150813 research outputs found
Sort by
Detecting and Mitigating the Clever Hans Effect in Medical Imaging: A Scoping Review
No full textThe Clever Hans effect occurs when machine learning models rely on spurious correlations instead of clinically relevant features and poses significant challenges to the development of reliable artificial intelligence (AI) systems in medical imaging. This scoping review provides an overview of methods for identifying and addressing the Clever Hans effect in medical imaging AI algorithms. A total of 173 papers published between 2010 and 2024 were reviewed, and 37 articles were selected for detailed analysis, with classification into two categories: detection and mitigation approaches. Detection methods include model-centric, data-centric, and uncertainty and bias-based approaches, while mitigation strategies encompass data manipulation techniques, feature disentanglement and suppression, and domain knowledge-driven approaches. Despite the progress in detecting and mitigating the Clever Hans effect, the majority of current machine learning studies in medical imaging do not report or test for shortcut learning, highlighting the need for more rigorous validation and transparency in AI research. Future research should focus on creating standardized benchmarks, developing automated detection tools, and exploring the integration of detection and mitigation strategies to comprehensively address shortcut learning. Establishing community-driven best practices and leveraging interdisciplinary collaboration will be crucial for ensuring more reliable, generalizable, and equitable AI systems in healthcare
Meta-Learning Exploration Strategies with Decision Transformers
No full textThe problem of pure exploration in sequential decision-making is to identify strategies for efficiently gathering information to uncover hidden properties of an environment. This challenge arises in many practical domains, including clinical diagnostics, recommender systems, and educational testing, where data collection is costly and the effectiveness of exploration is critical. Efficient exploration in these contexts strongly depends on exploiting underlying structural relationships within the environment. For instance, recognizing that multiple medical tests may provide overlapping information can reduce the number of tests required to make a diagnosis. Existing exploration approaches drawn from reinforcement learning and active hypothesis testing typically rely on heuristic strategies that require explicit prior assumptions about such structural information. However, when this information is unknown, heuristic methods often lead to redundant exploration, significantly limiting their practical utility in high-stakes domains. Furthermore, these existing approaches do not leverage past experience to improve their exploration efficiency over time. To overcome these limitations, we introduce In-Context Pure Exploration (ICPE), a novel meta-learning framework capable of autonomously discovering and exploiting latent environmental structures across related tasks to guide efficient exploration. ICPE leverages the in-context learning and sequence-modeling capabilities of transformers, combined with supervised learning and deep reinforcement learning techniques to learn exploration strategies directly from experience. Through extensive experiments on synthetic and semi-synthetic exploration tasks, we demonstrate that ICPE is able to efficiently explore in deterministic, stochastic and highly structured environments without relying on any explicit inductive biases. Our results highlight the potential of ICPE to enable more practical exploration strategies suitable for real-world decision-making contexts.M.Eng
A polyurethane-urea elastomer at low to extreme strain rates
No full textA finite strain nonlinear constitutive model is presented to study the extreme mechanical behavior of a polyurethane-urea (PUU) well suited for many engineering applications. The micromechanically- and thermodynamically based constitutive model captures salient features in resilience and dissipation in the material from low to extreme strain rates. The extreme deformation features are further elucidated by laser-induced micro-particle impact tests for the material, where an ultrafast strain rate ( > 1 0 6 s−1) incurs. Numerical simulations for the strongly inhomogeneous deformation events are in good agreement with the experimental data, supporting the predictive capabilities of the constitutive model for the extreme deformation features of the PUU material over at least 9 orders of magnitude in strain rates ( 1 0 − 3 to 1 0 6 s−1)
Intracellular proteomics and extracellular vesiculomics as a metric of disease recapitulation in 3D-bioprinted aortic valve arrays
No full textIn calcific aortic valve disease (CAVD), mechanosensitive valvular cells respond to fibrosis- and calcification-induced tissue stiffening, further driving pathophysiology. No pharmacotherapeutics are available to treat CAVD because of the paucity of (i) appropriate experimental models that recapitulate this complex environment and (ii) benchmarking novel engineered aortic valve (AV)–model performance. We established a biomaterial-based CAVD model mimicking the biomechanics of the human AV disease-prone fibrosa layer, three-dimensional (3D)–bioprinted into 96-well arrays. Liquid chromatography–tandem mass spectrometry analyses probed the cellular proteome and vesiculome to compare the 3D-bioprinted model versus traditional 2D monoculture, against human CAVD tissue. The 3D-bioprinted model highly recapitulated the CAVD cellular proteome (94% versus 70% of 2D proteins). Integration of cellular and vesicular datasets identified known and unknown proteins ubiquitous to AV calcification. This study explores how 2D versus 3D-bioengineered systems recapitulate unique aspects of human disease, positions multiomics as a technique for the evaluation of high throughput–based bioengineered model systems, and potentiates future drug discovery
Multimessenger signatures of compact binaries
No full textGravitational waves and electromagnetic observations provide complementary views into some of the most extreme objects in the Universe. In this thesis, I present studies of multimessenger compact binaries from two angles: electromagnetic follow-up of gravitational-waves, and gravitational-wave follow-up of electromagnetic sources. I first describe technical and computational efforts to enable the distribution of alerts of kHz gravitational-wave sources as a member of the LIGO--Virgo--KAGRA collaboration, and to improve localizations of these events by folding in galaxy catalog information. I then detail work to enable electromagnetic follow-up observations of binary neutron star and neutron star--black hole mergers with two telescopes, the Transiting Exoplanet Survey Satellite (TESS) and the Wide-field Infrared Transient Explorer (WINTER). Approaching multimessenger observations from the opposite direction, I describe a search for gravitational waves coincident with fast radio bursts from the only Galactic fast radio burst source. Lastly, I perform an electromagnetic study of Type Ia supernovae in the mid-infrared, whose white dwarf binary progenitors will be mHz gravitational-wave sources for the future LISA space mission.Ph.D
Coplanarity of rooted spanning-tree vectors
No full textEmploying a recent technology of tree surgery, we prove a “deletion–constriction” formula for products of rooted spanning-trees on weighted directed graphs that generalizes deletion–contraction on undirected graphs. The formula implies that, letting τ x ∅ , τ x + , and τ x - be the rooted spanning-tree polynomials obtained, respectively, by removing both directed edges between two vertices, or by forcing the tree to pass through either edge, the vectors ( τ x ∅ , τ x + , τ x - ) are coplanar for all roots x . We deploy the result to give an alternative derivation of a recently found mutual linearity of stationary currents of Markov chains. We generalize deletion–constriction and current linearity for two pairs of edges and conjecture that similar results may hold for arbitrary subsets of edges
Nationwide Trends in Hospitalizations for Sudden Cardiac Arrest Before and During the COVID Outbreak
No full textBackground/Objectives: Sudden cardiac arrest (SCA) accounts for ~50% of cardiovascular mortality in the U.S. Cardiovascular complications are common in acute and post-acute COVID-19 infection. We aimed to examine nationwide trends in SCA-related hospitalizations in the United States before and during the COVID-19 outbreak. Methods: Using data from the National Inpatient Sample, we conducted a retrospective analysis of hospitalizations for SCA in the U.S. between 2016 and 2020. Sociodemographic and clinical characteristics and in-hospital mortality were compared between the pre-COVID (2016– 2019) and COVID (2020) eras. Multivariable analysis was performed to identify factors associated with mortality. Results: Among a weighted total of 153,100 SCA hospitalizations between 2016 and 2020, the median age was 65 years, 62.7% were male, and 66.6% were white. There was a trend towards fewer hospitalizations in 2020 compared to prior years (n = 28,585 vs. naverage = 32,129, p = 0.07). In-hospital mortality remained unchanged between the pre-COVID and COVID eras (47.7% vs. 47.3%, p = 0.66). Increased mortality was associated with female sex (OR: 1.21; 95% CI: 1.15–1.28; p < 0.001), non-white race (OR: 1.24; 95% CI: 1.15–1.28; p < 0.001), history of renal failure (OR: 1.08; 95% CI: 1.02–1.15; p = 0.007), and diabetes (OR: 1.32; 95% CI: 1.25–1.39; p < 0.001). In 2020, 1.5% of the study population was diagnosed with COVID-19 infection, which was found to be independently associated with increased in-hospital mortality (OR: 1.57; 95% CI: 1.27–1.95; p < 0.001). Conclusions: In 2020, there was a trend towards a decrease in hospitalizations for SCA, while COVID-19 infection was independently associated with higher in-hospital mortality among patients admitted with SCA
Digital Twin Technology Applied to Automotive Diagnostics
No full textThere is currently a lot of interest in the area of Digital Twin (DT) Technology. Physical product oriented organizations are increasingly looking for ways to stay ahead of the technological innovation curve in order to not get disrupted by more agile entrants. Therefore, the promise of a technology like DT is alluring for the sake of maintaining a competitive edge. This thesis seeks to explore the potential benefits of DT technology alongside what challenges might be faced in implementing one. To this end, a problem statement is formulated in the field of automotive diagnostics. This is a key value addition field for automotive companies seeking to better manage the diagnosis and repair of their automobiles in the field or the manufacturing environment. The problem is further concretized with a study of some user-driven use cases and needs in a real automotive company. From these needs, a set of requirements is formulated to guide the architecture and design of a DT demonstration. The process of architecting and designing the DT is documented. This includes a deep dive on the modeling approaches considered, the solution space for the architecture and the detailed design and implementation of a DT demonstration from a selected architectural concept. The DT demonstration is then operated under controlled conditions in order to showcase some of its capabilities. Pursuant to all this, a reflection on the effectiveness of the demonstration and the lessons learned about the implementation process are discussed. The results of the study and demonstration show some promise for organizations seeking to adopt DT technology, in this particular case for automotive diagnostics. The benefits are mainly in terms of better system architecture planning and the increased potential for better incorporating lessons learned from products operating in the field back into the design process. These benefits are weighed against the socio-technical challenges of implementing DTs from the outset of a system design exercise.S.M
A holistic model for understanding the dynamics of outsourcing
No full textOutsourcing is a complex process as many external and internal factors that look convincing in the first place might, however, lead to a failure in the long run. Motivated by this, we wanted to get a holistic understanding of such outsourcing decisions. Thus, we created a comprehensive System Dynamics simulation model including all relevant variables to examine the dynamic nature of outsourcing in a holistic manner and over time that consists of more than 200 interrelated variables. Our results show, amongst others, that higher process specialisation that requires substantial investments by the supplier appears to be favourable for an outsourcing company and shifting a larger quantity to such a supplier achieves better cost savings and thus accounts for a better overall outsourcing result. On an operational level, we identified an innovation trap, a bargaining power shift, a plagiarism trap, and a knowledge trap. Based on that, we give specific managerial recommendations to tackles these aspects. We conclude that, amongst others, it is important for innovative companies with rather complex processes and parts to carefully plan which and how many employees to release so as not to lose the knowledge on those outsourced processes and parts
Large Language Models in Qualitative Research: Uses, Tensions, and Intentions
No full textCHI ’25, Yokohama, JapanQualitative researchers use tools to collect, sort, and analyze their
data. Should qualitative researchers use large language models
(LLMs) as part of their practice? LLMs could augment qualitative
research, but it is unclear if their use is appropriate, ethical, or
aligned with qualitative researchers’ goals and values. We interviewed twenty qualitative researchers to investigate these tensions.
Many participants see LLMs as promising interlocutors with attractive use cases across the stages of research, but wrestle with their
performance and appropriateness. Participants surface concerns
regarding the use of LLMs while protecting participant interests,
and call attention to an urgent lack of norms and tooling to guide
the ethical use of LLMs in research. We document the rapid and
broad adoption of LLMs across surfaces, which can interfere with
intentional use vital to qualitative research. We use the tensions
surfaced by our participants to outline recommendations for researchers considering using LLMsin qualitative research and design
principles for LLM-assisted qualitative research tools