Hello! I am a PhD candidate in the Computer & Information Science & Engineering at the University of Florida. I am interested in exploring ways to make Machine Learning trustworthy, fair, and safe. I have explored explainability and fairness aspects of Machine Learning to make it fair in different areas such as justice and computer vision. I hope to continue exploring this area in future. I majored in Computer Science from Amirkabir University of Technology (AUT) and graduated with a minor in Mathematics in 2014. At AUT, I worked on research projects about use of reinforcement learning in mobile networking. In my free time, I enjoy reading biographies, cross stich and ride my bike.
I have diverse background in domains such as Machine Learning/AI, Human Computer Interaction, and Software Development
Deep Learning Framework: Tensorflow/Keras (using frequently), PyTorch (using if needed) Other: python, NLP, Computer vision, reinforcement learning
Technologies: Dialog fow,Affinity Diagrams, Balsamique, Invision, prototyping, wireframe,task flow Skills: Qualitative analysis, Quantitative analysis, Hypothesis Testing, Survey Design
Web Development: HTML5, CSS3, ReactJs, NodeJS, Javascript, jQuery. Other: Java, Agile methodologies, Heroku, Vercel, Object-Oriented Design
I have diverse work experience in academics, non-profit organizations and Tech startups related to Machine Learning/AI.
Built strong communication with other data scientists for better assessment of techniques in the ML pipeline by leveraging strong knowledge of ML reporting and fairness paradigm
Built object detection models for different type of objects in the city environment for 3D modeling of images. Formulated styleGAN method to generate realistic pictures of different type of trees to establish 3D models of jungles or state parks using national data points.
- Developed an open-source Risk Assessment Simulator based on information gathered through multiple FOIA requests. This simulator has been developed using react.js and JavaScript. - Evaluated the current differential privacy effort happening at U.S. Census Bureau and bolded the current problems. - Investigated the current development of content tracing app and their consequences for Public privacy.
Manage and execute multiple projects related to machine learning, artificial intelligence, algorithmic fairness, and human computer interaction. Collaborate with researchers from different disciplines, including epidemiology, consumer sciences, and philosophy to provide in-depth technical information for software development process.
- Analyzed time series data to find the existing trend of using Stimuli drug in Schools. - Designed the research agenda and most efficient ways to visualize the data using Tableau. - Predicting the readmission rate for patients who over utilize the insurance. Using unstructured four-year ER visits data, I was able to improve the accuracy by %10.
Ethical and fairness study of Predictive Policing
NeighborHood Rides
Fairness-Aware Methodology in Juvenile Recidivism
Application Quest
Sentiment and Trust in AI
AR Therapy
Mobile Decision Aid (MODA)
Trust and QOS optimization in adhoc networks
Outstanding International Student Award at UF|2020
Cornell Summer School on Designing Technology for Social Impact Scholarship|Summer 2021
Bank of America Travel Award to attend Grace Hopper Celebration|Fall 2020
Among the three reciepents of Media Democracy Fund fellowship|Summer 2020
Google travel award for BPDM conference at Howard University|Febraury 2019
Gartner Graduate Fellowship CISE department at UF|March 2020
Induction to AEL Top Graduate Student Honor Society | Fall 2019
Concepts: Neural networks, structure of ML projects, CNN, RNN
Concepts: MRI segemntation, transfer learning, cox survival analysis
Concepts: Data cleaning, problem solving, critical thinking, data ethics, and data visualization
As time progress, autonomous vehicles may be a common mode of transportation, and companies like Lyft and Uber will adopt them in place of human drivers. In this paper, the barriers to adopting autonomous vehicles in rural areas are discussed by examining the current struggles of rural communities with respect to finance, transportation infrastructure, policy, and demographics. This project resulted into one paper in ISTAS conference 2021 and one open piece paper in Technology and Society Magazine. stay Tuned!
Application Quest, known as AQ, could be used in domains such as HR and admissions to reduce the implicit bias and human error. Using unsupervised learning methods, AQ will select the most holistic and representative sample to increase diversity. We conducted experiments to compare AQ with other state-of-art undersampling methods. The paper is accepted in the International Conference of Machine Learning and Data Mining.
Risk assessment tools are used throughout the nation to treat and rehabilitate juvenile delinquents. However, Racial disparity is a significant problem in these tools, which leads to a harsher sentencing process for adolescents of color. Prior research has shown that the neural network outperformed the other existing methods in predicting recidivism by far. PACT data is used for indicating the recidivism in the Florida Juvenile Justice Department. This proposal aims to develop a methodology to assess the predictive performance and fairness of the machine learning methods used in juvenile recidivism prediction. We use ML explainability combined with data analysis techniques to explain the existing disparities. Moreover, we aim to find fair learning representation based on the current sensitive attributes and their proxies. Lastly, we will use interpretable ML to provide interpretations of how the performance could be improved while preserving fairness.
Dynamicity and infrastructure-less nature of MANETs expose the routing in such networks to a variety of attacks, and moreover, make the conventional fixed policy routing algorithms inefficient. To deal with the routing challenges and varying behavior of malicious nodes in such networks, employing reinforcement learning algorithms and proper trust models seem promising. In this paper, we introduce a cognition layer in parallel and interacting with the network layer which comprises two cognitive processes: path learning (routing) and trust learning. The first process is based on machine learning algorithms and the latter is based on trust management. We compare our algorithm, TQOR, with a well known trust-based routing protocol, TQR, in terms of three measures of performance. The simulation results show better end-to-end delay and communication overhead which further improve as time progresses, without sacrificing the data packet delivery ratio.
Over the past 20 years, researchers have investigated the potential of Virtual Reality (VR) to enhance rehabilitative therapies by improving motor control, supporting motivation, and offering analgesic effects. Prior work indicates that patient adherence to prescribed in-home regimens has significant impact on recovery time. Though Connected Health Technologies and Virtual and Augmented Reality (AR/VR) may maximize in-home adherence and recovery, questions about design and deployment remain. We designed a first-person Augmented Reality (AR) experience to elicit user and practitioner perspectives about AR for rehabilitative contexts. We found significant differences between patient and practitioner-report of regimen adherence. We also identified key attitude barriers to adopting VR/AR for clinical practice which may impact support for in-home VR/AR use. Findings from these studies inform directions for future research and development about the use of VR/AR in a therapeutic context. This work recieved the best student award in the annual Meeting of Human Factors and Ergonomics Society in 2019.
Conversational Voice User Interfaces (VUIs) help us in performing tasks in a wide range of domains these days. While there have been several efforts around designing dialogue systems and conversation flows, little information is available about technical concepts to extract critical information for addressing the users’ needs. AI could help us in extracting dialogue information and address user needs. We developed an AI-based mobile-decision-aid (MODA) that predictively models and addresses users’ decision strategies to facilitate users’ in-store shopping decision process. Here we share our design and subsystems to make our research reproducible. To make our research reproducible, the code of backend server and dialogflow agent used will be published on my github!
AI and Machine Learning gained so much popularity recently. However, there are many conversations about the disparities and biases coupled with these technologies.In this project, we conducted user studies to learn more about the sentiment and trust of participants toward AI-powered technologies. It resulted into a paper which is accepted to ISTAS 2020. It is currenlty in press.
Machine Learning has become a popular tool in a variety of ap-plications in criminal justice, including sentencing and policing. Media hasbrought attention to the possibility of predictive policing systems causing dis-parate impacts and exacerbating social injustices. However, there is little aca-demic research on the importance of fairness in machine learning applicationsin policing. Although prior research has shown that machine learning modelscan handle some tasks efficiently, they are susceptible to replicating systemicbias of previous human decision-makers. While there is much research on fairmachine learning in general, there is a need to investigate fair machine learningtechniques as they pertain to the predictive policing. Therefore, we evaluatethe existing publications in the field of fairness in machine learning and pre-dictive policing to arrive at a set of standards for fair predictive policing. Wealso review the evaluations of ML applications in the area of criminal jus-tice and potential techniques to improve these technologies going forward. Weurge that the growing literature on fairness in ML be brought into conversa-tion with the legal and social science concerns being raised about predictivepolicing. Lastly, in any area, including predictive policing, the pros and consof the technology need to be evaluated holistically to determine whether andhow the technology should be used in policing. This paper is accepted into the Artificial Inteligance and Law journal 2021. Stay tuned!!