Lightning Talks (11:15AM-1PM) Guoyu Lu (School of Electrical and Computer Engineering) "3D Structure Modeling and Assessment for Crops and Beyond" Crop 3D structures and traits are crucial for understanding plant phenotyping and optimizing crop growth. However, constructing accurate 3D models for complex plant roots is challenging. To address this, we propose an unsupervised learning scheme using mobile phone images to build precise 3D root structure models. To avoid destructing roots, we also developed deep convolutional neural networks for GPR signal detection and graph neural networks for root shape reconstruction, enabling non-destructive root analysis. In addition to underground root reconstruction, our work leverages unmanned aerial systems (UAS) to provide a comprehensive view of crop growth. By deploying a cost-effective and precise 3D sensing system on UAS, we empower agricultural researchers, growers, and service providers to assess crop growth accurately and identify potential risks. Our approach utilizes specialized AI systems for large-scale, fine-scale 3D crop structure modeling. Furthermore, our methods extend beyond agriculture, as we demonstrate their applicability to other mobile robot applications such as autonomous driving and nighttime navigation. This showcases the versatility and potential impact of our 3D modeling techniques. Ari Schlesinger (School of Computing): “Working Towards Human-Centered AI” Abstract: Concerns about the impacts of AI are on the rise. The tech community has a responsibility to design and build AI ecosystems that benefit our diverse society. But this is easier said than done. Despite our best efforts, power imbalances, bias, and discrimination continue to impact AI systems. We need to understand and address the roots of AI's discrimination problems in order to build more human-centered AI systems. In this talk, I will discuss early-stage work focused on supporting developers and organizations in creating more equitable AI systems. He Li, School of Chemical (Materials and Biomedical Engineering): “Physics-informed machine learning for infectious disease forecasting” Abstract: As witnessed in the Coronavirus disease 2019 (COVID-19) pandemic, accurate forecasting of the spread of contagious illnesses has become increasingly important to public health policymaking and could prevent loss of millions of lives. To better prepare for the future pandemic, it is essential for epidemiologists and biomedical engineers to continuously improve the capabilities and accuracy of the disease forecasting models through introducing novel computational methods. In this work, we propose to employ physics-informed neural networks (PINNs), one of the most popular models in the emerging area of scientific machine learning, to boost the capability of predicting the spread of infectious diseases. Since the original work was published in 2019, PINNs has been cited more than 4000 times and led to hundreds of follow-up applications crossing the areas of mathematics, physics, chemistry, engineering, economics and so on. Although several attempts have been made to implement PINNs in infectious diseases forecasting, these studies have shown inconsistent model performance. The objective of this work is to systematically investigate the performance of PINNs model on assimilating epidemiological data and mechanistic models, aiming to develop specific strategies that can optimize the design of PINNs model for enhanced forecasting in infectious diseases. Gerald Kane (Terry College of Business): "Avoiding an Oppressive Future of Machine Learning: A Design Theory for Emancipatory Assistants" Abstract: Widespread use of machine learning (ML) systems could result in an oppressive future of ubiquitous monitoring and behavior control that, for dialogic purposes, we call “Informania.” This dystopian future results from ML systems’ inherent design based on training data rather than built with code. To avoid this oppressive future, we develop the concept of an emancipatory assistant (EA), an ML system that engages with human users to help them understand and enact emancipatory outcomes amidst the oppressive environment of Informania. Using emancipatory pedagogy as a kernel theory, we develop two sets of design principles: one for the near future and the other for the far-term future. Designers optimize EA on emancipatory outcomes for an individual user, which protects the user from Informania’s oppression by engaging in an adversarial relationship with its oppressive ML platforms when necessary. The principles should encourage IS researchers to enlarge the range of possibilities for responding to the influx of ML systems. Given the fusion of social and technical expertise that IS research embodies, we encourage other IS researchers to theorize boldly about the long-term consequences of emerging technologies on society and potentially change their trajectory. Soheyla Amirian (School of Computing): “AI-Powered Healthcare: A Computational Journey with the UGA AMIIE Lab” Abstract: Artificial Intelligence and Healthcare Informatics transform the landscape of medical research and patient care. In this quick talk, we will delve into the innovative research and solutions emerging from the UGA AMIIE Laboratory, showcasing our medical imaging informatics projects. This presentation will offer an overview of the AMIIE lab's mission, methodologies, and recent achievements, highlighting the potential of AI to address critical challenges in the digital healthcare domain. Neal Outland (Department of Psychology): "Harmonizing AI Integration in the US Economy: Bridging Diverse Stakeholder Perspectives for Seamless Transition" Abstract: In the rapidly evolving landscape of Artificial Intelligence (AI), its integration into the United States economy presents both unprecedented opportunities and complex challenges. This presentation delves into the intricacies of AI integration, focusing on the divergent preferences and expectations of various stakeholders, including CEOs, industry leaders, and the general population. The primary objective is to explore pathways towards a seamless and frictionless integration of AI technologies across different sectors of the economy. One of the critical areas of discussion is the disparity in AI adoption preferences between high-level decision-makers and the broader public. CEOs and industry leaders often prioritize efficiency, innovation, and competitive advantage, while the general population shows a strong inclination towards job security, ethical considerations, and the socio-economic impacts of AI. This dichotomy poses a significant challenge in creating a cohesive strategy for AI integration that satisfies all parties. To address these challenges, the presentation proposes the development of a common language and understanding across the various fields involved in AI research and development. This involves fostering interdisciplinary collaboration, standardizing AI terminologies, and ensuring that AI education and awareness are accessible to all layers of society. By establishing this common ground, it becomes possible to align the diverse expectations and priorities of different stakeholders, paving the way for a more harmonious integration of AI into the economy. Furthermore, the presentation will highlight the importance of regulatory frameworks, ethical guidelines, and public-private partnerships in facilitating a smooth transition. It will also touch upon the need for continuous dialogue and feedback mechanisms to adapt to the dynamic nature of AI and its impact on the economy. In conclusion, the presentation aims to provide a comprehensive overview of the strategies and approaches necessary to achieve a seamless integration of AI in the US economy, emphasizing the importance of balancing innovation with inclusivity and ethical considerations. Hongyue Sun (Environmental, Civil, Agricultural and Mechanical Engineering): “Data Science Enabled Decision-making in Advanced Manufacturing and Personalized Safety” Abstract: The advancements of sensing and information technology have brought significant opportunities to engineering systems, where data containing rich streaming and heterogenous information are collected in manufacturing and occupational environments. However, there is a lack of systematic methodologies to address these high dimensional, streaming, and heterogenous information and support engineering systems decision-making. In this talk, I will present data science enabled decision-making to address the above challenge in advanced manufacturing and personalized safety. Tianming Liu (School of Computing): “When Brain-Inspired AI Meets Artificial General Intelligence (AGI)” Abstract: ChatGPT and GPT-4 have demonstrated remarkable performance in many natural language processing (NLP), reasoning, content generation, and multimodal tasks, particularly, in zero-shot learning settings. In some sense, ChatGPT/GPT-4 already exhibits human intelligence traits, or artificial general intelligence (AGI). In this talk, I will share my understanding of core technologies in ChatGPT/GPT-4 including foundation models, large language models (LLM), in-context learning, prompt engineering, reinforcement learning from human feedback, and multimodal integration, and in particular, I will offer brain science perspectives to these innovative methodologies and their applications in medicine, healthcare, pharmacy, and public health. Lakshmish Ramaswamy (School of Computing): "Robust Environmental AI for Urban and Coastal Sustainability"
