Despite the growing prominence of Artificial Intelligence (AI) chatbots used in education, there remains a significant gap in our understanding of how interface design elements, particularly avatar representations, influence learning experiences. This paper explores the impact of different AI chatbot avatar representations on students’ learning experiences through a mixed-methods within-subjects study, where participants interacted with three distinct types of AI chatbot interfaces with a common large language model (LLM) over a 14-week university course. Our findings reveal that preferences vary according to factors such as learning habits and learning activities. Avatar design also exhibits affordances for specific prompting behaviors, while the perceived human touch influenced learning experiences in nuanced ways. Additionally, real-world relationships with the individuals behind deepfakes influence these experiences. These insights suggest that the thoughtful integration of diverse avatar representations in AI chatbot systems for different learners and settings can greatly enhance learning experiences.

This workshop is part of the ACM Multimedia 2025 Conference and is organized by the ACM I2M Chapter, consisting of both industry and academia members. The rapid convergence of Artificial Intelligence (AI), Human-Computer Interaction (HCI), and immersive multimedia is redefining the landscape of intelligent and adaptive digital experiences. As ACM Multimedia 2025 emphasizes cutting-edge multimedia systems, this workshop directly contributes to its vision by exploring AI's transformative role in immersive media. Through AI-driven multimedia interaction, adaptive virtual environments, and intelligent content generation, this workshop will showcase how AI is enhancing the creation and experience of digital worlds. The workshop proceedings can be found at: https://dl.acm.org/doi/proceedings/10.1145/3728487

This innovative practice full paper investigates the impact of using a Question Answering (QA) tutor to enhance an adaptive online chemistry course, specifically Chem Quest (CQ), at Singapore Institute of Technology. The study aims to examine the relationship between QA-immediacy and affective learning, cognitive learning, and motivation in an online learning environment. Two research questions are addressed: (1) whether online students who experience instructor-immediacy or QA-immediacy through synchronous communication have improved learning outcomes compared to a control group, and (2) whether QA-immediacy explains significant variance in student affective learning, cognitive learning, and motivation in online classes. The study hypothesizes that students who experience instructor-immediacy through synchronous communication via the QA tutor will demonstrate enhanced learning outcomes compared to those who do not, and that QA-immediacy will contribute to a significant variance in student affective learning, cognitive learning, and motivation in online classes. Quantitative results show that though there is no significant difference between the average CQ pre-test and post-test scores of students who engage with the QA tutor versus those who do not, there is an observed difference in terms of motivation, affective and cognitive learning. Students who interact with the QA tutor have a more positive attitude towards CQ. These same students also have a lower learning loss and have an overall more positive impression of the CQ course. These results show the promising impact that QA tutors can have on student learning.

Reverberation is a physical phenomena caused by room impulse reflection from walls, ceilings and other plane surfaces. When occurring in speech, it contributes an additive echo to the speech which is proportional to the geometry of the environment in which that speech was recorded. Reverberation is not just noise though, as the echo timing in speech affects its intelligibility as well as the listener's interpretation of the playback environment. It gives listeners a sense of the size, shape and materials of the room it is in. For communications scenarios where playback reverberation differs significantly from the visual display (e.g. speech recorded in a small room but played back in a cavernous virtual reality (VR) space), a psychoacoustic dissonance arises between visual and auditory perception. This paper presents an AI-based algorithm to remove source room reverberation impulse effects from speech. We then add reverberation to match the target playback environment. To assess effectiveness, we simulate a set of recording room geometries, and perform psychoacoustic assessment to estimate the effect on a human listener.

This paper presents an approach to object detection in cartoon images. This method is a critical component of a newly developed educational platform aimed at enhancing mother tongue language usage among Primary 1 and Primary 2 students. The platform's visual engine incorporates image analytics of cartoon imagery to support language learning. While most existing object detection models are optimized for real-world images, our work addresses the underexplored area of cartoon image analysis. By fine-tuning pre-existing models, our method demonstrates performance improvements in detecting objects within cartoon images. This research contributes to the broader goal of integrating visual learning tools into early childhood education, with the goal of fostering improved language skills through engaging and interactive content.

Preparing students well for the Information Communication Technology (ICT) industry is challenging as the needs of the industry constantly evolve. As society becomes more receptive to technological solutions, their appetite for more complex and impactful solutions increases too. To cope with this changing demand and better prepare students entering the workforce, we propose a blended learning method with real-life medium-level complexity industry projects and industry clients in a Software Engineering specialization course that emphasizes applied learning with authentic assessment. The interaction with real-life industry clients with real problem statements exposes students to the uncertainty of software requirement gatherings and the design process of medium-level complex software engineering projects. This paper reports on the experience of these proposed approaches and students' self-perceived responses to the implementation. Our study results show significant improvement in students' self-perceived rating for preparedness for real-world industry problems after completing the project and an increase in the rating for most lifelong learning skills.

The demand for accessible telerehabilitation services in Singapore has surged, driven by an aging population and the COVID-19 pandemic. Traditional physiotherapy often entails lengthy and costly sessions. Exercising at home, while more convenient, requires vigilance to prevent further injury. To address this, we have developed a mobile application that enables users to effectively manage chronic pain through bitesize-at-home-physiotherapy exercises. The mobile app offers a personalized exercise program curated by an in-house physiotherapist coupled with an AI motion tracking algorithm to provide instantaneous corrective feedback while a user is doing their exercise. A pilot initial study assessed usability, effectiveness, accuracy, and user experience, demonstrating strong performance and high user satisfaction. This app meets the growing need for affordable and convenient telerehabilitation in Singapore, empowering users to manage chronic pain effectively and safely at home. The pilot study's positive results underscore its potential to significantly improve the lives of those with chronic pain.

Accurate and automatic power line anomaly detection is critical to smart grid. However, effective solutions are yet available due to the insufficiency of anomaly data. In this paper, we first collect a dataset from various sources consisting both normal and abnormal power line images. With this dataset, anomaly detection becomes feasible though with limited accuracy due to the limited size of the dataset. As such, we propose TransLine, an approach based on transfer learning to apply the existing knowledge extracted from large-scale datasets to complement the data insufficiency of power line anomaly detection. TransLine customizes and optimizes the knowledge to automate the power line anomaly detection with high accuracy. The experiment results show that TransLine can achieve superb accuracy of 96.1% on average and up to 98.1% given only a hundred abnormal images for model training. TransLine can be a key enabler of smart grid for great stability and efficiency and can inspire the other industrial applications facing data insufficiency issues.

This Innovative Practice Full Paper presents the development and implementation of an innovative online adaptive gamification platform for learning Computational Thinking (CT). CT is an essential problem-solving skill set in this modern era of digitization and technological advancements. To build students' knowledge and skills in CT while maximizing students' motivation and engagement in learning, a novel online adaptive gamified course called Computational Thinking Quest (CTQ) was introduced. The CTQ was designed and developed by a multidisciplinary team of students and faculty members. The key features of CTQ are (1) an interactive storyline with animated avatars, mini-games, and questions created using Unity three-dimensional cross-platform game engine and Blockly block-based visual programming language; (2) questions at three different levels of difficulty for effective adaptive and self-learning approach; (3) an answer and feedback to each question for increased students' confidence and enthusiasm towards learning; (4) hyperlinks to online learning resources for further reading; (5) a badge and a leaderboard to motivate active participation and encourage success; and (6) a course management system with automatic data saving capability to enable learning at own pace, anytime, anywhere. The CTQ was rolled out to some newly matriculated first-year undergraduate Engineering (ENG) and InfoComm Technology (ICT) students. A total of 54 ENG and 53 ICT students' learning performance and feedback were collated and analyzed. Statistical results from the paired Student's t-test and the Wilcoxon signed rank test consistently reveal that (1) the median of post-test marks is significantly higher than that of pre-test marks (p < 0.001); (2) the median of CT knowledge scores after taking CTQ is significantly higher than that of before CTQ (p < 0.001); and (3) the median time taken to complete the post-test is significantly lower than the pre-test (p < 0.001). Furthermore, more than 75% of ENG and ICT students, separately, stated that CTQ is an engaging or very engaging learning platform; 81% of them indicated that the educational content of CTQ is enriching or very enriching; and 83% of them commented that CTQ has motivated independent learning. The CTQ can also serve as a bridging course to narrow the heterogeneity gap among students with heterogeneous prior knowledge on computer programming languages and ease students into programming-related modules, thereby enhancing teaching and learning effectiveness.

In this paper, we develop a novel framework for action recognition in videos. The framework is based on automatically learning the discriminative trajectory groups that are relevant to an action. Different from previous approaches, our method does not require complex computation for graph matching or complex latent models to localize the parts. We model a video as a structured bag of trajectory groups with latent class variables. We model action recognition problem in a weakly supervised setting and learn discriminative trajectory groups by employing multiple instance learning (MIL) based Support Vector Machine (SVM) using pre-computed kernels. The kernels depend on the spatio-temporal relationship between the extracted trajectory groups and their associated features. We demonstrate both quantitatively and qualitatively that the classification performance of our proposed method is superior to baselines and several state-of-the-art approaches on three challenging standard benchmark datasets.

We address the problem of modeling and classifying American Football offense \nteams’ plays in video, a challenging example of group activity analysis. Automatic play \nclassification will allow coaches to infer patterns and tendencies of opponents more ef- \nficiently, resulting in better strategy planning in a game. We define a football play as a \nunique combination of player trajectories. To this end, we develop a framework that uses \nplayer trajectories as inputs to MedLDA, a supervised topic model. The joint maximiza- \ntion of both likelihood and inter-class margins of MedLDA in learning the topics allows \nus to learn semantically meaningful play type templates, as well as, classify different \nplay types with 70% average accuracy. Furthermore, this method is extended to analyze \nindividual player roles in classifying each play type. We validate our method on a large \ndataset comprising 271 play clips from real-world football games, which will be made \npublicly available for future comparisons.

Craniofacial disorders commonly result in various head shape dysmorphologies. The goal of this work is to quantify the various 3D shape variations that manifest in the different facial abnormalities in individuals with a craniofacial disorder called 22q11.2 Deletion Syndrome. Genetic programming (GP) is used to learn the different 3D shape quantifications. Experimental results show that the GP method achieves a higher classification rate than those of human experts and existing computer algorithms [1], [2].

Retrieval of 3D models have attracted much research interest, and many types of shape features have been proposed. In this paper, we describe a novel approach of combining the feature types for 3D model retrieval and relevance feedback processing.Our approach performs query processing using pre-computed pairwise distances between objects measured according to various feature types. Experimental tests show that this approach performs better than retrieval by individual feature type.

Recent advancement in 3D digitization techniques have prompted the need for 3D object retrieval. Our method of comparing 3D objects for retrieval is based on 3D morphing. It computes, for each 3D object, two spatial feature maps that describe the geometry and topology of the surface patches on the object, while preserving the spatial information of the patches in the maps. The feature maps capture the amount of effort required to morph a 3D object into a canonical sphere, without performing explicit 3D morphing. Fourier transforms of the feature maps are used for object comparison so as to achieve invariant retrieval under arbitrary rotation, reflection, and non-uniform scaling o the objects. Experimental results show that our method of retrieving 3D models is very accurate, achieving a precision of above .086 even at a recall rate of 1.0.

We present a prototype system, Predator-Miner, which extends Predator with an relational-like association rule mining operator to support data mining operations. Predator-Miner allows a user to combine association rule mining queries with SQL queries. This approach towards tight integration differs from existing techniques of using user-defined functions (UDFs), stored procedures, or re-expressing a mining query as several SQL queries in two aspects. First, by encapsulating the task of association rule mining in a relational operator, we allow association rule mining to be considered as part of the query plan, on which query optimization can be performed on the mining query holistically. Second, by integrating it as a relational operator, we can leverage on the mature field of relational database technology. We extend Predator to support a variant of DMQL, and allow SQL and DMQL to be intermixed in a query. We also demonstrate a cost-based mining query optimization framework.

3D digitization devices produce very large sets of 3D points sampled from the surfaces of the objects being scanned. A mesh construction procedure needs to be applied to derive polygon mesh from the 3D point sets. As the 3D points derived from digitization devices based on digital imaging technologies are inherently non-uniformly distributed over regions that may contain surface discontinuities, existing methods are not suitable for polygonizing them. This paper describes a novel polygonization algorithm for constructing triangle mesh from unorganized 3D points. In contrast to existing methods, this algorithm begins the mesh construction process from 3D points lying on smooth surfaces, and advances the mesh frontier towards 3D points lying near surface discontinuities. If 3D points along the edges and at the corners are sampled, then the algorithm will form an edge where two advancing frontiers meet, and a corner where three or more frontiers meet. Otherwise, the algorithm constructs approximations of the edges and corners. It can be shown that this frontier advancing algorithm performs 2D Delaunay triangulation of 3D points lying on a plane in 3D space.

This paper describes a novel method for constructing triangle meshes from noisy and non-uniformly distributed 3D points. It consists of two steps: noise point removal uses a clustering algorithm and epipolar constraint method to identify and remove noise points from the 3D points; and a constructive polygonization algorithm interpolates cleaned 3D points to construct a triangle mesh of the object.