22505 research outputs found
Sort by
An Investigation of Active Inference for Reinforcement Learning Control
Get PDFThis thesis investigates the application of active inference framework on different reinforcement learning (RL) tasks. We specifically consider the following OpenAI Gym environments: CartPole, MountainCar, and LunarLander. In this thesis, our primary goal is to explore whether active inference could provide better performance and stability compared to traditional RL methods. The proposed model consisted of three main components: a variational autoencoder (VAE) model to infer hidden states, a transition model predicting latent states, and a Double deep Q-network (Double DQN) as the actor selecting optimal actions. To achieve this, extensive experiments are carried out using grid searches across several hyperparameters, including learning rate, discount factor gamma, KL-divergence weights and soft update factor tau. Models achieving stable and rapid convergence across multiple trials were selected as optimal. Custom reward shaping techniques were implemented for more challenging environments such as MountainCar and LunarLander. The experimental results demonstrated that while the active inference agent successfully achieved the desired performance thresholds in each environment, its performance was not stable, often increasing early before subsequently decreasing. This behavior suggested issues related to catastrophic forgetting where the agent might implicitly treat different state regions as separate tasks, continuously overwriting previously beneficial policy parameters. Elastic weight consolidation (EWC) was explored to solve the instability issue. However, incorporating EWC yielded limited improvement, suggesting that the instability could originate from factors beyond traditional catastrophic forgetting. These results indicate that active inference, combined with Double DQN, is capable of effectively solving standard RL tasks. However, challenges remain in terms of policy stability. Therefore, it is important to conduct further research to understand and overcome these instabilities as it potentially deliver great utilities to solving more complex tasks with active inference
Global Trail: Interactive Travel Memory System for Unesco Sites
Get PDFGlobal Trail is an interactive mobile app designed for travellers or culture seekers to reimagine how they discover new places, engage with cultural sites and remember their travel experiences. In this digital world, photos get easily lost in phone galleries and sometimes lack an organized way to find specific memories to places visited. Through design, the app experience addresses three core objectives: Discover — users explore Unesco sites represented by badges through an interactive world map; Engage — they unlock and earn those badges when they visit those sites where they can post their memories on a community platform; and Remember — the app uses AI to select specific photos and videos from each sites visited and create a memory page tied to each location. Inspired by gamified learning, logging travel experiences and social media patterns, Global Trail balances fun and functionality. It makes memory organization simpler, gamification through badges rewarding, and connections through a community platform meaningful. Through research, user surveys and prototyping, this project aimed at designing a user-friendly experience combining emotion, education and technology together. Global Trail redefines travel as more than just moving to places but as the act of discovering, engaging and remembering. This thesis serves as a design exploration of how digital tools can enhance location-based memories, encourage engagement, and foster a sense of shared experience for both travelers and culture seekers
Transparent Display Design for Vending Machines
Get PDFThis thesis explores the innovative application of transparent display technology in vending machines, addressing the limitations of traditional vending interfaces in consumer engagement. By integrating comic book and manga-inspired aesthetics with cyberpunk visual effects, the project transforms vending machines into dynamic, interactive canvases. These designs incorporate transparent overlays, comic panel UI layouts, and glitch-inspired animations to enhance user interaction and visibility of products. This thesis critically evaluates the design process, methodologies, and outcomes while highlighting research findings that inform the integration of emerging display technologies with bold visual storytelling. The final solution presents a seamless fusion of futuristic visual elements with traditional vending machine functionality, offering a transformative experience for users
Enhancing APTV Reliability: A Study of Design and Operational Efficiencies at An Aluminium Manufacturing Company
Get PDFThis paper aims to focus on enhancing the reliability of Anode Pallet Transport Vehicles (APTVs) used in Aluminium Manufacturing Company. The primary objective of the project is to establish the causes of failure modes, assess the operational problems, and implement strategies that would prevent, reduce, or minimize the frequency of failure, low equipment utilization, and high maintenance costs. APTV performance analysis involves the use of maintenance logs, operational data, and engineering specifications to obtain the best results. Applying Failure Mode and Effects Analysis (FMEA), the study highlights the most critical parts causing the largest effect on operations, which include traction motors, leaf springs, and tires. Moreover, unlike conventional brainstorming methods, tools such as RCA Fishbone, and 5 Whys are integrated to identify deep-rooted problems spanning organizational, mechanical, and environmental systems. The project also takes a closer look into the efficiency of the fleet with time utilization and cycle time analysis, where time gap can be identified. The study indicates that design changes, proactive maintenance, and other related management activities are effective ways of improving vehicle reliability. The project identifies the potential failure modes and provides a specific set of engineering solutions to reduce risk while providing guidelines on the frequency of maintenance. Overall, this report offers realistic solutions that would help enhance the reliability and operational effectiveness of APTV fleet across Aluminium Manufacturing Company, whether in the short term or for the long-term strategy formulating
Structure/Property/Processing of Covalent Adaptable Networks Bearing Hindered Urea Bonds
Get PDFDynamic covalent chemistry provides the functional basis for the efficient and reliable exchange of reversible binding groups which enable self-repair thus avoiding structural impairment or permanent damage of material properties. Covalent adaptable networks (CANs) have garnered attention for the sustainable development of polymers capable of mending damages thus facilitating eco-friendly recyclability. This work aims to address the role of reaction pathway and manufacturing process on the subsequent properties of CANs bearing hindered urea bonds (HUBs). HUBs have proven easy to incorporate into polymers and exhibit moderate temperature thermo-reversibility and creep resistance. In this investigation we study the effects of preparing CANs bearing HUBs using both free radical and thiol-ene chemistries. A novel HUB-based poly(urea urethane) prepolymer (PUUP) with diacrylate functionality enables self-healing under thermal stimulus. The PUUP was incorporated into networks with thiol building blocks at different stoichiometries; the diacrylate functionality enabled the formation of networks using either free radical photo-initiator or base catalyst. By preparing distinct networks with various molecular weight per elastically effective network chain (MC), this work addresses the interdependencies of network topology, mechanical properties, and self-healing resulting from these two synthetic strategies. Dynamic mechanical analysis (DMA) was utilized to measure the modulus of prepared networks, which enabled an assessment of crosslink density. Mechanical testing was then employed to investigate the bulk and self healing properties of these networks. A wide range of mechanical behavior was exhibited, depending on the preparation strategy: for photo-polymerized networks, elongation at break (emax) ranged from 88 to 152% and toughness values (UT) ranged from 0.45 to 0.62MJ/m3. On the other hand, base catalyzed networks exhibited elongation to break as high as 904%, and toughness as high as 2.75MJ/m3. A high healing efficiency of 100% recovery of toughness (hU) was exhibited by the base catalyzed sample with highest MC whereas photopolymerized networks exhibited significantly reduced performance (hU=29%). The reduced performance of photopolymerized networks is attributed to a competition between the desirable thiol-ene reaction and the competing free radical polymerization of acrylate groups. Additionally, intrinsically self-healing polymers are constructed via masked stereolithographic additive manufacturing (MLSA AM). The presence of strong hydrogen bonding between urea and urethane groups of the PUUP was reduced by incorporating reactive diluent tetrahydrofurfuryl acrylate (THFA), thus enabling the preparation of resins with sufficiently low viscosities (\u3c 10Pa.s) for vat polymerization. Resin formulations relied on traditional free radical polymerization or thiol-ene chemistry, where the incorporation of thiol building blocks facilitated a dual cure approach to building networks. The developed CANs are designed to contain HUBs within the side-chain (ladder) or main-chain (thiol). The polymerization mechanics of thiol systems resulted in networks with enhanced self-healing performance (83% \u3c he \u3c 98%) in comparison to ladder networks (he \u3c 21%). We further note that thiol networks exhibit high self-healing efficiencies over a wide range of HUB concentrations. These results suggest that thiol-ene chemistry can act as a powerful tool in generating AM network structures which enable sufficient topologic- and self-diffusion of HUBs to facilitate healing without a strong dependence on HUB concentration. Ultimately, this investigation has facilitated the development of advanced functional materials amendable to vat polymerization AM with utility in biomedical engineering, conductive devices/sensors, soft robotics, and for the extended shelf-life of energetics. This research provides the comprehensive characterization of the synthesis, scale-up, and utilization of AM to prepare CANs. By introducing a novel prepolymer capable of thermo-reversibility, the development of traditional cast-and-cure polymers are explored. These materials are further expanded towards use in the AM paradigm. An intrinsically healable HUB-based synthon was integrated into networks and scaled sufficiently to enable the 3DP workflow. This work highlights how this technology can be translated across multiple domains, acting to bridge the gap between small scale synthesis and AM of CANs
Inclusive Design Archive
Get PDFThe Inclusive Design Archive (IDA) is a conceptual research and design project that responds to the ongoing reliance on Eurocentric, male-dominated narratives in design education. Many art and design curricula continue to exclude underrepresented creators, movements, and methodologies, limiting students’ exposure to the full scope of global design practice. Even well-intentioned educators often lack accessible tools to bring diverse perspectives into the classroom in a structured, research-supported way. IDA offers a solution in the form of a contextual search application that helps educators and students explore topics through three intersecting lenses: Broader Perspectives, Intersectional Approaches and Alternate Narratives, and Concepts, Methodologies, and Critiques. Users might search broadly—like “modernism”—or more specifically—like “indigenous textile systems”—and receive layered results that include summaries, frameworks for understanding, and curated resources. The project was developed through research and iterative visual exploration, leading to a radially structured interface that organizes information by context and resource type. Final deliverables include a prototype UI, an informational website, poster series, and a short promotional video. Though conceptual, IDA proposes a feasible model for rethinking how inclusive research tools might support teaching, expand inquiry, and bring more voices into the stories we tell through design
Multiplex Leiden Optimized Dynamic Network Microsegmentation and Flow Validation for Software Defined Networks
No full textModern digital infrastructure is fundamentally built upon robust networks, which underpin seamless communication, data exchange, and the operation of critical systems. These networks facilitate everything from internet access and real-time patient vital sign transfers in healthcare to the management and control of industrial processes. As demands on networks have surged, their complexity and functionality have evolved significantly, continuously adapting to diverse sectoral needs. However, this rapid evolution has also amplified security challenges. The escalating scale and intricacy of contemporary networks have exposed vulnerabilities that traditional perimeter-based security models are often insufficient to counter. Static defenses prove inadequate in dynamic environments characterized by fluid traffic patterns, interconnected devices, and the prevalence of cloud-based and virtualized infrastructures. To mitigate these escalating risks, innovative network security approaches are imperative. Microsegmentation has emerged as a particularly effective solution, significantly enhancing protection and reducing the internal attack surface by dividing networks into smaller, isolated segments. This strategic partitioning inherently restricts lateral movement, limits unauthorized access, and effectively contains the impact of potential breaches. Microsegmentation is especially well-suited for cloud and virtualized environments, offering granular control over traffic flows and enabling the enforcement of security policies precisely tailored to specific applications, workloads, and users. This thesis presents a comprehensive study on microsegmentation for networks, thoroughly examining lateral movement techniques and corresponding defense measures. Crucially, a novel framework featuring an open-source implementation of dynamic microsegmentation for incident response, powered by Software-Defined Networking, has been developed and evaluated. Our rigorous testing demonstrates that the proposed framework offers a robust foundational security layer against lateral movement, proving particularly effective against known ransomware threats. Furthermore, testing conclusively shows that dynamic microsegmentation does not adversely affect network performance, establishing its suitability as an effective defense measure for east-west traffic. Finally, future research directions are offered to further develop on this proposed framework
People’s Plate: The Recipe for Equitable Food Distribution
Get PDFFood insecurity remains a pressing issue in the United States that disproportionately affects vulnerable populations, leaving millions without access to nutritious meals. In 2023 alone, over 50 million people turned to food assistance programs for additional support1. While traditional nutrition programs like food banks and soup kitchens provide a vital service to many communities, they often lack the capacity to address the multifaceted nature of hunger. The social stigma and alienation associated with seeking food assistance become of little concern when organizations are grappling with daunting challenges of limited resources, logistical hurdles, and the need to provide quality, nutritious meals. In order to offer a comprehensive and holistic solution to food insecurity, the immediate need for sustenance and the underlying social and logistical challenges need to be addressed simultaneously. This thesis proposes People\u27s Plate, a user-centered and community-based food security distribution program designed to provide nutritious food and support to vulnerable populations within any given region, while ensuring accessibility, adaptability, and sustainability. In a rigorous user-centered design process, the project involved extensive research, including interviews, surveys, and observations, to gain a deep understanding of the needs and preferences of food-insecure individuals. Iterative prototyping was employed to refine the design and optimize the user experience. The thesis explores a multifaceted solution that tailors to the needs of specific vulnerable communities, and the final design prioritizes accessibility, sustainability, and welcoming community, creating a self contained space for individuals to receive nutritious meals. People\u27s Plate aims to make an impact on our most vulnerable food insecure communities by being a source of reliable, dependable, and adaptive food solutions. People\u27s Plate has the potential to open a pathway for the development of a multifaceted food distribution program that can address national food insecurity more effectively than current methods, poising it to be part of the blueprint for a more equitable and successful part of the national food distribution system
Neuro-Evolution for Multivariate Time Series Anomaly Detection
Get PDFGiven the exponential growth of data in various sectors, the timely identification of anomalies has become crucial in mitigating risks such as financial losses and data breaches. This research focuses on advancing multivariate time series anomaly detection through the innovative application of neuro-evolutionary methods. Utilizing the Evolutionary eXploration of Augmenting Memory Models (EXAMM) algorithm, the study automates the design and optimization of Long Short-Term Memory (LSTM) autoencoder networks, enabling enhanced forecasting capabilities in recurrent neural network architectures. Leveraging these networks along with a One-Class Support Vector Machine (OCSVM) to generate boundary thresholds, this project delivers a robust and scalable procedure for identifying anomalies that outperforms existing proven methods in terms of accuracy and cumulative F1 score, while at the same time requiring orders of magnitude fewer trainable parameters. By comparing the reconstruction error of trained forecasts, the proposed approach facilitates accurate anomaly detection across diverse datasets against unseen anomalous data and informs efficient design decisions for future autoencoder network architectures