Commonsense reasoning in computer vision encompasses integrating visual data and contextual knowledge, crucial for enhancing AI's understanding of everyday scenarios. This understanding not only improves machine learning models but also enhances their ability to interact meaningfully with humans and the environment. Unlike CNN-based conventional vision models, which are designed to identify objects within a specific image, incorporating commonsense knowledge enables models to interpret scenes in a more holistic manner, thereby improving their spatial ability to reason about relationships among objects and actions. This integration not only enhances object recognition but also facilitates a deeper understanding of the contextual factors, ultimately leading to more precise predictions and interactions in real-world applications. This paper presents a comprehensive survey of recent developments that integrate commonsense knowledge into computer vision tasks. We systematically review approaches based on knowledge graphs, scene graphs, neuro-symbolic models, and commonsense-augmented transformers. We also outline current limitations related to dataset bias, knowledge incompleteness, and integration challenges. Finally, we highlight prospective research trajectories in cross-modal reasoning, scalable commonsense knowledge injection, and neuro-symbolic hybrid architectures to develop truly intelligent visual systems.

Members of the Western Michigan Transformative Interdisciplinary Human+AI Research Group have been engaged in two consecutive NSF-funded projects to promote AI readiness in diverse STEM disciplines. Putting equal emphasis on theory and practice, our goal is to instill knowledge and competency in safe, secure, and reliable AI across a wide range of learners from high school students through to university students and practitioners who wish to upskill. The second project that is currently underway has a specific focus on machine-assisted processing of massive data. This presentation focuses on the development of immersive learning experiences.

The groundbreaking invention of ChatGPT has triggered enormous discussion among users across all fields and domains. Among celebration around its various advantages, questions have been raised with regards to its correctness and ethics of its use. Efforts are already underway towards capturing user sentiments around it. But it begs the question as to how the research community is analyzing ChatGPT with regards to various aspects of its usage. It is this sentiment of the researchers that we analyze in our work. Since Aspect-Based Sentiment Analysis has usually only been applied on a few datasets, it gives limited success and that too only on short text data. We propose a methodology that uses Explainable AI to facilitate such analysis on research data. Our technique presents valuable insights into extending the state of the art of Aspect-Based Sentiment Analysis on newer datasets, where such analysis is not hampered by the length of the text data.

As more “Big-Tech” companies are developing IoT devices for the consumer, many users and professionals alike, have qualms about the direction that IoT devices are taking. Issues such as how IoT devices can protect user’s privacy while optimizing for efficiency and modularity have arose. This paper addresses these worries by developing a Smart Mailbox system to address the shortcoming of current IoT technologies. This paper also shows that numerous functionalities can be easily added to existing devices to ease our daily lives. For mailboxes, these enhancements could be to identify whether the mailbox is open or closed, whether a small package or a letter was delivered in the mail, whether the mail inside the box is subjected to high heat, etc. Our proposed design uses alternative networks, such as LoRaWan, which provides benefits and assurances to the end-user such as decentralization and security. While we specifically focused on creating a Smart Mailbox for the average consumer, the technologies used in this paper can be very quickly utilized for other IoT products and specialized for enterprises’ needs. When producing our design, we experimented with different types of distance-detecting sensors to determine how to produce the most accurate results that would be resistant to external factors such as temperature. Our results indicated that for the case of a Smart Mailbox, a combination of a temperature sensor and an ultrasonic sensor created the most consistent results. Finally, we found that the most optimal placement for the sensor would be on the mailbox’s door itself.

Information systems are increasingly using artificial intelligence (AI). However, AI can be tricked into misbehaving, showing bias, or committing abuse. The root causes of these errors and uncertainties can be hidden away while parallelizing AI algorithms on high-performance computing (HPC) infrastructure. The project outlined in this paper aims to use artificial intelligence from the ground up to generate teaching materials and curricula for student-teachers. Students embark on a journey of discovery, taking calculated risks in a learning environment. The main purpose of this document is to present the primary research results of the two-year pilot project. A secondary purpose of this paper is to disseminate information about this exciting endeavor to encourage like-minded educators and researchers to participate in this project.

In this paper, insights and data analyses are presented for tracking the use of phrases in the representation of key domain areas in scientific publications, over time. A domain refers to a particular branch of scientific knowledge and hence largely defines the main topic or theme of any scientific research paper. These domains can be extracted from scientific publications over a fixed time period. Varied phrases can then be analyzed whether they are representative of the same domain. The representative phrases are then analyzed as to whether they are trending in usage over time or are being phased out. Thus, insights are proposed for a domain to be qualified by its most trending representative phrase. These insights are supported by rigorous data analyses.

The Resource-defined fitness sharing (RFS) algorithm combines resource and fitness sharing techniques to solve exact cover problems. These problems are defined in terms of resources and cooperating species which attempt to solve the problem via co-evolution. For the testbed, Sudoku is used due to the scalability and complexity of the problem. However, two primary factors limit the performance study of the algorithm. First, the algorithms serial implementation betrays its natural parallel structures. Data analysis is also tedious due the limited number of large Sudoku puzzles and the time required to assess the algorithms performance. This research presents several solutions aimed at decreasing the compute time of RFS, streamlining testing, and automating the data analysis process. The computation time of RFS is addressed by exploiting the natural parallel structures inherent between RFS and the Sudoku algorithm. Testing and data analysis automation is streamlined by incorporating a user interface (UI). The UI accepts a Sudoku puzzle and provides an easy-to-read view of the current solution state of the RFS algorithm. Data analysis is interleaved with the UI to show the performance of the algorithm on the current puzzle as well as tracking the performance history of all puzzles introduced to the algorithm. These puzzles are separated by size to address the combinatoric increase in complexity to solve a given puzzle. Once computed on the backend, the UI populates the respective performance graphs.

Artificial Intelligence (AI) has experienced a strong revival recently. From autonomous vehicles to smart factories, AI is increasingly being used in STEM fields beyond computer science (CS). While AI is regularly taught in CS curricula, treatment of AI varies in other STEM disciplines. This paper describes an on-going project aimed at elevating AI knowledge and skills of non-CS STEM students and professionals. It emphasizes trustworthiness as a key factor of effective AI usage for large scale data analysis. Specifically, ten initial modules, which take an experiential learning and problem-solving pedagogical approach, have been developed and are now being piloted in fall 2021. They have been designed to be used both in a traditional classroom setting and as self-guided learning aid. This paper reports on the findings to date and aims to disseminate this exciting venture broadly for like-minded researchers to consider.

Artificial intelligence (AI) is increasingly applied to IT systems. However, AI can be manipulated to perform undesirably, exhibit biases or abusive behaviors. When AI algorithms are parallelized on high-performance computing-based cyberinfrastructure (CI), such misbehaviors and uncertainty can multiply to obscure the root causes. Secure, safe, and reliable computing techniques can mitigate these problems. The project described in this paper aims to inform curriculum and develop materials to educate students who use AI from the outset, so that they will first become aware of the issues and secondly practical considerations will be integrated with theory in classes. Intensive, multi-faceted, modular, experiential learning units are designed to rapidly upgrade the skills of current and future CI users, so they can apply new skills to their tasks. The loosely coupled modules can be taken as standalone self-directed units or integrated into existing classes, starting with CS 1 and CS 2, which are taken by many non-CS STEM students. In a sandpit environment, learners take measured risks when guided on a journey of discovery. The primary purpose of this paper is to present key findings of the research following a 2-year pilot. A secondary purpose of the paper is to disseminate this exciting endeavor broadly, so that likeminded educators and researchers can consider participating in the project.

Machine commonsense reasoning (MCR) systems can significantly improve the way we interact with machines. MCR systems are therefore an important element in any human-centric applications. Recent advances in machine learning (ML) have enabled breakthroughs in MCR technologies. This paper aims to improve healthcare outcomes by making human-machine interactions more intuitive than before. It presents learning models developed for MCR. Specifically, it presents a critical analysis of state-of-the-art deep learning (DL) models for MCR. These include recurrent neural network (RNN), transfer learning (TL), and transformers. Transformers, in particular, have been found to be effective for a range of natural language processing (NLP) applications, including MCR. Based on the analysis, another contribution of this paper is to assemble useful MCR tools into an adaptable MCR toolbox. To ensure broad applicability, the toolbox can be customizable for different MCR applications. Our research focuses on two specific MCR applications: commonsense validation and commonsense explanation. The former concerns identifying statements that do not make commonsense. The latter aims at explaining the reason why a given statement does not make commonsense. The paper presents some preliminary results of applying elements of the assembled toolbox to the two MCR applications. These results indicate that it is possible to achieve near human performances using finely-tuned state-of-the-art DL methods for the two MCR applications.

Artificial intelligence (AI) is increasingly applied to IT systems. However, AI can be manipulated to perform undesirably, exhibit biases or abusive behaviors. When AI algorithms are parallelized on high-performance computing-based cyberinfrastructure (CI), such misbehaviors and uncertainty can multiply to obscure the root causes. Secure, safe, and reliable computing techniques can mitigate these problems. The project described in this paper aims to inform curriculum and develop materials to educate students who use AI from the outset, so that they will first become aware of the issues and secondly practical considerations will be integrated with theory in classes. Intensive, multi-faceted, modular, experiential learning units are designed to rapidly upgrade the skills of current and future CI users, so they can apply new skills to their tasks. The loosely coupled modules can be taken as standalone self-directed units or integrated into existing classes, starting with CS 1 and CS 2, which are taken by many non-CS STEM students. In a sandpit environment, learners take measured risks when guided on a journey of discovery. The primary purpose of this paper is to present key findings of the research following a 2-year pilot. A secondary purpose of the paper is to disseminate this exciting endeavor broadly, so that likeminded educators and researchers can consider participating in the project.

Computational natural language processing (NLP) is indispensable in a humanized ambience intelligence environment. NLP facilitates ambient intelligence by making machines understand, and be understood by, humans. This in turn makes machines behave more human-like than they typically are today. Technological advances in machine learning (ML), and especially deep learning (DL), have been a key enabler of NLP research. This paper begins with a survey of recent developments of ML/DL for NLP. It then identifies some of the most promising techniques reported in recent literature. These most promising techniques are then assembled into a reusable toolkit for computational NLP. The adaptable nature of the assembled toolkit allows it to be reused in a broad range of NLP applications. The paper then describes experimental evaluation of our implemented solutions for comparative analysis. Two specific NLP applications form the basis of comparative evaluation. The first involves identifying one of M English sentences that does not make sense. The second, which is harder than the first, involves choosing from among N sentences the one that best explains why a presented sentence is invalid. Human baseline accuracies for these applications are 99.1% and 97.8%, respectively. The observation that these results are somewhat less than perfect demonstrates that even humans can occasionally find these tasks difficult. It further underscores the difficulties involved in some of these computational NLP applications. Experiments conducted on benchmark data show that advanced ML/DL can achieve near-human performance in both computational NLP applications with accuracy scores of 96.1% and 93.7%, respectively.

Western Michigan University, together with public and private partners, have developed ten learning units that promote artificial intelligence (AI) competency across STEM disciplines. The learning units are modular, experiential, customizable, and fun to use. They have been developed for both traditional classroom and self-directed learning. The modules are loosely coupled, so that learners can choose different pathways and modes of usage to suit. With an emphasis on “learn by doing”, the modules are experiential and can be customized for different STEM disciplines. The primary aim of the proposed workshop is to provide participants with hands-on experience of the learning units for themselves and go through a guided journey of discovery that is fun and engaging. An integral part of this dissemination effort involves a “train the trainers” component, encourages participants to share their experiences among others in their workplaces, etc., thereby creating a multiplier effect.

Although artificial intelligence (AI) promises to deliver ever more user-friendly consumer applications, recent mishaps involving fake information and biased treatment serve as vivid reminders of the pitfalls of AI. AI can harbor latent biases and flaws that can cause harm in diverse and unexpected ways. Before AI becomes interwoven into human society, it is important to understand how and when AI can fail. This article presents a timely survey of AI-induced mishaps that relate to consumer applications. The article also offers suggestions on mitigating strategies to manage the undesirable side effects of using AI for consumer applications. It, therefore, serves a dual purpose of creating awareness of current issues and encouraging other researchers in the consumer technology community to build better AI consumer applications.

Algorithms for ensemble methods (EM) based on bootstrap aggregation often perform copious amount of redundant computations (RC) thus limiting their practicality. Given this constraint, we propose a framework that views these algorithms as a collection of computational units (cu), a tightly coupled set of both mathematical operations and data. This view facilitates a reduction in RC (RRC), thereby allowing for faster execution plans. Inspired by the floor tiling approach in VLSI, we look to engineer solutions for RRC while possibly reconfiguring the underlying computing system's compiler technology stack. We start by showing that under the assumption that the computational system has unbounded but finite memory (i.e., the memory is large enough to hold all intermediate values) and that each cu has a uniform cost, our approach reduces to a well-studied directed bandwidth problem for the directed acyclic graphs (DAGs). Next, we consider a more realistic scenario where the computing system has limited memory and concurrent execution while still assuming a uniform cost. Using a new notion of (r,s) set cover of a DAG (nodes representing computational units and edges representing their interdependencies) we formulate the problem of reducing redundant computational steps in EM as a variation of a directed bandwidth problem. We show that the graph's minimum bandwidth is closely related to memory requirements for studying RRC. Finally, our preliminary experimental results are supportive of the proposed approach for RRC and promising that it can be applied to a broader set of algorithms in decision sciences.

The pervasive nature of the Internet of Things has resulted in generating a huge amount of data about the lives of IoT users. This data includes Personally Identifiable Information (PII) that reflects people's behaviors, interests, lifestyles, and everyday routines. Protecting PII from privacy violations is a challenge since IoT data need to be handled by public networks, servers, and clouds, which are untrusted parties for data owners. In this paper, a solution called Policy Enforcement Fog Module (PEFM) is proposed for protecting sensitive IoT data whenever they are accessed throughout their entire lifecycle. PEFM uses the power of policy enforcement in the edge-fog infrastructures for protecting data accessed within users’ local domains. For data that need to be sent to remote domains, PEFM uses Active Data Bundle (ADB); an executable and self-protecting construct that can run on any visited host and enforces privacy policies automatically for data accessed by these hosts. To test the feasibility of PEFM in realistic IoT systems, a framework of using PEFM as a privacy control for Foscam home security system is simulated. The experimental results show that PEFM assures data privacy via data minimization due to selective data disclosures. Better privacy controls with minimal overhead can be achieved if most PEFM processes are executed by the local fog nodes. Migrating parts of PEFM processes to remote fog nodes or the cloud incurs more overhead than using strictly local fog nodes. This overhead is the cost for a higher level of privacy regarding lifecycle data protection.

Future buildings will offer new convenience, comfort, and efficiency possibilities to their residents. Changes will occur to the way people live as technology involves into people's lives and information processing is fully integrated into their daily living activities and objects. The future expectation of smart buildings includes making the residents' experience as easy and comfortable as possible. The massive streaming data generated and captured by smart building appliances and devices contains valuable information that needs to be mined to facilitate timely actions and better decision making. Machine learning and big data analytics will undoubtedly play a critical role to enable the delivery of such smart services. In this paper, we survey the area of smart building with a special focus on the role of techniques from machine learning and big data analytics. This survey also reviews the current trends and challenges faced in the development of smart building services.

AbstractResearch into risky decision making (RDM) has become a multidisciplinary effort. Conversations cut across fields such as psychology, economics, insurance, and marketing. This broad interest highlights the necessity for collaborative investigation of RDM to understand and manipulate the situations within which it manifests. A holistic understanding of RDM has been impeded by the independent development of diverse RDM research methodologies across different fields. There is no software specific to RDM that combines paradigms and analytical tools based on recent developments in high-performance computing technologies. This paper presents a toolkit called RDMTk, developed specifically for the study of risky decision making. RDMTk provides a free environment that can be used to manage globally-based experiments while fostering collaborative research. The incorporation of machine learning and high-performance computing (HPC) technologies in the toolkit further open additional possibilities such as scalable algorithms and big data problems arising from global scale experiments.

•A method for manifold approximation where the low dimensional space is a PCA model with the mean and principal vectors modeled as smooth functions of a parameter that depends on the position on the manifold.•Generalizations where the manifold dimension is not constant.•Generalization where the dimensionality of the ambient space is not constant.•Comparison with PCA, Sparse PCA, and independent PCA models across the manifold, for simulated data, faces in the presence of in plane rotation and faces with different out of plane rotations.

SummaryTarget‐Decoy database is currently the method of choice to assess the quality of Proteins' search engines. Decoy versions of real peptides are generated and injected to the same database of real ones with different labels. Quality of search engines results is assessed based on the number of decoys retrieved as hits. In Crux‐Tide search engine, which is one of the fastest search engines currently available, the process of indexing and generating decoys is computationally expensive. In this paper, we analyze the serial algorithm in detail and show improvement possibilities, and then describe a parallel‐shared memory solution using OpenMP. To completely break up the dependency in the serial algorithms, a clever hashing technique is utilized to localize the process. The parallel solution and the hashing technique together are able to reduce the computation cost by approximately 70‐80% using few threads. Besides the parallelization, we redesign part of the serial code so that the memory consumption becomes more efficient. The parallel version can index the same files using around two‐third of the memory space that the serial version consumes. This solution could impact and support future distributed developments of Crux‐Tide searching phase, where each parallel unit could rank the observed spectra independently.