150813 research outputs found
Sort by
Report to the President for year ended June 30, 2025, Institute Discrimination and Harassment Response Office
No full textThis report contains the following sections: Academic year 2025 overview, Incident reports, Initiatives, Committee progress, and IDHR staff updates
Location Verification for Spoofing Detection in Non-Terrestrial Networks
No full textReliable location awareness is essential for the development of new services and applications in non-terrestrial networks (NTN). The ability of malicious users to report false location information poses a significant threat to NTN performance. This threat introduces the need for a flexible and robust location verification system (LVS) that can reliably detect malicious users. This paper proposes a single-satellite LVS based on round-trip time and angle-of-arrival measurements. We characterize several sources of uncertainty unique to the NTN scenario and examine their combined effect on positioning error. To detect spoofing probabilistically, we approximate the likelihood function for the unknown user position using a Gaussian mixture model and employ a likelihood ratio decision rule for location verification. Results display receiver operating characteristic curves to evaluate the LVS performance under various satellite ephemeris error conditions, spoofing distances, number of measurements available to the system, and wireless channel properties. The proposed LVS is shown to reliably detect spoofing among malicious users.S.M
Robust and expert-agnostic digital twin calibration via ensemble learning and Bayesian optimization
No full textBUILDSYS ’25, Golden, CO, USADigital twins have emerged as a critical tool in tackling climate change. Considering the data scarcity of complex systems, a promising approach to developing digital twins involves combining physics-based models with data assimilation. However, model calibration remains challenging due to uncertainties in both the physical models and observational data, and the reliance on domain knowledge. In this study, we develop an ensemble learning-based approach that aggregates sub-models with diversified calibration configurations. The proposed method streamlines calibration without expert-driven parameter screening and improves the digital twin's extrapolation capability, enabling more robust predictive applications. We demonstrate the effectiveness of our approach by calibrating the energy model of an office building, significantly reducing the extrapolation error and the associated risks. To the best of our knowledge, this is the first study to facilitate the calibration of physics-based models using ensemble learning, especially in the parameter space
Injection of Domain-Specific Knowledge for Enterprise Text-to-SQL
No full textThis work examines the current state of using large language models (LLMs) to solve Text-to-SQL tasks on databases in an enterprise setting. Benchmarks on publicly available datasets do not fully capture the difficulty and complexity of this task in a real-world, enterprise setting. This study examines the critical steps needed to work with enterprise data as well as using knowledge-injection to enhance the performance of LLMs on Text-to-SQL tasks. We begin by evaluating the baseline performance of LLMs on enterprise databases, revealing that a predominant source of failure stems from a lack of domain-specific knowledge. To improve performance, we explore knowledge-injection: the process of incorporating internal and external knowledge. Internal knowledge consists of database-specific information such as join logic, while external knowledge refers to institutional acronyms or group names. We present a hybrid retrieval pipeline that combines embedding and text based searching with LLM-guided ranking to supply models with relevant external knowledge during Text-to-SQL generation. We evaluate the impact of the knowledge-injection by testing the performance of LLMs on the table retrieval task after being augmented with appropriate external knowledge. We demonstrate that knowledge-injection significantly improves accuracy on table retrieval using BEAVER: an enterprise-level Text-to-SQL benchmark. Our findings highlight the importance of domain-specific knowledge-injection and retrieval augmentation in bringing LLMs closer to deployment in enterprise-grade database systems, as well as common failure modes that occur when executing enterprise Text-to-SQL.M.Eng
Computational Design of Architected Lattices for Construction Applications
No full textArchitected lattices have been utilized in aerospace and research applications for their modularity, scability, reconfigurability, and high strength-to-weight properties. However, voxels have yet to find widespread integration in the residential or commercial construction industry because of the industry’s distinct system needs. This study identifies the pain points unique to the construction industry that have slowed or disabled the adoption of new practices, highlighting the importance of utilizing known materials, methods, and the transparency of the design process, as major hurdles to adoption of innovation in the industry. This study presents a computational approach to designing architected lattices that seeks to undermine these core issues by making building with architected lattice structures agnostic to material and manufacturing methodology. Three open source computational approaches to architectural design are proposed: 1) integration of support structures for additively manufactured structures; 2) parametric design of voxels from 2D material, their manufacturing molds, and optional alignment features; and 3) generation of two-dimensional cut files for assembly with 3D printable joinery. These files are computationally designed and arranged for instantaneous production to demystify the lattice architectural design process, establish a pathway for utilizing all available materials in lattice construction, reduce the overhead costs for experimentation with lattice structures, and eliminate barriers to the fabrication process by enabling accessible manufacturing methods.S.M
Ideator Explorer: Enhancing AI-Assisted Ideation through Interactive Visualization
No full textCurrent AI-assisted ideation systems, often based on linear chat interfaces, struggle to help users effectively manage the complexity of creative exploration, hindering both divergent thinking across multiple paths and the convergent synthesis of ideas. This thesis introduces and evaluates Ideator Explorer, a human-AI ideation system built upon an interactive graph visualization interface designed to overcome these limitations. The core of the system is its spatial, tree-like representation of branching idea sequences. Formative user studies indicate that this visualization approach is preferred over chat interfaces for its organizational benefits and its effectiveness in helping users track parallel lines of thought during exploration. The spatial layout inherently supports both the exploration of diverse idea branches (divergence) and the identification of potential connections (convergence). This research focuses on the design and evaluation of this interactive graph interface, examining how its specific visualization and interaction techniques impact the user’s ability to navigate, organize, and develop ideas within complex ideation processes. The primary contribution is a novel, visually driven interface paradigm for human-AI collaboration that enhances the management and exploration of the creative solution space.M.Eng
Report to the President for year ended June 30, 2025, Norman B. Leventhal Center for advanced Urbanism
No full textThis report contains the following sections: Finance and Funding, Accomplishments, Administrative Initiatives, Personnel Updates, Teaching Impacts, Research Activities, Conferences and Presentations, Press and Publications, and Affiliated Faculty
Dipole Contact Engineering for Field-Effect Transistors Based on Two-Dimensional Materials
No full textIn the next several years and decades, the expanded use of artificial intelligence and edge computing will demand more powerful and energy-efficient electronics. Two-dimensional (2D) semiconductors, and in particular transition metal dichalcogenides (TMDs) such as molybdenum disulfide (MoS₂), are promising candidates for future field-effect transistors. TMDs can enable aggressive lateral and vertical device scaling, and they can add computing power density and new memory and sensing capabilities via 3D integration. However, several key challenges remain before 2D-channel transistors become commercially viable, including large contact resistances at the source and drain due to the van der Waals surface of 2D materials and the Fermi level pinning effect. A variety of methods have been explored to make ohmic contacts to MoS₂, the most promising of which so far is to use semimetals such as Bi and Sb, however these materials suffer from thermal instability. This thesis addresses these challenges by (1) exploring the ultimate limit of contact metal workfunction scaling to better understand the metal-MoS₂ interface, and (2) introducing a new method of reducing contact resistance to 2D materials by inserting dipole layers at the contact interface. Initial work on ultralow-workfunction (ULWF) metal deposition on MoS₂ and subsequent device fabrication is presented, though further study is required to mitigate effects from deposition equipment and the reactive nature of these metals. In parallel, the Janus TMD MoSSe is explored as an example system for dipole contacts, with extensive material characterization of the Janus TMD MoSSe being performed, and the effect of a dipole layer on the contact properties of FETs being established. Together, these results are a significant step towards solving one of the major hurdles for the commercial introduction of 2D-channel transistors.S.M
Evaluating the Impact of Equipment Investments on Olympic Medal Probabilities for Australian Professional Cyclists
No full textAusCycling is the National Sporting Organization for cycling in Australia. Their oversight includes all aspects of the sport, including the high-performance program. As AusCycling begins preparations for the 2028 LA and 2032 Brisbane Olympics, they look to invest in new cycling equipment to boost their expected medal counts. This research takes a three-phased approach at selecting an efficient equipment investment portfolio for AusCycling that results in high medal probability. Using machine learning techniques, we first build a CATBoost prediction algorithm that classifies future Olympic performance into the categories of “earn a medal”, “close to earning a medal”, and “not close to earning a medal”. The model predicts Olympic performance from performance at competitions prior to an Olympic Games with an overall accuracy of 96.3%. In the second phase of the research, a methodology is built to compute the percentage in race time an athlete needs to improve between two World Championship competitions in order to meet a probability threshold for earning a medal at the next Olympics. The final phase of this research combines the models and methodology of the first two sections by creating a Mixed-Integer Nonlinear optimization model which selects optimal equipment investments to maximize predicted Olympic performance while minimizing cost. When used on synthetic data similar to that available to AusCycling, the optimization selects an investment portfolio that yields an expected number of medals of 2.15 across men’s and women’s Team Pursuit and men’s Team Sprint for the 2028 Olympics. This methodology may help AusCycling determine which equipment investments to make ahead of future Olympic competitions.MN
Numerical Analysis of Human-Informed Topology Optimized Lateral-Load-Resisting Systems of Tall Buildings under Seismic Excitation
No full textIn the construction industry, structural, architectural, and environmental considerations can often be at odds with each other, leading to inefficient structures and, consequently, material waste. Topology optimization has shown promise as one potential solution to this problem, offering designs that are both structurally efficient and aesthetically interesting. However, topology-optimized designs are often difficult to manufacture or do not take into consideration other aspects that are crucial in the construction industry. Human-informed topology optimization, or HiTop, is a previously-developed algorithm that allows users to edit areas of interest, providing a computationally-efficient solution to address concerns with the designs. This paper uses MATLAB to apply HiTop to the design of the lateral-load-resisting systems of tall buildings, comparing results to those of three other designs: a “human” design with standard cross bracing, a optimized design using classical topology optimization, and a previously-developed algorithm which optimizes designs under a sum of modal compliances formulation, similar to how structures are analyzed in seismic codes. The designs are evaluated quantitatively, comparing natural periods, modal displacements, sum of modal compliances using modal decomposition, as well as computation time. They are also evaluated qualitatively, as HiTop is used to modify designs to improve constructability and aesthetics. The HiTop algorithm successfully created manufacturable, aesthetic designs in line with the user’s goals across a range of H/B ratios within a brief time frame. HiTop designs also performed similarly to the classically optimized designs, indicating that modifications to an optimized design to improve manufacturability, aesthetics, or other potential goals of a user do not significantly decrease structural performance under seismic loading.M.Eng