAVE/SAE corpus will improve academic performance in reading comprehension for African American students in US public education . This research effort also aims to increase the participation of African American students in AI education and research, which will further reduce this performance gap . Diversifying research participation in AI will translate to diverse AI tools and mitigate unwanted bias in applied AI. This research will have two specific aims:

1. Create a parallel corpora for AAVE and SAE: We will recruit research participants based on specific criteria that establish their proficiency in both AAVE and SAE dialects to construct a parallel corpus. Alternative methods such as online resources and large pre-translated text are not sufficient, as they are either rare or non-existent.
2. Develop AAVE/SAE translation model: We will construct a novel AI-driven dialect-to-dialect translation model that will be applied in educational settings to facilitate African American develop code-switching skills and be more proficient in SAE . We will base this model on a multi-head self attention-based sequential model that will capture the differences between the sentence pairs from the two dialects without losing any long-term dependencies.

The proposed research effort is creative and original because it develops an AAVE/SAE parallel corpus and a translation model that can be used for additional research that relates directly to the underrepresented African American community. This research addresses the lack of applied AI for the educational benefit of African American students in public education and can potentially change the social views of AAVE as a dialect of English. Accurate AAVE translation can develop tools used to create entire bodies of text in AAVE, which can be used to foster the development of code-switching skills of AAVE speakers. Specifically, a novel educational tool will be built to aid the development of code-switching skills in AAVE-speaking students. This translator will detect and suggest changes in written text in real-time. In addition to being used as a standalone tool, the translator can also be used as an add-on feature to an existing tool, like Grammarly. Our long-term plan includes collaborating with local K-12 schools with high African-American student enrollment to deploy this tool, iteratively improve based on feedback, and measure the impact.

• Sunandan Chakraborty (PI), Assistant Professor, Data Science
• Ted Hall, Clinical Associate Professor, School of Education

Much like other forms of bias, oppression, and hate, the incidence of antisemitism has been on the rise in the United States since 2015. Data collected by the Antidefamation League (ADL) indicate that schools, in particular, have been sites of antisemitic incidents. In the current sociocultural environment of permissiveness towards hatred, hundreds of incidents ranged from high school students using Nazi symbols, imagery, and antisemitic language (Pink, 2019a) to swastikas chalked onto the playgrounds of elementary schools. This collaboration between members of the IU School of Education-Indianapolis and the IU School of Liberal Arts at IUPUI draws on the ADL’s database of reported antisemitic incidents in schools across the nation, layering these reported incidents with a variety of demographic, historical, geographic and media-related data to develop a set of “community types.” From a curricular, training, and intervention-based perspective, a school-based incident of antisemitism would need to be addressed differently in Driggs, ID, than Brooklyn, NY. The community types will provide a guide for curating different packages to offer to different kinds of communities based on the factors explored in this research project. We will later partner with the non-profit organization Facing History and Ourselves to create targeted interventions—both pro-actively and responsively—for these schools and communities based on the community type in the form of curriculum, training, and workshops.

• Jeremy F. Price (PI), Assistant Professor, School of Education
• Carly Schall, Associate Professor, Sociology
• Jeffrey Wilson, Professor, Geography
• Mohammad Al Hasan, Associate Professor, Computer Science
• Xiao Luo, Assistant Professor, Computer Science

Stroke is a leading cause of morbidity, mortality, and disability in the U.S., accounting for around 795,000 new or recurrent strokes, 190,000 deaths, and $53 billion in medical costs each year [1]. Carotid arterial stenosis, a condition that involves blockage of blood flow from the heart to the head due to arterial lumen reduction, is the major cause of large-vessel ischemic strokes. Severe asymptomatic carotid stenosis (ACS) (lumen reduction above 85%) has been continuously challenging physicians to decide when and how to treat it because of the bifold stroke risks: an ACS will place a patient at more than a 3% increased risk of having a stroke in the next year whereas the risk of a patient having a stroke during endovascular therapy is 4.4% [2]. Uncertainties remain regarding the optimal technique for long-term prevention of stroke events in asymptomatic patients and, indeed, whether revascularization is sufficiently better than the best medical therapy [3]. Physicians would better address such stroke risks if they could access the lesional blood flow behavior and characteristics in carotid arteries. The literature has shown that hemodynamic flow parameters (HFPs) and local geometrical and flow patterns (LGFPs) are highly tied to stroke risks. HFPs include wall shear stress (WSS) [4-8], tensile stress9, trans-stenotic pressure gradient (TSPG) [10], and pulsatility index [11]; LGFPs include carotid web7, eccentric plaques5, recirculation [4,7,8], and vortex shedding [12]. Listed in Table 1, they can be obtained from the 4-dimensional blood flow that varies in 1-D time and 3-D space. However, they are not available concurrently in current clinical measurement.

The newly emerged engineering technique, called image-based computational hemodynamics (ICH) [13-15], is a promising non-invasive approach to address this medical need. Many studies [16-24] have demonstrated that ICH could reliably quantify the 4-D velocity and pressure fields in the diseased arterial system based on radiological imaging information such as Computed Tomography Angiography (CTA) and Doppler ultrasonography (DUS), from which one can, but is not limited, obtain the FHPs and visualize the LGFPs concurrently. Therefore, there is a critical need for clinicians to utilize the ICH capability to advance stroke prediction using FHPs and LGFPs.

ICH has been increasingly used in human cardiovascular systems to assess the severity of various vascular diseases, such as arterial stenosis [16-19] and arterial aneurysms [20-22]. It has become a promising noninvasive approach to quantifying hemodynamics (cardiovascular function) [23,24], resulting in innovation in medical device design, nonsurgical procedure planning, and tissue engineering. Despite the proven potential, ICH is not clinically accessible due to two bottlenecks. First, a typical ICH process from image data input to FHPs and LGFPs output requires proficient skills for image segmentation, flow modeling, high-performance computing, and post-processing of fluid data, as illustrated in Figure 1. Clinicians usually do not have these skills and cannot afford the training curve. Second, the ICH demands computation expense, typically taking days to weeks per case. It will take a much longer time to compute FHPs and visualize LGFPs, as our studies [25,26] indicate. The demanding software skills and high computation cost make ICH impractical in clinical settings. To address such a critical need, we propose to build a software system, 4D-ICH, to remove the bottlenecks and enable clinicians to use HFPs and LGFPs directly for predicting the lesional stroke risk of ACS.

• Huidan (Whitney) Yu (PI), Associate Professor, Mechanical Engineering
• Alan Sawchuk (PI), Professor of Surgery, IU School of Medicine
• Xiao Luo, Associate Professor, Computer Science
• Xiaoping Du, Professor, Mechanical and Energy Engineering

The number of STEM occupations (6.1% of all US jobs in the year 2019) is projected to grow at 8% between 2019 and 2029, in contrast to a growth rate of 3.4% for non-STEM jobs. Further, the median annual wage of STEM jobs in 2019 was $86,980, in contrast to that of non-STEM jobs ($38,160) (Bureau of Labor Statistics, 2020). Yet, at the same time, an estimated 48% of STEM higher education students switch to a non-STEM field or leave studies altogether (Seymour and Hunter, 2019), with women, underprepared students, and students of color being among the students at a high risk of switching and leaving. Further, 35% of the decisions to switch can be attributed to negative effects of STEM classroom experiences. A recent study has also found that only 26% of the observed STEM classrooms used research-based instructional strategies. Active learning, which underlies most research-based instructional strategies, refers to "a method of learning in which students are actively or experientially involved in the learning process and where there are different levels of active learning, depending on student involvement". IUPUI, the premier Urban Research University in Indiana, has been at the forefront of active learning research and implementation. One of the successful research, development and entrepreneurship projects in active STEM learning conducted at IUPUI is Peer Led Team Learning (PLTL) and Cyber PLTL.

• Snehasis Mukhopadhyay (PI), Professor, Computer & Information Science
• Pratibha Varma-Nelson, Associate Professor, Founding Executive Director of SEIRI
• Shiaofen Fang, Professor, Computer Science
• Lin Zhu, Senior Lecturer, Academic Advisor

Insects represent a large majority of the biodiversity on Earth, with approximately 5.5 million insect species, of which only ~20% are described. Traditionally, the identification of a new species requires a skilled taxonomist to visually extract morphological (physical) characters on the new insect that are not present in any other species that may be related. If the species is a new species, then, this species will contain morphological characters that are at odds with the existing keys, and thus a conclusion can be made it is a new species, and new characters are extracted and described so that other scientists can then make meaningful discoveries about the biology of the insect. Ultimately, the most important aspect of describing new species is rooted in the ability to define differentiable characteristics. What may not be differentiable by human recognition, may be differentiable by artificial intelligence (AI). Our long-term vision in this research project is to seek to answer whether AI can ultimately surpass human experts in extracting these subtle, yet potentially differentiable, morphological characteristics.

We believe that the AI model for insect identification should be designed by recognizing beforehand the open-world nature of the species identification problem and should not only identify future samples of species represented during training (seen classes) but should also be capable of detecting and identifying samples of unrepresented species (unseen classes). Identifying samples of unseen classes is an ill-defined problem. Although AI models can be tailored to operate in an open-set classification setting to detect insect samples with no matching classes in the training data, such an approach can only detect an insect sample as outlier but cannot identify its species of origin. As a by-product, the project will produce an AI system that can streamline image-based insect identification not just for common insect classes but also for rare and undescribed ones as well through image analysis and character extraction.

• Christine Picard (PI), Associate Professor, Biology
• George Mohler, Professor, Computer & Information Science
• Murat Dundar, Associate Professor, Computer & Information Science

Drug addiction, a chronic neuropsychiatric disorder characterized by compulsive drug seeking and taking even after prolonged withdrawal periods, causes a serious burden to individuals and societies around the world. It is estimated that 11.8 million people die each year due to substance abuse, which is more than the number of deaths from all types of cancer. Preventing relapse to drugs after prolonged periods of abstinence is one of the most challenging aspects of treating addiction in the clinic. Therefore, we developed a new concept “addiction-related motor engrams (ARMEs)” to define the potential involvement of CPNs, which originate from the supplementary motor area (M2) and project to, although not exclusively, the habit related-dorsal striatum including both dorsolateral and dorsomedial striatum (DLS, DMS), as key nodes connecting motor programs and drug-seeking behaviors. The mechanisms of relapse will be investigated in two underexplored aspects, i.e., the motor CPNs and the SCN4B subunits (a sodium channel subunit potentially responsible of increased neuronal excitability) in M2 and their striatal synapses. Our hope is to pave the way not only as a novel perspective in understanding addiction behaviors, but also to provide novel targets for clinical management of drug relapse.

• Yao-Ying Ma (PI), Assistant Professor, Department of Pharmacology and Toxicology, Indiana University School of Medicine
• Gavriil Tsechpenakis, Associate Professor, Computer Science

This project will produce preliminary results for the newly proposed concept at IUPUI – Connected, autonomous, recharging, and articulating vehicles for agile networks (CARAVAN) for enabling intelligent logistics and transportation. This project will advance fundamental knowledge in interdisciplinary fields that intersect with cooperative multi-agent control, vehicle-to-vehicle communications, reliability assessment, and human-computer interaction. The management and control of connected and autonomous heavy trucks will be studied using reinforcement learning and other AI methods to improve system safety and reliability over a wide spectrum of driving modes, including CARAVAN formation or dissolution, merging and splitting, disturbances from surrounding traffic, and resilience to component failures. Investigating human cognition in road scene understanding will also help the AI agents in CARAVAN interact with other road users efficiently. Driving videos recorded during different interaction scenes will be annotated via crowdsourcing experiments, from which visual cues and relationships with decisions will be retrieved with specifically trained algorithms. The labels will prepare a graph convolutional network for long-term intention and trajectory prediction for other road users.

• Lingxi Li (PI), Associate Professor, Electrical and Computer Engineering
• Renran Tian, Assistant Professor, Computer and Information Technology
• Xiaoping Du, Professor, Mechanical and Energy Engineering
• Snehasis Mukhopadhyay, Professor, Computer and Information Science

Kidney disease is a major health burden and a leading cause of mortality and healthcare expenditure. Chronic kidney disease (CKD) is highly prevalent, caused predominantly by diabetes and hypertension. A significant hurdle to develop effective treatment against kidney disease is a lack of understanding of cellular and molecular pathways underlying human CKD. The goal of this application is to solidify and extend our preliminary work to be applicable on large datasets and in the setting of disease. Our central hypothesis is that: when tissue is sparse, a machine learning approach to classify cell types based solely on 3D nuclear staining with DAPI staining will provide a comprehensive cell map of the human kidney and characterize the changes in the state of specific cells during kidney disease. To test this hypothesis, we will use the following specific aims:

1. Specific Aim 1: Comprehensively expand classification and spatial mapping to encompass major cell types, subtypes and rare cells in healthy kidney tissue based solely on 3D nuclear staining.
2. Specific Aim 2: Classify new cell subtypes induced by early diabetic kidney disease and link to disease pathogenesis by spatially mapping new cells classes to the nephron units.
3. Specific Aim 3: Enhance the interpretability and accessibility of our classification approach by devising measures of robustness and human readability.

• Tarek M. Ashkar (PI), Professor, IU School of Medicine
• Mohammad Al Hasan, Associate Professor, Computer Science

Diabetic retinopathy (DR) is a microvascular complication of Diabetes Mellitus (DM) and remains a leading cause of preventable blindness. It is asymptomatic in its early stages, but progression of the disease leads to an irreversible vision loss. However, adequate management of DM and regular comprehensive eye examination can preserve vision in DR in 98% of cases. DR screening and diagnoses currently involves highly trained and qualified human personnel and less automation. This results in a greater cost for both developing and developed countries. The major aims of this project are 1) develop novel deep learning models pretrained with transfer learning approaches for accurately detecting DR status 2) develop imaging feature-based clinically interpretable models, which can explain the contribution of retinal alterations in patients extracted by retinal image segmentation for detecting diabetic retinopathy and 3) validate and iteratively improve the developed models over time.

• Sarath Chandra Janga (PI), Associate Professor, Bioinformatics
• Ashay D. Bhatwadekar, Department of Ophthalmology, Eugene and Marilyn Glick Eye Institute, Indiana University
• Amir R. Hajrasouliha, IU Health Physicians Ophthalmology, Glick Eye Institute

Decision-making requires an agent to be open to novel information, but also resistant to distraction. Advancements in neuroscience and artificial intelligence (AI) provide a powerful heuristic to identify how the computational properties of a neural system is optimized to perform behaviors requiring either flexibility or intense focus. Understanding how the brain can move between these states is a critical and unmet need. The overarching goal of this proposal is to determine how computation in the brain is altered such that optimal control states emerge to guide the immediate needs of decision-making. Preliminary data indicate that actuating neural activity in an intelligent agent in the form of a recurrent neural network, yields quantitatively similar results to those seen in animal models of decision-making. This project will extend this work by identifying ways to increase the biological feasibility of intelligent agents.

• Christopher C. Lapish (PI), Associate Professor, School of Science
• Daniel Durstewitz, Professor for Theoretical Neuroscience, CIMH/ Heidelberg University
• George Mohler, Associate Professor, Department of Computer and Information Science

Lithium-ion batteries (LIBs) are the state-of-the-art energy storage systems and power suppliers for consumer electronics, electric vehicles, and the smart grid. In comparison to alternative rechargeable battery technologies (lead-acid, Ni-Cd, and Ni-MH), LIBs offer higher gravimetric energy density, higher volumetric energy density, higher voltage, higher cycle life, low shelf-discharge rate, faster charging time, and lower toxicity. Thanks to tremendous research and development efforts, the energy density of LIBs continues to grow at a rate of 10% per year. However, the required power for portable electronic devices is predicted to increase at a much faster rate of 20% per year. Similarly, the global electric vehicle battery capacity is expected to increase from around 170 GWh per year today to 1.5 TWh per year in 2030, this is an increase of 125% per year. Currently, there are no methods or design tools to support a faster LIB development. To keep up with the increasing energy demand, a breakthrough in battery design technology is required.

To significantly accelerate the development of LIBs and maximize their energy density, a multidisciplinary research team from Indiana University-Purdue University Indianapolis (IUPUI) proposes the development of a multitask Bayesian machine learning (BML) method. This approach offers the unique ability to process data from physical experiments, material databases, and multi-scale, multi-fidelity computational models, to guide the experiments and simulations that support the discovery and optimization of the electrode microstructures. In order to experimentally verify the effectiveness of the proposed Bayesian machine learning method, the IUPUI team is partnering with Advanced Renewable Power (ARP), LLC (arpteck.com), a company located in Indianapolis that develops and manufactures compact, high-performance power solutions to power the next generation of Hybrid Electric transit buses and delivery trucks. For this verification, ARP and the IUPUI team will apply the proposed multitask BML method to develop an optimal LiNixMn(2-x)O4 electrode, which will be fabricated and tested in a LIB cell.

• Andres Tovar (PI), Associate Professor, Department of Mechanical and Energy Engineering
• Likun Zhu, Associate Professor of Mechanical Engineering
• Hazim El-Mounayri, Associate Professor of Mechanical Engineering
• Khosrow "Nema" Nematollahi, CEO of Advanced Renewable Power, LLC and CAE-net; Associate Faculty at Purdue University School of Engineering

Alcohol use disorder (AUD) is among the most expensive health care problems faced by our society, with upwards of 184.6 billion dollars spent annually on treatment, alcohol-related morbidity, and mortality. Increasing evidence suggests a high level of comorbidity between AUD and psychiatric disorders. One hypothesis that has drawn considerable attention is that alcohol use exacerbates the dysregulation of mitochondrial function, contributing to the comorbidity of AUD and various mental illnesses. Our new findings suggest that Fkbp5 KO is a precisely controlled genetic animal model with a phenotype related to neuron function. However, current manual data counts of mitochondrial and neuronal differences are low throughput and lack the quantification of dynamic alterations. Our long-term goal is to develop a novel treatment for the comorbidity of AUD and mental illness. In this direction, the objectives of our proposed research are to investigate Fkbp5 in the regulation of mitochondrial morphology and networks, and neuronal development. Alcohol is a cellular stressor that exacerbates normal biological processes. The co-chaperone FKBP5 protein appears essential for hypothalamic-pituitary adrenal (HPA) axis function as well as cellular response to various insults. Our extensive preliminary data strongly support that Fkbp5 is involved in cell and organelle morphology changes, and could lead to enhanced mitochondrial function via regulation of microtubule-associated proteins. Therefore, our hypothesis is that Fkbp5 regulates mitochondrial morphology via regulation of microtubule-associated proteins.

• Tiebing Liang (PI)
• Gavriil Tsechpenakis
• Caroline Miller
• Kent E. Williams

Deep neural networks (DNN) or deep learning provide a major breakthrough technology in AI domain and are widely adopted recently for increasingly complex and large-scale machine learning problems including image classification, speech recognition, language translation, drug discovery, and self-driving vehicles. While DNN achieve remarkable success in terms of performance, recent studies have shown that DNN lack robustness, and hence are highly vulnerable to perturbations in the input space. This iAI white paper proposal is devoted to a systematic study on how to address the serious vulnerability in current DNN models and their impacts on applications. We seek not only the better understanding of the cause of the vulnerability of DNN models, but also attempt to propose an effective and efficient approach to make DNN much more robust for practical applications. Motivated by our previous research on multiresolution learning for the improvement of generalization and robustness of shallow neural networks, we propose to systematically investigate multiresolution learning for DNN models. The basic idea is to exploit the different resolutions of training data in DNN training process, allowing DNN to learn and extract better the underlined patterns and structures of data at different resolution levels. Consequently, the better the DNN models learn, the better their robustness. To the best of our knowledge, no existing work explores multiresolution learning paradigm for DNN yet, except for our preliminary study.

• Yao Liang (PI), Professor, Computer Science

Stroke is a leading cause of death in the western world and a significant cause of disability and loss of quality of life. It is critical to identify the asymptomatic carotid stenosis (ACS) patients at higher risk of stroke and provide corresponding invasive treatment. However, in a busy clinical set-up, it is difficult to do all tests or exams to identify all different features to identify the ACS patients with a higher stroke-risk potential. The objective of this research is to investigate whether artificial intelligence (AI) models can systematically identify the plaque and patient features of the ACS patients with higher stroke risk based on the real-world Electronic Health Records (EHR) data. If high accurate prediction can be achieved, early treatment and intervention can be planned to prevent strokes in the ACS patients at high risk for a stroke while avoiding unnecessary surgery with its risks in other patients. The AI models will permit the inclusion of all previous ultrasound studies, CTA studies, patient demographics, clinical exams, and different clinical phenotypes to be evaluated simultaneously looking for the possible development of stroke.

• Xiao Luo, Ph.D. (PI), Assistant Professor at the School of Engineering and Technology
• Alan Sawchuk, M.D., Professor of Surgery at the Indiana University School of Medicine

Chronic rhinosinusitis (CRS) affects ~12% of the US adult population, accounting for $10-13 billion in direct costs and $20 billion in indirect costs annually. Its large quality of life (QoL) burden results from nasal symptoms of obstruction, drainage, smell loss, and facial pain, as well as systemic issues including sleep difficulty, cognitive dysfunction, fatigue, depression, and asthma exacerbation. Traditional CRS management follows a linear algorithm ending with the consideration for treatment escalation to surgery when symptoms are not adequately controlled by medical therapies. However, in chronic diseases where different therapeutic options exist, there is a need for individualized evidence-based decision support at the point-of-care. Therefore, we hypothesize that application of artificial intelligence machine learning (ML/AI) can be used to uncover patterns.

• Vijay R. Ramakrishnan, M.D. (PI), Professor, Department of Otolaryngology-Head and Neck Surgery Director of Rhinology Research. Co-Director, Advanced Rhinology Fellowship Affiliate Scientist, Regenstreif Institute, Inc.

Stochastic games have been widely used to study network security where the attacker and the defender are considered as the players in the game. This proposal aims to study the existence of game theoretical equilibrium strategies in a stochastic game with a finite state space and finite action sets, and then apply the game theory analysis in Cybersecurity applications. AI algorithms will be employed to compute the equilibrium solutions of such multi-player games in a more efficient way. Also, multi-agent reinforcement learning can be studied well in the multi-player game of the cybersecurity problem.

• Subir K. Chakrabarti (PI), Professor of Economics, School of Liberal Arts
• Qin Hu, Assistant Professor, Computer Science

This interdisciplinary project will use state-of-art machine learning techniques to identify modifiable risk and protective processes in suicidal trajectories from a biopsychosocial perspective. By leveraging artificial intelligence in unique multimodal data across the biomedical image, social networks, clinical, psychological, and behavioral domains collected over time, this study will ultimately produce an easy-to-interpret, culturally appropriate, and accessible tool to clinicians, social workers, community health practitioners, and policymakers in suicide prevention programs.

With close collaboration between social work and AI experts, this project will make fundamental scientific contributions to computer or information sciences, social, behavioral, cognitive, and economics. The team will address the alarming disparities in suicide across developmental trajectories from children, adolescents to young adults. The integrated algorithm will allow clinicians, families, teachers, and community health practitioners to identify young people at risk before them being in crisis so that clinicians can intervene earlier, which could have a significant contribution.

The findings of this project will advance understanding of how AI-learning computation, combined with advances in social, psychological, and biomedical data analysis. Clinically, it will improve our knowledge of the origins, modifiable risks, and protective processes of disparities in suicide trajectories.

• Yunyu Xiao, Assistant Professor, School of Social Work
• Joan Carlson, Associate Professor, School of Social Work
• Shiaofen Fang, Professor, Computer Science
• George Mohler, Associate Professor, Department of Computer and Information Science