143258 research outputs found
Sort by
Minutes: Staff Senate, March 12, 2025
No full textUniversity of Minnesota Duluth. Staff Senate. (2025). Minutes: Staff Senate, March 12, 2025. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/275522
Robustness and safety of deep learning models.
No full textUniversity of Minnesota Ph.D. dissertation. May 2025. Major: Electrical/Computer Engineering. Advisor: Ju Sun. 1 computer file (PDF); xx, 150 pages.Deep learning (DL) refers to a data-driven machine learning technique in which neural networks with many layers are used to model and understand complex patterns and relationships in the data. DL models have revolutionized numerous complex real-world tasks ranging from image recognition to natural language processing, demonstrating significant performance gain over the `traditional' approaches. Despite the impressive performance of DL models, concerns about their robustness and reliability persist: DL models are noted to be sensitive to adversarial attacks, data distribution shifts, and other perturbations, which can lead to significant performance degradation.
As a result, the adoption of DL models in high-stakes real-world applications is still limited today, and addressing the robustness issue of DL models is an emerging and critical research area. In this thesis, we present our findings on the robustness issue of DL models. First, we point out the robustness challenge on DL classifiers --- current adversarial robustness evaluation may not be rigorous, and the robustness conclusions drawn based on such evaluation may not be trustworthy. Based on our analysis, we express our pessimistic view that universal robustness for DL classifiers is a goal too ambitious to achieve. Next, we discuss the robustness challenge in the application of DL-based watermarks. Despite that existing DL based watermarking systems are shown to be robust to traditional digital corruptions (e.g., jpeg compression, additive noise), we show that small but carefully crafted perturbations can easily break existing watermarking systems, requiring no knowledge about the watermarking system itself. We also show that incorporating low-frequency component in the image watermark is necessary to robust image watermarks. Then, we discuss the idea of selective classification (prediction with a reject option) to accept the imperfection of DL models and make best use of them. We propose a confidence score using raw logit output from the DL classifiers and show its better potential in performing selective classification to reduce the liability of mistakes made by DL models. Lastly, we discuss future research directions based on our work, including potential ways to make DL classifiers more robust, how to develop more reliable DL-based watermarking systems, and ways to achieve reliable selective classification in practice.Liang, Hengyue. (2025). Robustness and safety of deep learning models.. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/275904
Gophers Women's Track and Field soar to No. 29, break records amid season challenges
No full textRuntime 09:27 minutesSophia Bablitch and Megan Davis break down major updates across Gopher sports, including big jumps in women’s track and field rankings, standout performances in golf and gymnastics, new basketball transfers and Gophers in the NBA.Bablitch, Sophia; Davis, Megan. (2025). Gophers Women's Track and Field soar to No. 29, break records amid season challenges. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/274656
Investigating the Mississippi River: A qualitative analysis of perceptions, engagement, and values
No full textThis report analyzes pre-existing interviews with residents along the Mississippi River from the headwaters to the Gulf of Mexico. Our goal is to provide qualitative data on respondent’s perceptions, values, and engagement with the river to better inform federal policy decisions on water quality investments and support government and nonprofit communications and strategic planning efforts along the river. Our guiding research questions were:
How do resident’s perceptions, engagement, and values of the river differ throughout the diverse geophysical regions of the Mississippi River corridor?
What correlations emerge between how people engage with the river and how they talk about the river throughout the regions?
This work summarizes data into a table to compare geographic features, river access, engagement, perceptions, values, and threats across the nine qualitative regions outlined by the Relay of Voices project. We incorporated respondent quotes in a narrative about the ways in which perceptions, engagement, and values differ throughout the regions and distilled key takeaways from the interview data to provide insight into the varied relationships people hold with the river throughout its diverse regions.This work is part of a three-year project (2022-2025) called "Investigating the distribution and value of water quality benefits along the Mississippi River" funded by the Environmental Protection Agency (EPA) to research the value of a cleaner Mississippi River with the goal of informing federal water policy. This $742,000 project is in collaboration with researchers from Michigan State University and the University of Washington deploying multiple methods to better understand how different communities perceive the benefits and costs of a changing Mississippi River.Warren, Natalie; Keeler, Bonnie. (2025). Investigating the Mississippi River: A qualitative analysis of perceptions, engagement, and values. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/276422
Neuromodulation for refractory epilepsy: biomarkers and stimulation strategies
No full textUniversity of Minnesota Ph.D. dissertation. May 2025. Major: Biomedical Engineering. Advisor: Theoden Netoff. 1 computer file (PDF); x, 105 pages.Implantable brain sensing and stimulation devices offer transformative potential for treating neurological disorders such as epilepsy. Currently, electrical stimulation is primarily adjusted and improved through “physician in the loop” control. Harnessing the full potential of these technologies requires appropriate tuning of the stimulation therapy, reliable biomarkers, and closed-loop control strategies with proper neural signal processing. This thesis investigated biomarkers and tuning strategies for further improving current stimulation therapy options as well as the development of a novel real time closed-loop neuromodulation approach.As part of an ongoing anterior nucleus of the thalamus deep brain stimulation clinical trial, we developed and implemented critical in clinic Medtronic Percept™ data collection workflows. Local field potential data processing, analysis, and patient response to different stimulation parameters were estimated at every neurologist followup to determine a research optimized setting that minimized their LFP broadband response. Through retrospective analysis of 11 participants in the trial, a significant association was observed between responders and participants that exhibited a slow gamma oscillation. In 4/6 responders, DBS suppression of slow gamma oscillations was observed within clinical visits under multiple settings. Furthermore, chronic stimulation across visits demonstrated a long term reduction in baseline (“stimulation off”) slow gamma activity in all 6 responders. These data demonstrate the potential for slow gamma oscillations as a biomarker of ANT DBS responder candidacy, effective therapy management, and a potential signal for optimizing stimulation parameters over time.
Additionally, we present an Adaptive Real-TIme STate Space (ARTISTS) closed-loop controller based on an Auto-Regressive with eXogenous input (ARX) model and Linear Quadratic Regulator (LQR) for state estimation and optimal stimulation. This low-complexity controller enables dynamic adjustment of stimulation parameters without prior system knowledge and was validated in biologically grounded models of epilepsy and Parkinsonian dynamics where neural activity was significantly suppressed during stimulation.
Both findings in ANT DBS and the development of the ARTISTS controller show potential for new closed-loop ANT DBS therapy opportunities and methodologies for improving neuromodulation to be tuned based on quantitative biomarkers, representative of the patient state.Sanger, Zachary. (2025). Neuromodulation for refractory epilepsy: biomarkers and stimulation strategies. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/275921
Effects of electrostatic spray drying encapsulation on flavors and its comparison with traditional spray drying.
No full textUniversity of Minnesota M.S. thesis. 2025. Major: Food Science. Advisor: Gary Reineccius. 1 computer file (PDF); xi, 70 pages.This study compared the flavor encapsulation properties of electrostatic spray drying (ESD) with traditional spray drying (TSD). Four flavor compounds—limonene, citral, cinnamaldehyde, and t-2-hexenal—were encapsulated using two different carrier system formulations: one with gum acacia and maltodextrin (Formulation 1), and another with HICAP-100 (a chemically modified starch) and maltodextrin (Formulation 2). The physical characteristics of the spray-dried flavors were assessed by measuring total oil, particle size, surface oil, and the formation of oxidation products over a 21-day accelerated storage period. ESD was performed at various atomizer voltages (+18, -18, +2, and -2 kV) to investigate how these voltages influenced the physico-chemical properties of the flavor encapsulant. Oxidation levels of the flavor compounds were determined using gas chromatography with flame ionization detection.ESD has been proposed as an alternative to traditional spray drying for encapsulating bioactive and volatile compounds due to its lower inlet temperatures and the use of nitrogen gas as drying medium. The particle size of TSD was found to be larger than that of any of the ESD samples, with no significant variation in particle size across different voltage settings. The total limonene content in the spray dried formulation with HICAP was similar for both TSD and ESD samples, except for the +18 kV sample, which exhibited the lowest oil content. Similarly, for the formulation with gum acacia, there was no significant difference in total oil content between TSD and ESD. However, oxidation levels of the flavor compounds were higher in ESD samples compared to TSD, which could be attributed to the higher surface oil content in ESD. Oxidation was evident immediately after spray drying in all ESD samples, though the degree of oxidation remained consistent across the different voltages. Further research is required to explore the underlying causes of unexpectedly high rate of oxidation in ESD samples.Jagdale, Yash. (2025). Effects of electrostatic spray drying encapsulation on flavors and its comparison with traditional spray drying.. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/273520
Farm legal series: Agricultural Production Contracts, 2025
No full textThis publication may not reflect current laws, scientific knowledge or recommendations. Current information may be available from the University of Minnesota Extension at www.extension.umn.edu.Peterson, Jeffrey A.. (2025). Farm legal series: Agricultural Production Contracts, 2025. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/273707
Pre-service principals' understanding of equity and inclusion
No full textUniversity of Minnesota M.A. thesis. 2025. Major: Organizational Leadership, Policy, and Development. Advisor: Nathaniel Stewart. 1 computer file (PDF); iii, 40 pages.The qualitative study explores how the University of Minnesota (UMN) administrative licensure principal candidates develop and understand equity framework principles through required course documents. I conducted a qualitative analysis of candidate documents completed by graduates of the principal preparation program. The UMN administrative licensure program will offer a space to deeply explore and code evidence for two equity frameworks. The first equity framework is the Principal Preparation Answerability Rubric (PPAR). The PPAR is a co-created, critical evaluation rubric that centers the needs, wants, and criticisms of principals from Black, Brown, and Indigenous (BBI) students and teachers. The second equity framework is the state-required Minnesota Rule 3512.0510 Competency C: Equity and Culturally Responsive Leadership. This required competency and list of sub-competencies must be demonstrated and approved for candidates to receive principal licensure by the state of Minnesota. This collaborative study uses grounded and state-level standards for equity and offers principal preparation program staff a chance to investigate their equity efforts within the program and understand areas of improvement.Bigirindavyi, Malaika. (2025). Pre-service principals' understanding of equity and inclusion. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/275819
Three essays in child development
No full textUniversity of Minnesota Ph.D. dissertation. May 2025. Major: Applied Economics. Advisor: Paul Glewwe. 1 computer file (PDF); vii, 132 pages.This dissertation consists of three independent essays on educational and health issues in child development. The first and second chapters are independent empirical studies authored solely by me, while the third chapter is co-authored with Yue Bao. The first essay investigates the development of personality traits and cognitive skills during adolescence. I find evidence of cross-effects and self-productivity of personality traits and cognitive skills, using Australian data. Both parental investments and child rearing are critical for the development of personality traits and cognitive skills during adolescence. I also find that personality traits and cognitive skills are important for adult earnings. The second essay examines the effect of living in the vicinity of an electronic-waste dumpsite on infant health outcomes in Ghana. I find that exposure from the e-waste dumping site significantly increases the probability of diarrhea and respiratory illness. I find two mechanisms, tetanus infection and water pollution. I find that premature birth is more likely to occur when mothers have not received a tetanus vaccine during or before pregnancy in the exposed area. I also find that diarrhea can be mitigated by consuming safer sources of drinking water. In the third essay, we develop a two-period discrete choice model to examine how higher education decisions are shaped when student loans play an increasingly crucial role in the US. We still find disparities in higher education attainment under a constructed scenario in which all students, regardless of family income, rely on student loans. The model highlights and helps us better understand whether disparities in higher education attainment are due to unequal access to financing or to deeper structural inequalities (for example, inequalities in K-12 preparation).Deng, Jiuchen. (2025). Three essays in child development. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/275877
Minutes: Student Government Association, April 7, 2025
No full textUniversity of Minnesota Duluth. Student Government Association. (2025). Minutes: Student Government Association, April 7, 2025. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/276134