Artificial Intelligence

Artificial intelligence (AI) has the potential to generate transformative solutions that enhance human life and societal well-being in Maine and beyond. Through innovative technologies and applications, AI can help Maine industries improve their operations and compete in the global economy .

Goal 1: Research Objective
Develop transformative AI-based solutions that enhance the social and economic well-being of the citizens of Maine and beyond. Priorities include making AI more efficient, ethical, and secure.
Goal 2: Enterprise Objective
Increase the number of Maine businesses and organizations using AI solutions to improve their products and operations.
Goal 3: Workforce Objective
Provide training and expertise to researchers and practitioners throughout Maine whose work could benefit from AI, and incorporate AI training into existing postsecondary programs in related fields.
Goal 4: Climate Change Objective
Use AI to help advance climate-smart practices and policies in agriculture, aquaculture, forestry, fisheries, clean energy, and related fields, and research ways to reduce AI’s large carbon footprint.

With the advancement in computing technology and change in market demand, computing systems are now employed for performing complex and involved tasks. Developing algorithms to solve complex problems is not always easy; hence, techniques were developed to have the algorithm itself learn from observing the raw data. These learning techniques and other rule-based techniques for solving complex problems gave rise to a new computing paradigm called Artificial Intelligence (AI). Today in the U.S. and worldwide, AI techniques are employed in diverse domains such as:

Home Automation: AI-based automation techniques are used to optimize power utilization in different appliances around the house and automated home security systems routinely rely on AI techniques to detect anomalies and predict future consumption.
Intelligent Transport System: From traffic control algorithms to self-driving cars, many transport system applications use some form of AI to improve efficiency, control device behaviors and make decisions that have social and economic consequences.
Environmental Protection: Pollution prediction systems and wildlife monitoring techniques use AI to improve accuracy and coverage.
Cybersecurity: AI-based methodologies are increasingly used for predicting system vulnerabilities, detecting fraud, and detecting intrusions.
Medicine: AI is widely used in the medical domain to monitor patients, develop pharmaceutical products, and remote robotic surgeries and other tasks .
Precision Agriculture: Drone-based crop monitoring, satellite image-based yield estimation, and automated farming systems are all using AI.
Industrial Automation: AI is also used in industrial settings to automate the pipeline, operate warehouse robots, and perform quality control.
Retail Automation: Smart carts and automatic item detection techniques are also being introduced into the retail world with the help of AI techniques.

AI has penetrated nearly all U.S. markets and AI applications are only predicted to grow.

Challenges Facing Artificial Intelligence Growth

Artificial Intelligence has come a long way from simple rule-based expert systems to deep neural networks. AI is now widely used in diverse cyber- physical systems (CPS) for solving real-life problems. Although the ability to solve complex problems using AI techniques has increased dramatically in recent years, there are several existing challenges to implementing reliable and accurate AI systems:

Ethical Issues and Bias: AI models are susceptible to undesirable bias if proper training methodologies are not followed and guardrails are not implemented. Also, relying on certain data features for generating a prediction can lead to unfair or unjust outcomes.
Inefficient Implementation: Techniques to implement AI in real systems are far from perfect. Model accuracy in lab settings must translate to the real world to make an impact. Standards must also be developed for these techniques to ensure reliability.
Lack of Data Privacy and Security: Data mining, processing, and collection of data for training AI models is a topic of strong public interest and debate. While compliance standards are increasingly adopted and enforced, there are additional steps that need to be taken to ensure high privacy and security standards are adhered to .

Inefficient Data Storage and Access: AI systems require the storage and rapid access of a very high volume of data for different purposes. Innovations are required to ensure efficient data management in connected systems and the cloud .
High Carbon Footprint: Training/using AI models often requires a high amount of computation, which indirectly leads to negative environmental outcomes .

Solving these core issues in AI can lead to better AI- driven systems and a wider appeal.

The Roux Institute at Northeastern University

In 2020, Northeastern University launched the Roux Institute in Portland, Maine, an ambitious initiative to grow talent in AI and other advanced technologies. The institute, made possible by the vision and philanthropy of David and Barbara Roux, is partnering with Maine companies to advance workforce skills through graduate education and research opportunities. It seeks to transform Portland into a hub for innovation.

Davis Institute for Artificial Intelligence at Colby College

Established in 2021, the Davis Institute for Artificial Intelligence at Colby College explores applications of AI and machine learning grounded in the liberal arts. Its target outcomes include both education and enterprise development.

University of Maine Artificial Intelligence Initiative (UMaine AI)

The UMaine AI initiative, founded to promote AI research activities in the university, and has more than 40 affiliated faculty members (seven steering committee members). UMaine AI strives to create strong collaborative ties with other universities, government, industry, and broader community. This initiative has also led to different outreach activities, including monthly education webinars attracting over 1,200 participants.

Key Past & Current AI-Related Research Activities at the University of Maine

The University of Maine has fostered, developed, and perfected different technologies involving diverse AI techniques. Some of the research groups working in the area of AI and cyber-physical systems are listed below:

MaineSAIL: The Software Agents and Artificial Intelligence Laboratory (led by professor Roy M. Turner) at UMaine focuses on developing multiagent systems empowered with AI. Several interesting innovations such as ORCA, CODA, and ACRO were made possible through the involvement of this research team.
Advanced Structures and Composites Center: With more than 200 patents and 700+ publications, this research group (led by professor Habib Dagher) focuses on developing innovative technologies in diverse domains, such as material sciences, advanced manufacturing, and composites. Such technologies also involve the development of cyber-physical systems such as the DeepCLiDAR.
VEMI Lab: This lab was co-founded by Richard Corey and professor Nicholas Giudice, who lead the lab with professor Caitlin Howell. This research team specializes in developing human- computer interaction (HCI) technologies.
Sekeh Lab: Professor Sekeh and her team at UMaine contribute to diverse areas of AI. This group has worked toward developing novel AI algorithms, improving deep learning model efficiency, securing AI models, and many other innovations .
WiSe Net Lab: Established in 2005, this lab, led by professor Ali Abedi, has engineered and innovated several technologies in the areas of wireless sensor networks and space applications. AI and smart sensing play an important role in many of these technologies.
MIM Lab: The Multisensory Interactive Media Lab focuses on developing innovative technologies in the domains of virtual reality and augmented reality, with AI playing an important role in such innovations .
SKAI Lab: The Spatial Knowledge and Artificial Intelligence Lab, led by professor Torsten Hahmann, innovates in diverse areas of AI such as formal space representations, automated ontology modularization, and integration of high-level knowledge with low-level data.
Climate Change Institute: The hub of climate change research at the University of Maine. Some of the technologies employ cyber physical systems and AI for climate monitoring, climate prediction, and information transfer.

The University of Maine has a long history of innovating in diverse areas of AI. It has led research initiatives to advance the theoretical understanding of AI and machine learning while also developing innovative technologies, utilizing AI, that can make a real difference in the world.

Future Research Objectives

The University of Maine will leverage existing resources and past research experiences to solve different core problems of AI, apply AI to solve novel real-world problems, and continue to innovate in the space of cyber-physical systems. The following are some of the suggested research activities in the areas of AI and cyber-physical systems:

Green AI: Innovation is required to make AI training, inferencing, and architecture search more energy efficient. Future works should also focus on efficient implementations of AI in cyber- physical systems with greenhouse gas emissions in mind .
Ethical AI: Although some work has been done in this domain, scientists are far from creating truly unbiased (in terms of irrelevant features) AI models. More research is required in this domain to make AI more trustworthy in the future.
Toward Efficient Hardware Implementation: Efficient implementation of AI models in real systems is a challenging task due to real-world constraints that may have been overlooked in a lab setting. Research is required to bring real system implementation constraints into the loop while developing AI models.
Secure AI: The security of the data being processed by an AI model and the security of the AI model itself is crucial to building trust among the user communities. UMaine has past contributions in this area and will continue to drive research innovations to solve these issues .
Efficient Data Management for AI: Efficient data handling in connected devices (edge) has emerged as a new challenge in recent years . Innovations in this area are required to make AI- driven connected systems more efficient in the future.
New Frontiers: University of Maine ECE has hired two new faculty members (Dr. Tonkoski and Dr. Chakraborty) for creating potential research thrusts in the areas of applied AI for cybersecurity and sustainable energy.

References

Turner, Roy M. “Intelligent control of autonomous underwater vehicles: The Orca project.” 1995 IEEE International Conference on Systems, Man and Cybernetics. Intelligent Systems for the 21st Century. Vol. 2. IEEE, 1995. DOI: 10.1109/ICSMC.1995.538022

MaineSAIL ORCA, http://mainesail.umcs.maine.edu/ MaineSAIL/projects/orca

ASCC DeepCLiDAR, https://composites.umaine.edu/ deepclidar

Herbert, Valerie M., et al. “Developing a smartphone app with augmented reality to support virtual learning of nursing students on heart failure.” Clinical Simulation in Nursing 54 (2021): 77-85. https://doi. org/10.1016/j.ecns.2021.02.003

Soucy, Nicholas, and Salimeh Yasaei Sekeh. “CEU-Net: Ensemble Semantic Segmentation of Hyperspectral Images Using Clustering.” arXiv preprint arXiv:2203.04873 (2022). https://doi. org/10.48550/arXiv.2203.04873

WiSe Net Lab, https://umaine.edu/wisenetlab

MaineSAIL http://mainesail.umcs.maine.edu/ MaineSAIL

Advanced Structures & Composites Center, https:// composites.umaine.edu

VEMI Lab, https://umaine.edu/vemi

Sekeh Lab, https://salimehyasaei.wixsite.com/ sekeh-lab

MIM Lab, http://www.mimlab.info

SKAI Lab, https://ai.umaine.edu/skailab

Climate Change Institute, https://climatechange.umaine .edu