The UAlbany Artificial Intelligence Symposium

Presenter 1:  Hany Elgala, Associate Professor, Electrical and Computer Engineering (ECE)* 

Title: Machine Learning for Wireless Communications and Networks 

Abstract: Wireless communication has evolved to encompass a variety of application areas, from high-speed multimedia streaming for the exponentially growing mobile user devices to the massive deployment of resource-constrained devices to enable Internet-of-Things (IoT) tasks such as sensing and inference. The talk will highlight the research activities on machine learning (ML) as an alternative approach to realize classical signal processing and control operations required to build a transceiver and operate a wireless network. The talk will also address the concern about the associated electric and magnetic field (EMF) exposure, the data-driven study on population exposure, and the initial project to realize an ML-based EMF exposure map for the UAlbany campus. Finally, the talk will briefly introduce the ML for Engineers course offered by the ECE department. 

Presenter 2: Yiming Ying, Professor, Math and Statistics, UAlbany* 

Title: Topics in Foundational AI 

Abstract: In this talk, I will describe some topics in foundational AI including how to understand the generalization of deep learning and how to design privacy-preserving and fair machine learning algorithms. 

Presenter 3: Charalampos Chelmis, Assistant Professor, Computer Science* 

Title: Robust Learning with Noisy Label Detection and Counterfactual Correction 

Abstract: One of the critical factors in building an accurate classification model is training data quality. Although most classification algorithms implicitly assume perfect training data fidelity, real-world training data are often noisy. Most AI and machine learning models are instructed to ignore noisy training data instances. However, dropping suspected noisy label samples from the training set can result in potentially detrimental data loss, which can in turn lead to either an overfitted model or a totally unusable training dataset. In this talk, I will discuss my cutting-edge results on automatically learning how to identify noisy labeled data instances and repairing their labels so as while at the same time training a robust classification model with noisy labeled data. 

Presenter 4: Yunlong Feng, Assistant Professor, Mathematics and Statistics* 

Title: Robust Machine Learning: Theory, Methods, and Applications 

Abstract: The goal of robust machine learning (ML) is to develop models and algorithms to learn from imperfect data, such as data with errors, outliers, skews, or biases. My research focuses on robust ML are two-fold. On one hand, I am interested in developing foundations for well-established yet not fully-understood robust ML models and algorithms. For instance, we recently conducted some theoretical studies on a family of nonconvex risk minimization schemes that are widely used in computer vision and engineering. On the other hand, with my collaborators, we are developing new robust ML approaches and exploring their new application domains. As an example, we recently investigated their applications to biological spectral image data and Raman spectral data. The overarching goal of my research in this direction is to build an adaptive ML system that has theoretical guarantees and can accomplish a wide range of ML tasks with imperfect data. 

Presenter 5:  Ming-Ching Chang, Associate Professor, Computer Science* 

Title: AI Research Summary of the SUNY Albany CVML Lab 

Abstract: I will provide research summary of the CVML Lab, which covers 4 areas: Computer Vision, Digital Media Forensics, AI & Cybersecurity, and Scientific Machine Learning. 

Presenter 6:  Sahebi Shaghayegh, Assistant Professor, Computer Science* 

Title: Machine Learning for Human Learning 

Abstract: Artificial intelligence is beginning to transform education as we know it. Particularly, online educational environments provide an abundance of data that can be used to optimize the learning experience for students, facilitate teaching for instructors, and support strategic decisions for educational institutes. In this talk, I discuss my research on AI for student knowledge modeling and behavior modeling in online educational environments. 

Presenter 7:  Petko Bogdanov, Associate Professor, Computer Science* 

Title: Domain-informed machine learning for spatio-temporal data, biological networks, nanomaterial design and beyond 

Abstract: Complex data requires domain knowledge to be infused in the machine learning algorithms employed for analysis, prediction and decision making. For example, spatiotemporal weather phenomena follow physical rules and constraints and failure to incorporate those in predictive ML models may hurt performance and explainability. Similarly, making inferences based on gene expression measurements should account for known gene interactions from the literature. In this talk, I plan to summarize some of the inter-disciplinary work in my lab geared towards domain-aware machine learning. 

Presenter 8:  M. Dolores Cimini, Director, Center for Behavioral Health Promotion and Applied Research, Educational and Counseling Psychology* 

Title: Innovation for the Public Good: Blending AI and Behavioral Science to Address Suicide Risk, Substance Use, and Related Behavioral Health Concerns 

Abstract: Recent research has highlighted the critical role of artificial intelligence (AI) in predicting and preventing serious behavioral health issues such as suicide and substance use and providing timely and responsive mental health service access to individuals and populations which would otherwise not have had equitable access to such interventions. The proposed lightning talk will discuss how scientists are using AI and machine learning to predict and address mental health and substance use-related risk through predictive modeling, integration of AI as part of diagnosis, treatment, crisis intervention, and practitioner training, and providing timely, responsive, and life-saving access to services to populations experiencing health disparities. Barriers associated with the integration of AI in mental health and substance use service delivery, including safety, protection of privacy, accuracy and reduction of bias in AI algorithms, and the responsibility to respond appropriately and ethically to high-risk circumstances with the benefit of scientist and practitioner partnerships planned through our research agenda, will be addressed. 

Presenter 9:  Daphney-Stavroula Zois, Associate Professor, Electrical and Computer Engineering* 

Title: Cost-sensitive Machine Learning and Signal Processing 

Abstract: In many application domains (e.g., healthcare, criminal justice), it is necessary to reach an accurate decision in a timely manner using limited resources (e.g., costly tests, time-consuming evidence collection). At the same time, it is desirable to tailor decisions to each individual case (e.g., patient, defendant). Most of the existing machine learning and signal processing approaches, however, ignore resource constraints and/or acquire a general solution for all cases. This talk will present an overview of my group's recent research activities on enabling resource-efficient inference and decision-making with applications including but not limited to search and delivery of human services, cyber-physical-human systems and brain-computer interfaces. 

Presenter 1:  Vanessa Przybylo, Post-Doctoral Associate, Atmospheric Sciences Research Center* 

Title: Detecting the Presence of Precipitation in New York State Mesonet Imagery at Night using Convolutional Neural Networks 

Abstract: A variety of pre-defined convolutional neural network architectures are harnessed to predict the presence of precipitation in New York State Mesonet (NYSM) imagery at night.  126 stations across the state record panoramic images with a standard temporal frequency of five minutes.  Tens of thousands of images, each with diverse backgrounds, are labeled based on the presence or absence of precipitation streaks from infrared detection.  The two main categories of precipitation or no precipitation are expanded upon to predict obstructions such as spider web formation or lens-covering raindrops.  Procedures to enhance trustworthy output are integrated within the image labeling process through inter-coder reliability metrics, and uncertainty quantifications are recorded through probabilistic model predictions.  Unique model architectures are compared and performance metrics are evaluated to deduce model generalizability on future imagery or the geographic expansion of imagery.  The end goal is to create a deep neural network to provide real-time, operational capacities for the prediction of precipitation across New York State to be used alongside the suite of NYSM instrumentation for actionable insight.   

Presenter 2:  Justin Curry, Assistant Professor, Mathematics and Statistics* 

Title: Modeling Solar System Wide Internet via Topological Methods 

Abstract: Recent collaborative work between UAlbany's Math Department and NASA uses machine learning and topological data analysis to provide better models for time-varying networks with high latency. Understanding such structures is necessary for the development of scalable autonomous routing in a future solar system wide internet. Moreover it provides an interesting case study in how featurization in machine learning often has an interesting pipeline with various levels of inverse problems, illustrating another area of my research. 

Presenter 3:  Kevin Knuth, Associate Professor of Physics, Physics* 

Title: The Discovery and Characterization of Other Star Systems and their Planets 

Abstract: We are living in an unprecedented era of discovery in which missions like the Kepler Space Telescope and the James Webb Space Telescope are providing data allowing for the discovery and study of other star systems and their planets, which are referred to as exoplanets.  In this talk I will provide an overview of how machine learning makes these discoveries possible through model-based learning.  This will be followed by a summary of the discoveries our laboratory here at UAlbany has made, which include a potential (the first) Trojan planet and a remarkable Triple-Star System in which the stars orbit one another in a matter of hours! 

Presenter 4:  Matthew Szydagis, Associate Professor of Physics, Physics* 

Title: A Naive Bayesian Classifier for Identifying Unidentified Aerial Phenomena 

Abstract: Recently, Unidentified Aerial Phenomena or UAP, formerly called UFOs, have come under increasing scrutiny, with the Defense department and now even NASA taking them extremely seriously. AI/ML techniques are the right tools for identifying and classifying objects in our skies in photos/videos. Anomaly finding is a specialty of such techniques, which can help classify not only the "mundane" objects like birds and human-made aircraft, but find the outlier objects/phenomena possessing anomalous characteristics. In this talk we will introduce C-TAP or Custom Target Analysis Protocol, the only known unclassified, non-governmental software for motion detection in FLIR (Forward Looking Infrared) videos. It was written for UAPx (https://www.uapexpedition.org), a 501c3 federal non-profit whose mission it is to identify UAP in a scientific manner, mitigating bias but not shying away from the most speculative possibilities, should hard evidence present itself in their favor. Preliminary C-TAP results from ~600 hours of infrared imagery will be presented. 

Presenter 5:  Fangqun Yu, Research Faculty, Atmospheric Sciences Research Center* 

Title: Use of machine learning to improve global climate models 

Abstract: The radiative forcing of anthropogenic aerosols associated with aerosol–cloud interactions (RFaci) remains the largest source of uncertainty in climate change prediction. The calculation of particle number concentration (PNC), one of the critical parameters affecting RFaci, is generally simplified in global climate models. Here we employ outputs from long-term (30-years) simulations of a global size-resolved (sectional) aerosol microphysics model and a machine-learning tool to develop a Random Forest Regression Model (RFRM) for PNC. The PNC RFRM has been implemented in a global climate model and shown to improve the model performance. This work highlights a promising approach based on machine learning to reduce uncertainties of global climate models without compromising their computing efficiency. Future work can also include using Artificial Intelligence to generate high-resolution air quality data for health effect studies and environmental justice assessment. 

Presenter 6:  Shao Lin, Professor, Environmental Health Sciences* 

Title: Use of Data Science Method in Environmental And Health Research 

Abstract: My talk will include the following contents: 1) Present my completed and ongoing projects using data science methods (e.g., a) identifying environmental risk factors for asthma among children; b) developing community vulnerability index for climate change or extreme weather events; c) refining exposure indicators for meteorological factors; 4) detecting exposome for COVID-19 using machine learning method). 2) I propose to prepare a NIH R01 or training or center grant connecting AI and environmental health. 3) request collaborators at AI across campus. 

Presenter 1:  Mila Gasco Hernandez, Research Director / Associate Professor, Center for Technology in Government / Department of Public Administration and Policy* 

Title: Adoption and implementation of AI in the public sector from a public administration/management perspective 

Abstract: AI has advanced as one of the most prominent technological innovations to push the conversation about the digital transformation of the public sector forward. However, most existing literature concentrates on the areas of application, potential benefits, risks, and concerns. We have seen little systematic evidence of, for example, what the potential public value of automated public services is, what the actual implementation of these tools and services looks like in government agencies, and how AI impacts the design and delivery of a range of public services, potentially creating new service models. During this lightning talk, I will present my research agenda on the adoption and implementation of AI in the public sector from a public administration/management perspective, which includes current research on topics of organizational challenges and organizational routines as well as impact on processes and service delivery, and plans for funding opportunities related to AI accountability and citizen engagement. 

Presenter 2:  Meghan Cook, Program Director, Center for Technology in Government (CTG UAlbany)* 

Title: AI Modeling and Visualizations for Anomaly Detection in NYS’s Voter Registration Data: Informing NYS Board of Elections Next Generation Enterprise Investments 

Abstract: Voter registration has been recognized across the country as one key source of cybersecurity vulnerability in election processes. Aware of the advertised potential of artificial intelligence (AI) driven data analytics, NYS Board of Elections asked the Center for Technology in Government to explore AI techniques to monitor voter registration data; specifically going beyond current efforts to detect errors in the data to identifying patterns and anomalies in NYS’s voter registration database. CTG UAlbany led an interdisciplinary team of social and computer science researchers, faculty, and students from CTG UAlbany, the College of Engineering and Applied Sciences, and the College of Homeland Security and Emergency Preparedness and Cybersecurity in stakeholder needs assessment, data forensics, statistical and AI modeling, and development of visualizations of voter registrations data.  Producing prototypes designed to build a shared understanding among state and county election leaders, the UAlbany team helped inform NYSBOE’s future investments in AI solutions. 

Presenter 3:  Esra Gules-Guctas, Adjunct Professor, Political Science* 

Title: Dynamics of Perception in the Context of Algorithmic Accountability 

Abstract: Although, algorithmic tools and artificial intelligence applications have now become a vital part of our daily lives, reshaping the way we live and the way governments manage public services, we know little about how ordinary individuals understand their justice problems and respond to them in an increasingly automated world. Understanding the factors that influence decisions to mobilize the law have important implications for understanding the current state of judicial scrutiny of automated decisions. Attention to the role played by courts is crucial particularly because policy formulation is rarely swift; the judiciary may be the first to address the novel challenges posed by algorithmic governance. 

Presenter 4:  Zheng Yan, Professor, Educational and Counselling Psychology* 

Title: Understanding AI among Ordinary People: An Unprecedented Challenge 

Abstract: A technology typically goes through two stages: (1) invent and develop it by scientists and engineers and (2) access and use it by ordinary people. Modern AI as an emerging technology is currently moving from the development stage to the use stage. Thus, how ordinary people understand AI becomes increasingly important because it will impact how people access and use AI (e.g., people’s knowledge about an autonomous car will determine whether they support, learn, buy, or driving it). This talk will (1) review the extensive literature on ordinary people’s understanding of three technologies, computers, Internet, and cellphones, (2) discuss four key features of AI as a scientific concept, emerging vs. emerged, complex vs. simple, dynamic vs. stable, artifact vs. natural in the broad context of comparing these four technologies in terms of conceptual understanding, and (3) outline a proposal targeting NSF’s Human-AI Interaction and Collaboration grants. 

Presenter 5:  Aaron Benavot, Professor, Education Policy and Leadership/ School of Education* 

Title: Harnessing AI and Natural Language Processing (NLP) to Assess Country Policies and Programs in Climate Communication and Education 

Abstract: As a leading researcher in the Monitoring and Evaluating Climate Communication and Education project (www.mecce.ca), I will describe the suite of indicators being developed to evaluate global efforts in climate communication and education (CCE). I will highlight the construction of current and future indicators to assess countries’ CCE policies and programs and additional work needed to evaluate actual country implementation of CCE commitments. This requires careful analysis of hundreds of official documents in dozens of languages, many of which have already been compiled by the MECCE team. The envisioned analysis would employ AI and NLP to construct contextualized understandings of CCE based on national curricular frameworks, education sectors plans, technical/vocational education and training policies, subject syllabi and possibly textbooks, mainly in primary and secondary education. My talk will emphasize the benefits of this research for global monitoring and benchmarking, intergovernmental negotiations and improving the quantity and quality of CCE worldwide. 

Presenter 6:  Unni Pillai, Associate Professor, College of Nanoscale Science and Engineering+ 

Title: Diffusion of AI and its Impact on Productivity and Employment in the US Economy 

Abstract: The talk will focus on two central themes in my research: 

1. Modeling the diffusion of AI across different sectors of the economy. This strand of research tries to understand the factors that determine the take up of AI across different sectors in the economy.  

1. Understanding the impact of AI on Productivity and Employment in the US economy. As AI enables machines to perform many of the tasks that were previously done by humans, labor productivity in the economy will improve. But as machines become more capable at tasks, previously done by humans, unemployment in the economy would increase. This strand aims to understand the  extent to which advances further in AI will exacerbate these effects, as well as role of government policy in this context. 

Presenter 7:  Kabel Stanwicks, Senior Assistant Librarian & Head of Access Services, University Libraries* 

Title: Leveraging AI and user-generated description to enhance library cataloging, search, and discovery 

Abstract: My current research demonstrates that user-generated social tags in the popular music domain align with the constructs of Name Authority Records (NARs). These are the records that libraries and cultural institutions use to describe the people responsible for creating works found in our collections. I would like to present a general overview of this research and discuss the next steps in my research agenda, which will examine how social tags relate to the controlled vocabularies used in NARs, explore the use of artificial intelligence to enhance NARs and library search interfaces, and apply these methods to other knowledge domains. This research has the potential to enhance user search and discovery, while simultaneously reducing the significant resources libraries and cultural institutions invest in creating and maintaining NARs. 

Presenter 1:  Nathaniel Cady, Interim Vice President of Research, SUNY Polytechnic+ 

Title: AI Hardware Development 

Abstract: My research focuses on the development of novel microelectronics hardware for AI, machine learning and neuromorphic computing. My group utilizes advanced nanofabrication processes to build nanoscale memory devices that are capable of mimicking synaptic functions in the human brain. Combining these devices with conventional microelectronic circuitry enables us to demonstrate efficient in-memory computation and the first steps towards hardware acceleration for AI applications. 

Presenter 2:  Serge Oktyabrsky, Professor, CNSE+ 

Title_1: Multilevel Functionality of Al-Sb Phase Change Memory  for AI Hardware 

Abstract_1: “Compute-in-memory” hardware utilizing novel phase change memory (PCM) materials will improve the energy efficiency and computation density required for Artificial Intelligence applications. Multiple resistance levels with fine control and low stochasticity are essential for Resistive Processing Units acting similarly to biological synapses. Tellurium-free group III-Sb binary alloys show faster crystallization, lower power operations, and improved stability as compared to the most studied Ge-Sb-Te system. Specifically, Al-Sb PCM cells demonstrate record-high resistance contract, lower resistance drift allowing for 32 programmed levels, and iterative pulse programming. 

Title_2: Electrically - Tunable Infrared Sensor for AI Object Recognition 

Abstract_2: A joint team from SUNY Poly and University at Buffalo – SUNY is developing an electrically-tunable infrared sensor with the capability of on-the-fly control by an artificial intelligence agent for confident object recognition. The sensor utilizes a quantum well infrared photodetector (QWIP) with asymmetrically-doped pairs of  QWs that allows for control of electron population on the split ground-state level by the external electric field. The ratio of the responsivities of the two bands is controlled over an order of magnitude. We projected the benefits of the proposed sensor by collecting a dataset with two spectrally-offset commercial IR cameras and performing the task of object detection. 

Presenter 3:  Satyavolu Papa Rao, Vice President Research (NY CREATES) & Adjunct Professor, CNSE, Nanoscience 

Title: Superconducting Optoelectronic Neuromorphic Computing 

Abstract: 

• Review of superconducting approaches to neuromorphic computing, the differentiating aspects of these concepts - advantages and disadvantages. 
• Overview of AFRL-funded research into Superconducting Optoelectronic Neuromorphic Computing conducted by Papa Rao & Nate Cady over the past 3 years 
• Presentation of plans for the near future - and potential points of collaboration with U. Albany/SUNY Poly faculty 

Presenter 4:  Yuchi Young, Associate Professor, Health Policy, Management, and Behavior+ 

Title: Developing a wearable technology to reduce work-related injury among direct care workers 

Abstract: Work-related injuries among direct care workers (DCWs) in long-term care settings are common. About 70-80% of the 5,000,000 DCWs are home health aides, personal care aides, and certified nursing aids who provide the majority of long-term care to persons with disabilities on daily tasks (e.g., bathing).  These daily tasks require repetitive heavy lifting and transferring, often leading to work-related injuries. OSHA estimated the cost associated with workers’ compensation alone to be more than $50 billion annually, yet, work-related injuries are potentially preventable.  

Our feasibility study (n=7) found (1) risky patient transfer behavior was common in the assisted living facility, and (2) this behavior could be adequately detected via wearable motion’ tracking sensors. 

The proposal aims to develop a wearable technology for injury prevention (WeTIP), which provides immediate feedback on improper lifting and transferring practices as part of the work safety training for DCWs. 

Presenter 5:  Dola Saha, Assistant Professor, Electrical & Computer Engineering+ 

Title: AI for Wireless 

Abstract: Wireless signals have helped connect people across the globe, communicate beyond the Earth, sense signals originating from outer space or the Earth for understanding the Universe or our world through the window of radio frequency (RF). Traditional signal processing based models rely on simplistic channels to form a closed form representation, whereas emerging deep learning based models use neural networks as black box. This talk explores various ways of introducing domain knowledge in the neural network (NN) models to make it tractable, bounded and explainable. This talk will discuss ongoing projects on deep learning for wireless transmitter and receiver design for emerging Terahertz frequencies, waveform generation, interference cancellation and distributed learning. 

Presenter 1:  Scott Tenenbaum, Head of Nanobioscioence, Nanobioscience+ 

Title: Creating AI Genetic Algorithms for Designing Complex RNA Switches 

Abstract: We develop and use genetic algorithms for engineering 2-piece RNA switches that are grounded in classic Darwinian evolution principles.  Using this approach we sample enormous genomic sequence space to identify candidate sequences and structures for testing at the bench.  Winners are developed as potential mRNA therapeutics. 

Presenter 2:  Pradeep Atrey, Associate Professor, Computer Science* 

Title: Detecting and Characterizing Multi-rich Disinformation on Social Media 

Abstract: Falsifying multimedia asset is a type of social hacking designed to change a reader’s point of view, the effect of which may lead them to make misinformed decisions. This work focuses on detecting disinformation content on media-rich social media platforms, such as Facebook and Twitter. The goal is to design, develop, and evaluate novel and transformative tools and techniques to not only detect online media-rich disinformation but also analyze its causal impact, leveraging expertise from AI and other fields. The work will advance the current state of knowledge in determining the impact of and detecting and mitigating cross-platform disinformation containing multiple types of media (text, images, and videos) spread over social media. Moreover, the disinformation theories and models developed through the collaboration of the complementary fields will be transformed into a fully functional software application, which will be delivered for public use. 

Presenter 3:  Bryan Sotherden, Manager of Information Systems and Programming, Professional Development Program (PDP)* 

Title: Experimenting With Chatbots to Handle Routine User Support Requests 

Abstract: The Professional Development Program provides application development and hosts several high traffic public facing sites providing a wide range of services resulting in numerous repetitive user support requests. PDP is piloting using chatbots to improve customer service providing quicker replies while also removing staff from basic repetitive user questions and requests.  

Freshdesk, the ticket system PDP currently uses offers AI powered chatbots. This bot was trained using human generated support email replies from actual user questions. The content of these emails was modified to question and answer format to train the AI powered bot.  

While still early, results are favorable. Future plans include using APIs to further integrate the bot. For example, PDP supports NYS childcare providers looking to open a facility, resulting in thousands of requests to open new facilities. API integration will enable the bot to query our database, and instantly provide users with their daycare petition status." 

Presenter 4:  Igor Lednev, Distinguished Professor, Chemistry* 

Title: Machine learning for the analysis of spectral data as a universal tool for probing (bio)chemical composition of complex samples 

Abstract: The application of vibrational spectroscopy combined with machine learning for forensic purposes and medical diagnostics will be discussed. 

Presenter 5:  Janet Paluh, Associate Professor Nanobioscience, Nanobioscience+ 

Title: Neural network architectures and AI methods for optimizing drug treatments and biomedical imaging diagnostics in healthcare 

Abstract: We are developing neural network architectures and AI methods to expand a currently finite set of diagnostic methods in healthcare for applications in optimizing drug treatments, that includes multi-drug therapies, the design and execution of clinical trials, as well as applications for biomedical imaging. Implicit in the software design is the ability to be compatible with the supercomputing AI environment, and eventually embed the software into custom low-power hardware for an improved -better and faster- user experience. Our current software solutions such as SyNC, SyNC-iN, and DendrAItic focus on artificial synapses and neural communication with applications relevant to cognitive and psychiatric disorders. CNN architectures are also being developed and optimized via enabling tools to benefit disease diagnostics in MRI biomedical imaging. The ability to bridge the gap of biomedicine with AI requires teams with diverse and overlapping skill sets that can enable in-silicon experimentation and quickly advance real outcomes.