Friday, April 11, 2025 | 8:30 AM – 5:00 PM
Location:
STEW 279 – Keynote & Research Talks
STEW 310 – Student Poster Session
STEW 206 – Catering
No Registration Required – Open to All!
Join us for a full-day workshop designed to bring together faculty, staff, postdocs, students, and researchers from across Purdue. Applied AI Day will feature keynote talks from leading AI experts, research presentations by Purdue faculty, a panel discussion, and a student poster session.
Schedule of Events
| Time | Event | Location |
|---|---|---|
| 8:30 AM – 9:00 AM | Breakfast | STEW 206 |
| 9:00 AM – 9:15 AM | Welcoming Remark - Dr. Baijian “Justin” Yang, Associate Dean for Research, Polytechnic Institute | STEW 279 |
| 9:15 AM – 10:15 AM | Keynote Talk I – Dr. Navdeep Jaitly, Apple Machine Learning Research | STEW 279 |
| 10:15 AM – 10:45 AM | Research Talk I – Dr. Rua Mae Williams, CGT | STEW 279 |
| 10:45 AM – 11:00 AM | Break | |
| 11:00 AM – 11:30 PM | Research Talk II – Dr. Kyubyung Kang, SCMT | STEW 279 |
| 11:30 AM – 12:00 PM | Research Talk III – Dr. Xingyu Li, SOET | STEW 279 |
| 12:00 PM – 1:00 PM | Lunch | STEW 206 |
| 1:00 PM – 2:30 PM | Student Poster Session | STEW 310 |
| 2:30 PM – 2:45 PM | Break | |
| 2:45 PM – 3:45 PM | Keynote Talk II – Iris Pan, Amazon | STEW 279 |
| 3:45 PM – 4:15 PM | Research Talk IV – Dr. Gaurav Nanda, SOET | STEW 279 |
| 4:15 PM – 4:45 PM | Panel Discussion – Dr. Navdeep Jaitly, Apple Machine Learning Research; Iris Pan, Amazon; Dr. Rua Mae Williams, CGT; Dr. Kyubyung Kang, SCMT; Dr. Gaurav Nanda, SOET; Dr. Mustafa Abdallah, CIT | STEW 279 |
| 4:45 PM – 5:00 PM | Farewell | STEW 279 |
Keynote Speakers
Dr. Navdeep Jaitly
Research Scientist, Apple Machine Learning Research
Explorations in Generative Modeling without tokenization
Talk Abstract: Recent successes in machine learning have been driven in a large part through a successful, standardized recipe. In this recipe data is ‘tokenized' into discrete sequences, and language modeling techniques are applied to model these discretized sequences. However, this leaves open the possibility that tokenization may compress away important information that is needed for modeling. Even more importantly, it leaves open the question of what is needed to build models that can capture the properties of the raw data directly. In this talk we will show how examples of how tokenization maybe hurt the capabilities of these models. We will also describe our recent work on multimodal generative modeling - including generative models of images and audio, conditioned on text, models of audio conditioned on text and video, and joint models of text and audio - which build generative models of raw data directly. We show that these recent techniques can lead to powerful models of raw data directly without relying on additional models for tokenizing the inputs.
Bio: Navdeep Jaitly is a Research Scientist at Apple Machine Learning Research (MLR) where he leads a team of researchers working on fundamental techniques for Machine Learning. He got his PhD from University of Toronto under the supervision of Geoffrey Hinton in the foundational days of Deep Learning. During a PhD internship at Google in 2011, he demonstrated how Deep Learning could revolutionize speech recognition and this work was a part of a 2012 paper which received the test of time best award paper from IEEE Signal Processing Magazine in 2022. After his PhD he joined Google Brain, working on sequences models, introducing methods such as Listen Attend and Spell, Adversarial Autoencoders and Pointer-Networks. He has also held machine learning research positions at Nvidia, Google Brain Robotics, D. E. Shaw and the National Labs.
Iris Pan
Lead Designer, XR/AI, Amazon
Designing AI-Powered Shopping Experiences: Challenges, Opportunities, and the Future of Creativity
Talk Abstract: AI is transforming the way we shop—and the way we design. As a Lead Designer at Amazon, I work on AI-driven shopping experiences like View in Your Room (AR home shopping) and Rufus (AI conversational commerce). In this talk, I’ll explore the challenges and opportunities of designing AI-powered shopping, our collaboration with applied scientists, and how AI is reshaping creativity. Join me for a discussion on the evolving role of designers in intelligent commerce.
Bio: Xiaobi (Iris) Pan is a globally-recognized pioneer in the fields of Emerging Technology Design. She has been the inventor of the first sound wave interactive installation visited by over 1 million participants, the app that promises to plant 100 million trees on our planet by 2025, and a series of industry-leading new technology products and experiences.
Currently a Lead Designer at Amazon, she is dedicated to using technologies like Augmented Reality and Artificial Intelligence to revolutionize the consumer experience for both business and social good.
Outside of work, she likes to write, make music, and saber to her custom songs in virtual reality. Her writings on design, technology and their social implications have influenced over 10 million readers worldwide.
To learn more about her and her works,
https://www.linkedin.com/in/irisxpan/
https://irispan.com/
https://www.zhihu.com/people/iris.pan/
Research Talk Speakers
Dr. Rua Mae Williams
Assistant Professor, Computer Graphics Technology
AI's Eugenics Problem and What to do Next
Talk Abstract: Many scholars of science and technology history have noted the influences of the eugenics movement on statistics, computing, and machine learning. This historical traces of eugenic ideologies in applied AI are important for understanding "unintended" human impacts and consequences of decision-making support systems. However, many AI practitioners do not feel these critiques apply to them, as they find the premise absurd or believe that their work is immune to or stands apart from these influences. This presentation attempts to make the legacies of eugenics plain for AI practitioners, to explain how eugenic thought contaminates applied AI systems, and what AI practitioners can do to uproot these logics. We will cover real-world examples of eugenic logics driving AI design decisions and propose alternative approaches to Applied AI that center self-determination for marginalized people most at risk from AI harms.
Bio: Rua M. Williams is an assistant professor in User Experience Design at Purdue University and a former Just Tech Fellow (2022-2024) with the Social Science Research Council. As Principal Investigator of the CoLiberation Lab, Dr. Williams’s work explores how disabled people imagine and build their own sociotechnical worlds, and investigate how technology policy and research practice interact to disrupt disabled people’s bodily autonomy and access to meaningful public life.
.jpg)
Dr. Kyubyung Kang
Assistant Professor, School of Construction Management Technology
Beyond the Blueprint: AI-Powered, Low-Cost Data for Transforming Construction and Infrastructure Management
Talk Abstract: Traditional construction monitoring relies on manual inspections, which are time-consuming, error-prone, and inefficient for large-scale projects. While BIM and computer vision offer solutions, real-time integration with low-cost data sources like CCTV footage is limited, leaving critical information fragmented. The main challenge is that most AI methods are designed for rich datasets, yet real-world construction environments often lack high-quality data or find it too expensive to collect. This presentation explores how to extract meaningful insights from low-cost, noisy data and adapt AI to function effectively in constrained conditions. An AI-powered, low-cost monitoring framework will integrate CCTV feeds with BIM for real-time progress tracking and defect detection. Deep learning models will extract geometric information from low-quality video data, dynamically updating BIM models. A cloud-based system will process and synchronize visual and BIM datasets, enabling seamless, automated monitoring. By leveraging affordable, scalable technologies, this approach provides a novel framework for construction monitoring. Expected outcomes include a validated AI-driven system, an optimized deep learning model for low-resolution video analysis, and improved synchronization of visual and BIM data. These advancements in digital twin systems offer a cost-effective solution for enhancing construction management and infrastructure resilience in an AI-driven era.
Bio: Kyubyung Kang, Ph.D. is an Assistant Professor of Construction Management Technology at Purdue University and the President of the Korean-American Scientists and Engineers Association (KSEA) Indiana Chapter. His research focuses on leveraging AI-driven computer vision, digital twins, and automation to enhance the resilience and efficiency of infrastructure systems. With a background in civil engineering and construction management, Dr. Kang has led multiple funded research projects integrating AI, BIM, and data-driven decision-making to optimize construction monitoring, maintenance, and management. He actively collaborates with industry and government agencies, including the Indiana Department of Transportation (INDOT), to develop scalable, cost-effective solutions for infrastructure challenges. His work advances AI applications in construction, particularly in utilizing low-cost data for real-time defect detection, progress tracking, and asset management. Dr. Kang is also committed to education and workforce development, integrating AI-based digital twin applications into his courses to prepare the next generation of construction and infrastructure professionals.
.jpg)
Dr. Xingyu Li
Assistant Professor, School of Engineering Technology
A Large Manufacturing Decision Model for Human-Centric Decision-Making
Talk Abstract: To adapt to changing demands and disruptions, manufacturing systems necessitate dynamic reconfiguration, facilitated by growing digitalization, modularity, and autonomy. Such reconfiguration, however, heightens decision-making complexity and the need for human supervision. While Generative AI (GenAI), particularly large language models (LLMs), fosters natural human-resource interactions, existing methods lack manufacturing-specific context. This paper introduces a Large Manufacturing Decision Model (LMDM) leveraging image generative models to precisely represent and generate manufacturing-specific reconfiguration decisions using a digital twin, minimizing data requirements and reducing hallucination risks. Simulation results showcase LMDM’s ability to refine system configurations through human guidance, transforming digital twins into human-centric decision-making tools.
Bio: Dr. Xingyu Li is currently an Assistant Professor in the School of Engineering Technology at Purdue University, West Lafayette. Before joining Purdue, he was an Adjunct Assistant Research Scientist at the Department of Mechanical Engineering at University of Michigan - Ann Arbor. Dr. Li received his Ph.D. degree in Mechanical Engineering from the University of Michigan – Ann Arbor in 2018. His research interests include smart manufacturing systems, supply chain management, deep learning, artificial intelligence, and optimization. Dr. Li is currently a CIRP Research Affiliate, an ASTAR Visiting Fellow, an AnalytiXIN Fellow, and a corresponding expert in Engineering, he is also a recipient of the Best Reviewer of OMEGA in 2023, Journal of Manufacturing Systems Outstanding Reviewer, Best Paper Award at the 2019 IEEE Ai4i, Ford COVID-19 Innovation Challenge Award and Presidents Health and Safety Award.
Dr. Gaurav Nanda
Assistant Professor, School of Engineering Technology
AI and HCI Applications for Improving Decision Support in Safety
Talk Abstract: Unintentional injuries account for 3.16 million lives lost annually and cause many more cases of serious incapacitations. Injury surveillance efforts aim to understand the causes and circumstances leading to these injuries and act as an effective guiding mechanism for designing preventive interventions such as updated safety policies, and product recalls. Public health agencies collect and analyze injury data from different sources such as hospital emergency departments, workplace injury reports, and health surveys. The raw injury data collected is in unstructured form of incident narratives which are transformed into structured form through “injury coding” which involves assigning different types of injury codes to each case such as cause-of-injury, body part injured, and product involved. This structured data is subsequently analyzed to understand the circumstances leading to different types of injuries and prioritize prevention efforts. Manual injury coding and analysis is time and resource consuming. AI based approaches have been used for injury data analysis but have been observed to have limited effectiveness for rare and complicated injury cases. This talk will discuss various AI and HCI based approaches to improve accuracy and decision support aspects of machine learning based injury data analysis and injury prevention practices.
Bio: Dr. Gaurav Nanda is an Assistant Professor in the School of Engineering Technology at Purdue University with focus in Industrial Engineering Technology. He obtained his Ph.D. in Industrial Engineering from Purdue University and his Bachelors and Masters from Indian Institute of Technology (IIT) Kharagpur, India. He has worked as a postdoctoral researcher at Purdue University for two years and in the software industry for five years. His research interests include text mining, safety, human factors, and intelligent decision support systems leveraging complementary strengths of humans and AI models. He teaches courses on Human Factors, Operations Management, Manufacturing Information Systems, and Intelligent Manufacturing. He is affiliated as a core faculty member with the Applied AI Research Center at Purdue University.