Digital Commons Network^™

Frozen Lvlms For Micro-Video Recommendation: A Systematic Study Of Feature Extraction And Fusion, Huatuan Sun, Yunshan Ma, Changguang Wu, Yanxin Zhang, Pengfei Wang, Xiaoyu Du

Research Collection School Of Computing and Information Systems

Frozen Large Video Language Models (LVLMs) are increasingly employed in micro-video recommendation (MVR) due to their strong multimodal understanding. However, existing apporches typically deploy LVLMs as fixed black-box feature extractors without systematically comparing alternative representation strategies. To address this gap, we present the first systematic empirical study on various feature extraction paradigms and integration strategies, along with hierarchical representations from frozen LVLMs for MVR. Extensive experiments on representative LVLMs reveal that hidden states from multiple decoder layers provide richer and more effective representations for MVR. Guided by this insight, we propose the Dual Feature Fusion (DFF) Framework, a lightweight approach …

“Grandpa, Can You Speak Nicer?”: Envisioned Chatbot Roles And Design Tensions In Intergenerational Communication Conflicts, Tianyi Zhang, Emran Bin Elias Poh, Yueyue Hou, Yi-Chieh Lee, Renwen Zhang, Jiannan Li, Anthony Tang

Research Collection School Of Computing and Information Systems

Intergenerational conversations often break down when differences in tone, language, or expectations lead participants to feel dismissed or misunderstood. In this work, we explore how people envision AI-driven chatbot interventions for addressing communication problems in text-based intergenerational family chat. We conducted a scenario-based design interview with 10 pairs of family members from different generations, in which participants designed chatbot interventions that varied in intervention target and timing. Our findings show that participants expect chatbots to perform multiple themes of intervention, including mediating understanding, providing emotional support, offering evaluative commentary, and guiding interaction through behavioral suggestions. These expectations varied systematically across …

Group Conversational Agents: A Review Of Designs That Support And Shape Group Interaction, Shunyi Yeo, Tianyi Zhang, Scott Bateman, Gary Hsieh, Young-Ho Kim, Simon Tangi Perrault, Jiannan Li, Anthony Tang

Research Collection School Of Computing and Information Systems

Conversational agents that participate in or mediate group interaction introduce challenges that extend beyond supporting individual users, raising new questions about how agents participate in and influence groups. To characterise this emerging design space, we present a systematic review of 53 peer-reviewed studies on group conversational agents (GCAs). We analyse how GCAs intervene in group-level processes, including participation regulation, conflict mediation, task alignment, and execution support. Using concepts from group research as an analytic lens, we organise prior GCA work around recurring group interactional challenges (orientation, conflict, alignment, and execution), and examine the roles agents are designed to play in …

Towards Auto-Evaluation For Large Language Models, Jiahao Ying

Dissertations and Theses Collection (Open Access)

The rapid advancement of large language models (LLMs) has created an urgent need for evaluation methodologies that are timely, scalable, reliable, and informative. Conventional evaluation benchmarks, although essential for measuring model capabilities and guiding model development, are often constructed and maintained through labor-intensive human annotation. As LLMs continue to improve through increases in model scale, training data, and computational resources, static benchmarks may quickly lose discriminative power. Moreover, the growing use of large and diverse training corpora increases the risk of benchmark leakage, which can inflate evaluation results and obscure the true capabilities of models. These challenges call for a …

Integer Quantization And Embedded Deployment Of Cnn Star Trackers For Cubesats, Meora R. Giusiano

Master's Theses

A star tracker determines spacecraft orientation by photographing the star field, detecting stars in the image, matching them against a catalog, and computing the rotation between observed and cataloged directions. Convolutional neural networks (CNNs) have been proposed as replacements for the detection and centroiding stage, offering improved sub-pixel accuracy and recovering faint stars that classical thresholds lose to stray light and sensor noise. The improvement comes at higher computational cost; the PolySat systemboard targeted in this work lacks the floating-point hardware these networks assume.

This thesis closes the gap between floating-point desktop evaluation and embedded integer deployment. Nine encoder-decoder CNN …

What Makes A Modern Attention Implementation?, Brian H. Slonim

Master's Theses

Since the seminal assertion by Vaswani et al. in 2017 that “Attention Is All You Need,” transformer models have risen to ubiquity due to their ability to learn extremely complex patterns from sequence data, culminating in the unprecedented generative capabilities of large language models. These models’ strength lies in their scale: hundreds of millions (e.g., BERT-LARGE) to billions or trillions of learned parameters. Running inference with these models, let alone training them, would be intractable without significant innovations in the hardware and software that support them. This need has driven an enormous demand for GPU compute and associated software ecosystems, …

High-Resolution Queries, Low-Resolution Context: Scaling Vision Transformers With Asymmetric Spatial Reduction, James M. Dwyer

Master's Theses

Vision Transformers (ViTs) have demonstrated great performance on image classifica tion benchmarks, however, the quadratic complexity of the self-attention mechanism with respect to sequence length limits their scalability to higher resolution inputs. The attention score matrix grows as O(N2) in both compute and memory, where N is the number of patch tokens, making ViTs computationally expensive and memory intensive for applications that require real-time inference or operate under resource constraints.

This thesis investigates whether the key and value sequences of the self-attention mechanism can be compressed using the local spatial structure of the image — while keeping queries at full …

Scorespeak: An Agentic System For Natural Language Control Of Musical Scores, Nathan S. Lim

Master's Theses

With the recent popularization of large language models (LLMs), natural language has become one of the most accessible and powerful ways for people to interact with creative tools. Although they have become common in mainstream domains like image and audio editing, there is currently no robust AI-based system that can reliably turn free-form language into edits for symbolic musical scores. This gap represents a missed opportunity to improve human workflows for creating and editing sheet music, but it is also a fundamental limitation for other agentic music systems; without a robust mechanism for translating free-form language into structured scores, AI …

Political Inconsistency Detection Across Legislative Speech And Public Communications, Scott M. Pramuk

Master's Theses

Political actors communicate about legislation across multiple contexts, including committee hearings, recorded votes, and public-facing press releases. Differences between these forms of communication can provide useful signals for journalists and researchers seeking to understand how legislators present policy positions to different audiences.

This thesis extends the Digital Democracy Project, a legislative transparency initiative that provides access to California state legislative hearing transcripts, voting records, and related legislative data. Specifically, this work incorporates publicly accessible, legislator-authored news releases into the Digital Democracy Database and develops a pipeline for analyzing legislative communication across multiple sources. The system collects news releases from California …

Multimodal Machine Learning For Soil Burn Severity Mapping Across California Wildfires, Sanjana Checker

Master's Theses

Accurate mapping of soil burn severity (SBS) is critical for post-fire watershed management, erosion risk assessment, and ecological recovery planning, yet traditional field-based approaches remain costly, time-intensive, and spatially limited. This thesis presents a machine learning pipeline for wall-to-wall SBS classification across California wildfires using multi-sensor satellite imagery, terrain derivatives, and bioclimatic covariates. Field-collected SBS observations (n = 2,180) from 52 wildfires occur- ring between 2013 and 2025, sourced from the U.S. Forest Service and CAL FIRE, were used to train and evaluate multiple classification architectures within a Google Earth Engine and Google Cloud-based prediction framework. After upsampling the unburned …

Saag: Structured Agent Assessment And Grounding, Ritvik Garimella, Vedant Khandelwal, Anvi Kohli, Amit Sheth

Publications

Exact-match evaluation of agent-calling obscures qualitatively different failure modes: a model may select the right function yet hallucinate argument values, or satisfy a schema while choosing a agent for the wrong reason. Existing benchmarks collapse these distinctions into a single binary score, leaving practitioners unable to diagnose where agent calls fail. We propose SAAG a cascaded diagnostic framework that decomposes agent-calling evaluation into three sequential stages: registry conformance, structural completeness, and argument grounding, each producing interpretable stage-specific diagnostics. These diagnostics additionally enable iterative self-repair: on prediction failure, the stage-specific signal guides targeted correction without leaking ground-truth values. We evaluate this …

Rvit-Fusionnet: A Local Cross-Attention Feature Fusion-Based Hybrid Framework For Brain Tumor Classification, Naima Islam, Sajeeb Kumar Ray, Md Anwar Hossain, Syed Mohammed Shamsul Islam

Research outputs 2022 to 2026

Accurate brain tumor classification via MRI is essential for diagnosis and treatment. This study introduces RViT-FusionNet, a hybrid deep learning model that integrates convolutional and transformer architectures for enhanced tumor detection. The model utilizes ResNet-50 to capture textural details and a Vision Transformer for extracting global context. A Local Cross-Attention (LCA) module is proposed to align and merge these features, allowing the network to model local structures and long-range dependencies concurrently. To enhance generalization across varied imaging conditions and tumor types, a domain discriminator is included to discern spatial and domain-specific patterns, fostering the learning of domain-invariant representations. The approach …

Deployment-Aware Deep Learning For Computer Vision: Efficient Architectures From 3d Segmentation To Mixed Reality, Bahar Uddin Mahmud

Dissertations

Deep learning has become the dominant approach for solving vision-centric problems; however, its successful deployment in real-world applications remains limited by high computational cost, data dependency, and insufficient integration with practical and human-centered environments. While state-of-the art deep learning models often achieve impressive performance in controlled settings, they frequently fail to generalize or operate efficiently under deployment constraints such as limited resources, complex data modalities, and real-time interaction requirements. These limitations motivate the need for a deployment-oriented deep learning framework that balances accuracy, efficiency, and practical usability.

This dissertation investigates the design and deployment of efficient deep learning architectures for …

Cfalr: Collaborative Filtering-Augmented Large Language Model For Personalized Fashion Outfit Recommendation, Yujuan Ding, Junrong Liao, Yunshan Ma, Yi Bin, Wenqi Fan, Tat-Seng Chua, Qing Li

Research Collection School Of Computing and Information Systems

Personalized outfit recommendation poses a significant challenge in e-commerce and social media platforms, requiring systems that balance user preferences with aesthetic compatibility. Collaborative filtering (CF) provides a traditional solution for this, but it struggles with data-sparse scenarios and complex user-item-outfit relationships. Meanwhile, existing template-based approaches are constrained by rigid pre-designed structures. To bridge these research gaps, we introduce CFALR (Collaborative Filtering-Augmented Large Language Model for Recommendation), a novel framework that synergizes collaborative filtering with large language models for personalized outfit recommendation. Specifically, CFALR describes user-outfit interactions in natural language and leverages LLMs to capture fashion semantics while employing CF-enhanced embeddings …

Interfold: Learning Interpretable Diffusion Manifolds Beyond Binary Samples, Alexander Vincent Lewi, Rainer Tan, Shengfeng He

Research Collection School Of Computing and Information Systems

We propose InterFold, a framework for learning and applying interpretable semantic manifolds in latent diffusion models, without requiring binary or paired supervision. Existing methods for semantic editing either rely on limited paired data or uncover only coarse, unsupervised directions that fail to capture user-specific, fine-grained attributes. InterFold addresses these limitations by learning a target attribute manifold in the H-space of diffusion models using only a set of positive, unlabeled examples. To edit a new image, InterFold projects its H-space representation toward this learned manifold through test-time optimization, enabling precise, identity-preserving modifications of complex, non-binary concepts. To make these edits effective …

Adaptive Outlier Detection Over Data Stream, Rui Zhu, Mingyuan Jiang, Xiaochun Yang, Baihua Zheng, Bin Wang, Tao Qiu

Research Collection School Of Computing and Information Systems

Continuous distance-based outlier detection in streaming data poses significant challenges and has a wide range of practical applications. Traditional threshold-based methods perform well under stable streaming conditions, where fixed parameters remain effective. However, they often struggle with dynamic data distributions and high stream speeds, leading to suboptimal performance, limited control over the number of returned outliers, and failure to meet real-time detection requirements. To address these issues, this paper introduces a novel Recall and Proportion-Aware Outlier Detection (RPA-OD) query. In RPA-OD, ρ defines a distance relaxation that enables real-time outlier detection. Specifically, objects with fewer than k neighbors within the …

A Novel Hierarchical Multi-Agent System For Payments Using Llms, Donghao Huang, Joon Kiat Chua, Zhaoxia Wang

Research Collection School Of Computing and Information Systems

Large language model (LLM) agents, such as OpenAI’s Operator and Claude’s Computer Use, can automate workflows but unable to handle payment tasks. Existing agentic solutions have gained significant attention; however, even the latest approaches face challenges in implementing end-to-end agentic payment workflows. To address this gap, this research proposes the Hierarchical Multi-Agent System for Payments (HMASP), which provides an end-to-end agentic method for completing payment workflows. The proposed HMASP leverages either open-weight or proprietary LLMs and employs a modular architecture consisting of the Conversational Payment Agent (CPA - first agent level), Supervisor agents (second agent level), Routing agents (third agent …

“Alexa, Do Not Say That In Front Of My Boss!” A Cross-Cultural Comparison Of User And Ai Preferences For Privacy-Aware Smart Speaker Interactions Across Contexts, Lynne Warin, Emily Aurelia, Anthony Tang, Emily Aurelia, Delphine Reinhardt

Research Collection School Of Computing and Information Systems

Due to their limited ability to reason about the social context in which they are used, smart speakers pose significant privacy risks by responding in ways that may violate people's implicit social boundaries. We conducted a cross-cultural vignette study (N = 944) in Germany and Singapore to investigate how situational factors—specifically social context (bystander relationships and closeness), physical context (location), and interaction context (topic and deceptive intent)—regulate user preferences for smart speaker responses. Our results demonstrate that these factors are superior predictors of response preferences than dispositional user traits (i.e., intrinsic personal traits). We identify two distinct social dynamics: a …

Scaling Up Multi-Agent Reinforcement Learning For Large Agent Teams And Long-Horizon Tasks: A Survey, Minghong Geng, Shubham Pateria, Budhitama Subagdja, Ah-Hwee Tan

Research Collection School Of Computing and Information Systems

Multi-agent reinforcement learning (MARL) empowers multiple autonomous agents to acquire effective policies for collaborative problem-solving. Over the last decade, MARL has seen significant advancements, with numerous algorithms achieving impressive performance across various benchmarks and real-world applications. Nevertheless, the scalability of multi-agent systems, in terms of the number of agents and the length of the task horizon, remains a critical consideration for applying MARL methods to complex problem-solving. Given that a dedicated review of the existing approaches and challenges in scaling up multi-agent systems remains largely absent, this survey aims to bridge this gap by delivering a comprehensive review of MARL …

Georoad-Upernet: Geo-1-Based Weakly Supervised Multispectral Road Extraction Via Role-Aware Context Fusion And Semantic Regularization, Shaoqian Chen, Yunliang Chen, Jianxin Li, Ao Yang

Research outputs 2022 to 2026

Extracting roads accurately from remote sensing images is important for map updates, traffic analysis, and infrastructure monitoring. Medium-resolution multispectral images can provide useful surface and background information, but when used alone, the spatial details are limited for retaining narrow roads, intersection structures, and fine road topologies. To address this problem, this paper proposes GeoRoad-UPerNet, a Geo-1-centered weakly supervised multispectral framework for road extraction. In this framework, Geo-1 serves as the primary 16-band multispectral source, Sentinel-2 Level-2A imagery serves as auxiliary contextual support, and OpenStreetMap (OSM) road information is converted into proxy supervision rather than dense manual ground truth. GeoRoad-UPerNet contains …

Enhancing Pointing Gestures Of Non-Hmd Users In Asymmetric Collocated Mixed Reality Collaboration, Nam-Dang Vo, Van-Vinh Thai, Anthony Tang, Khanh-Duy Le

Research Collection School Of Computing and Information Systems

A common collocated group setting in mixed-reality (MR) collaboration is a person wearing a MR headset (HMD user) and presenting MR contents to audiences who are not provided with such specialized devices (Non-HMD users). In this setting, while Non-HMD users can view the MR environment shown on a large physical display, it still remains challenging for the HMD user to interpret their pointing gesture when they spatially refer to objects in the MR environment. To address this, we designed and evaluated two pointing techniques—SCREEN and SCREEN+SPACE—that support Non-HMD users in referring to MR content. Screen pointing allows users to refer …

Ai In Healthcare: Regulatory Guidelines And Judge-Made Negligence Principles For Ai Implementers, Gary K. Y. Chan

Research Collection Yong Pung How School Of Law

The use of artificial intelligence (AI) in healthcare may, notwithstanding its potential benefits, result in harm to patients from allegedly negligent acts or omissions by hospitals and medical doctors. In such circumstances, how should the principles in the tort of negligence (duty of care, breach, causation, remoteness of damage, and defences) respond to AI innovations in healthcare? In particular, how may the standard of care expected of hospitals and medical doctors be informed by regulatory guidelines? We refer to case law precedents and regulatory guidelines on the roles and responsibilities of doctors and hospitals as AI implementers. Importantly, they prompt …

Rode: Linear Rectified Mixture Of Diverse Experts For Food Large Multi-Modal Models, Pengkun Jiao, Xinlan Wu, Bin Zhu, Jingjing Chen, Chong-Wah Ngo, Yu-Gang

Research Collection School Of Computing and Information Systems

Large Multi-modal Models (LMMs) have significantly advanced a variety of vision-language tasks. The scalability and availability of high-quality training data play a pivotal role in the success of LMMs. In the realm of food, while comprehensive food datasets such as Recipe1M offer an abundance of ingredient and recipe information, they often fall short of providing ample data for nutritional analysis. The Recipe1M+ dataset, despite offering a subset for nutritional evaluation, is limited in the scale and accuracy of nutrition information. To bridge this gap, we introduce Uni-Food, a unified food dataset that comprises over 100,000 images with various food labels, …

Benchmarking Gaslighting Negation Attacks Against Multimodal Large Language Models, Bin Zhu, Yinxuan Gui, Huiyan Qi, Jingjing Chen, Chong-Wah Ngo, Ee-Peng Lim

Research Collection School Of Computing and Information Systems

Multimodal Large Language Models (MLLMs) have exhibited remarkable advancements in integrating different modalities, excelling in complex understanding and generation tasks. Despite their success, MLLMs remain vulnerable to conversational adversarial inputs. In this paper, we systematically study gaslighting negation attacks—a phenomenon where models, despite initially providing correct answers, are persuaded by user-provided negations to reverse their outputs, often fabricating justifications. We conduct extensive evaluations of state-of-the-art MLLMs across diverse benchmarks and observe substantial performance drops when negation is introduced. Notably, we introduce the first benchmark GaslightingBench, specifically designed to evaluate the vulnerability of MLLMs to negation arguments. GaslightingBench consists of multiple-choice …

Language Embeddings Meet Shallow Autoencoders, Rodrigo Alves, Vojtěch Vančura, Pavel Kordík, Antoine Ledent

Research Collection School Of Computing and Information Systems

Shallow autoencoders are appealing recommenders due to their simplicity, scalability, and competitive retrieval quality, but they struggle in strict cold-start settings where new items have no interactions. We propose an inductive shallow autoencoder that leverages item side information (language embeddings) by fixing the decoder to item features and learning only an encoder in the same semantic space. To prevent trivial self-reconstruction without enforcing a hard zero diagonal, we introduce diagonal gating: a leave-one-item-out objective that blocks the self-copy shortcut only for the item being updated while retaining context from the rest of the user history. An alternating-style optimization trains the …

Rc-Nf: Robot-Conditioned Normalizing Flow For Real-Time Anomaly Detection In Robotic Manipulation, Shijie Zhou, Bin Zhu, Jiarui Yang, Xiangyu Zhao, Jingjing Chen, Yu-Gang Jiang

Research Collection School Of Computing and Information Systems

Recent advances in Vision-Language-Action (VLA) models have enabled robots to execute increasingly complex tasks. However, VLA models trained through imitation learning struggle to operate reliably in dynamic environments and often fail under Out-of-Distribution (OOD) conditions. To address this issue, we propose Robot-Conditioned Normalizing Flow(RC-NF), a real-time monitoring model for robotic anomaly detection and intervention that ensures the robot's state and the object's motion trajectory align with the task. RC-NF decouples the processing of task-aware robot and object states within the normalizing flow. It requires only positive samples for unsupervised training and calculates accurate robotic anomaly scores during inference through the …

Anatomical Domain Shifts: Test-Time Heterogeneous Adaptation For 3d Human Pose Prediction, Qiongjie Cui, Pan Zhou, Jingjing Chen, Na Zhao

Research Collection School Of Computing and Information Systems

The research frontier in human pose prediction (HPP) is advancing toward continual test-time adaptation (TTA), where models must self-adapt to dynamic test distributions. To date, the homeostatic continual TTA remains the sole viable solution, which isolates the model parameters and update domain-sensitive ones. Despite mitigating full-body domain gaps, human anatomical heterogeneity (domain shifts often localize to specific regions) is ignored. This anatomical-agnostic approach forces uniform parameter adaptation across kinematically distinct segments, causing: over-adaptation of stable regions and under-adaptation of shift-prone articulations. To address it, we introduce TT-HA, a novel Test-Time Heterogeneous Adaptation that implicitly estimates domain changes for anatomical segments, …

History To Future: Evolving Agent With Experience And Thought For Zero-Shot Vision-And-Language Navigation, Guangzhao Dai, Shuo Wang, Zihan Wang, Guo-Sen Xie, Yang Yang, Jinshan Pan, Qianru Sun, Xiangbo Shu

Research Collection School Of Computing and Information Systems

Vision-and-Language Navigation in Continuous Environment (VLN-CE) requires an agent to follow language instructions to navigate the target destination. With the advancement of large language models (LLMs), recent efforts have explored adapting them for zero-shot VLN-CE, offering a promising solution in addressing the drawbacks of poor generalization in the training-based paradigm. However, existing LLM-based works primarily perform naive reasoning for decision-making and lack feedback, e.g., reviewing historical errors and predicting future potentials. Consequently, it may suffer from continuous failure for those initial error tasks. In this paper, we rethink LLM-based zero-shot VLN-CE and propose a new paradigm, named EvoNav, to improve …

Task Complexity Matters: An Empirical Study Of Reasoning In Llms For Sentiment Analysis, Donghao Huang, Zhaoxia Wang

Research Collection School Of Computing and Information Systems

Large language models (LLMs) with reasoning capabilities have fueled a compelling narrative that reasoning universally improves performance across language tasks. We test this claim through a comprehensive evaluation of 504 configurations across seven model families—including adaptive, conditional, and reinforcement learning-based reasoning architectures—on sentiment analysis datasets of varying granularity (binary, five-class, and 27-class emotion). Our findings reveal that reasoning effectiveness is strongly task-dependent, challenging prevailing assumptions: (1) Reasoning shows task-complexity dependence—binary classification degrades up to -19.9 F1% points (pp), while 27-class emotion recognition gains up to  +16.0 pp; (2) Distilled reasoning variants underperform base models by 3–18 pp on simpler tasks, …

How Do Machine Learning Models Change?, Joel Castaño, Rafael Cabañas, Antonio Salmerón, David Lo, Silverio Martínez-Fernández

Research Collection School Of Computing and Information Systems

The proliferation of Machine Learning (ML) models and their open source implementations has transformed AI research and applications. Platforms like Hugging Face (HF) enable this evolving ecosystem, yet a large-scale longitudinal study of how these models change is lacking. This study addresses this gap by analyzing over 680,000 commits from 100,000 models and 2,251 releases from 202 of these models on HF using repository mining and longitudinal methods. We apply an extended ML change taxonomy to classify commits and use Bayesian networks to model temporal patterns in commit and release activities. Our findings show that commit activities align with established …