22455 research outputs found
Sort by
Representation learning for domain adaptation and cross-modal retrieval
Get PDFMost machine learning applications involve a domain shift between data on which a model has initially been trained and data from a similar but different domain to which the model is later applied on. Applications range from human computer interaction (e.g., humans with different characteristics for speech or handwriting recognition), computer vision (e.g., a change of weather conditions or objects in the environment for visual self-localization), and neural language processing (e.g., switching between different languages). Another related field is cross-modal retrieval, which aims to efficiently extract information from various modalities. In this field, the data can exhibit variations between each modality. Such variations in data between the modalities can negatively impact the performance of the model. To reduce the impact of domain shift, methods search for an optimal transformation from the source to the target domain or an optimal alignment of modalities to learn a domain-invariant representation that is not affected by domain differences.
The alignment of features of various data sources that are affected by domain shift requires representation learning techniques. These techniques are used to learn a meaningful representation that can be interpreted, or that includes latent features through the use of deep metric learning (DML). DML minimizes the distance between features by using the standard Euclidean loss, maximizes the similarity of features through cross correlation, or decreases the discrepancy of higher-order statistics like the maximum mean discrepancy. A similar but distinct field is pairwise learning and contrastive learning, which also employs DML. Contrastive learning not only aligns the features of data input pairs that have the same class label, but also increases the distance between pairs that have similar but different labels, thus enhancing the training process.
This research presents techniques for domain adaptation and cross-modal retrieval that specifically focus on the following two applications. (1) Online handwriting recognition involves representing written characters as multivariate time-series data from sensor-enhanced pens and aims to classify the written text. We recorded and evaluated various datasets for single character and sequence-to-sequence classification, and made them publicly available. We evaluated the domain shift that can occur between right- and left-handed writers, as well as between different writing styles, using uncertainty quantification techniques. Our approach utilizes higher-order statistics or optimal transport to adjust the features between right- and left-handed writers in order to minimize this domain shift. The best transformation is selected using DML techniques. Additionally, we assess the effectiveness of contrastive learning and DML for adapting the domain between writing on tablet and on paper, as well as for cross-modal retrieval in offline and online handwriting recognition. (2) Visual self-localization aims to determine the absolute and relative position and orientation of a human or robot using only one monocular camera. We propose to enhance the task of predicting the absolute pose by incorporating an auxiliary task of predicting the relative pose using optical flow during the learning process and to pre-train on simulated data. In addition, we evaluate different fusion methods that utilize representation learning to combine information from visual and inertial sensors
Application of molecular dynamics simulations for developability assessment and formulation development of biologics
Get PDF
Die transkranielle Random Noise Stimulation zur Behandlung kognitiver Symptome aus der Domäne der Negativsymptomatik bei Menschen mit einer Schizophrenie
Get PDFUntersuchungen zur Wirksamkeit antimikrobieller Peptide gegen bovine Mastitispathogene in vitro und zu Komponenten des angeborenen Immunsystems in einem bovinen mammären Explantmodell
Get PDFEntrustable Professional Activities (EPAs) framework to inform surgical residency training programs in Ethiopia medical education
Get PDFBackground
Entrustable Professional Activities (EPAs) are activities that are essential to the discipline and can be delegated to individuals without direct supervision in a specific health care context once they have demonstrated sufficient competence (Amare et al., 2021; Ten Cate et al., 2015; Ten Cate et al., 2017).
Since EPAs are believed to have potential benefits, a wide range of specialty programs have proposed them, and they have become popular in medical programs (Amare et al., 2021; Beeson et al., 2014; Haines et al., 2017; McCloskey et al., 2017; Peters et al., 2017; Ten Cate, 2017; van Loon et al., 2014; Young et al., 2018). Even though core EPAs have become available globally (Amare et al., 2021; Touchie & ten Cate, 2016), they cannot automatically be adapted for use in other contexts (Amare et al., 2021; Shorey et al., 2019). With this in mind, the need to develop an EPA framework for surgical residency training in Ethiopia is imperative (Amare et al., 2021). The goal is for graduating surgical residents must be able to carry out these EPAs independently by the time they graduate. However, graduates of general surgery residents have also been criticized for their ability to perform EPAs (Amare et al., 2021; Bucholz EM, 2011 Aug 15; Friedell et al., 2014; Moore et al., 2017; Perone JA, 2017 Apr 1; Wagner JP, 2018 Apr).
The present study aimed to develop valid end-of-training EPAs for surgical residency training programs as a framework to inform curriculum design, teaching, and assessing competencies in the local context of Ethiopian medical education” (Amare et al., 2021), as well as to assess how faculty members judge residents' performance in executing EPAs, and how residents rate their own ability to systematically introduce and implement EPAs in the “surgical residency training programs” (Amare et al., 2021).
Methods
"A three-round Delphi method was used to establish consensus about important surgical EPAs among experts. A total of 136 experts representing all surgical residency training institutions in Ethiopia were invited to participate. Round 1 & 2 consisted of senior expert panelists (n = 8) to identify potential EPAs and determine the content validity. Round 3 consisted of a survey (n = 128) to further validate the identified EPAs by attending surgeons who work with them. Each EPA had to achieve at least 80% or higher agreement among experts to be considered having acceptable content validity" (Amare et al., 2021). In addition, the survey was conducted at “four surgical residency training institutions in Ethiopia” (Amare et al., 2021) to investigate resident and surgical team members judgments of a graduating general surgery residents' competency in carrying out EPAs.
Result
“In round 1, a total of 272 EPAs were proposed, reduced, and grouped to 39 consented EPAs. In round 2, the same experts rated each EPA’s relevance, resulting in 32 EPAs with a satisfactory item-level content validity index (I- CVI > 0.83). Overall, in the survey in round 3, 29 EPAs met the standard criterion for acceptability (S-CVI/Ave = 0.90) and achieved a high degree of final consensus (ICC =0.998, 95% CI [0.996, 0.999]; (F = 439.2, p < 0.0001)” (Amare et al., 2021). In carrying out EPAs, there was a statistically significant difference in judgments between residents and surgical team members (P =0.03, CI: 0.51-0.95) as well as between surgical faculty members (P =0.001).
Conclusion
“The framework of 29 validated and accepted EPAs can guide future surgical residency training programs in the Ethiopian medical education context. The framework allows programs to move from a time-dependent to an outcome-based model and transforms traditional assessment into entrustment decisions. Thus, the use of the framework can improve the quality of training and patient care in Ethiopia” (Amare et al., 2021). The perception/judgment gap that exists between residents and surgical teams and among faculty members could pose a problem in education and healthcare systems. Our study emphasizes the need to describe EPAs in sufficient detail and to make performance criteria transparent and understandable before fully implementing an EPA-based assessment
The social role of AI advisers
Get PDFArtificial Intelligence (AI) profoundly affects how people communicate, work, and perceive the world. While autonomous AI systems are the focal point in societal and academic discussions, advisory AI systems, which influence human decisions but don't undertake independent actions, often remain unexplored. Examples range from automated purchase recommendations to medical diagnoses. This dissertation seeks to understand what advisory AI systems truly are. Are they capable of autonomous, human-like action? Or can they be reduced to inert tools? And what happens when advisory AI systems are closely linked with human perception, especially through Augmented Reality and sensory augmentation? Does their ontological status change? This dissertation concludes that, regardless of their implementation, advisory AI systems occupy an ontological status between tools and humans. They are more than just tools but less than humans.Künstliche Intelligenz (KI) beeinflusst massiv, wie Menschen kommunizieren, arbeiten und die Welt wahrnehmen. Während autonome KI-Systeme gesellschaftlich und akademisch im Fokus stehen, bleiben beratende KI-Systeme, die menschliche Entscheidungen beeinflussen, aber keine eigenständige Handlung übernehmen, oft unerforscht. Beispiele reichen von automatisierten Kaufempfehlungen bis hin zu medizinischen Diagnosen. Die vorliegende Dissertation untersucht, was beratenden KI-Systeme wirklich sind. Sind sie zu eigenständiger, menschenähnlicher Handlung fähig? Oder lassen sie sich auf handlungsunfähige Werkzeuge reduzieren? Und was passiert, wenn beratende KI Systeme eng mit der menschlichen Wahrnehmung gekoppelt werden, insbesondere durch Augmented Reality und sensorische Augmentation? Verändert sich ihr ontologischer Status? Die vorliegende Dissertation schlussfolgert, dass unabhängig von Ihrer Implementation beratende KI-Systeme einen ontologischen Status zwischen Werkzeugen und Menschen einnehmen. Sie sind also mehr als nur Werkzeuge, aber weniger als Menschen.
Louis Longin ist wissenschaftlicher Mitarbeiter am Lehrstuhl für Philosophie des Geistes an der Ludwig-Maximilians-Universität München, wo er 2023 mit der vorliegenden Dissertation promoviert wurde. Seinem Interesse gilt der wachsende Einfluss der Künstlichen Intelligenz auf den menschlichen Nutzer, besonders in den Bereichen der sozialen Interaktion, Ethik und sensorischen Augmentation
Hirnregionsspezifische Reaktion von Gliazellen auf invasive und nicht-penetrierende Verletzungen des Rückenmarks
Get PDFDie reaktive Gliose ist als physiologische Reaktion auf pathologische Veränderungen innerhalb des zentralen Nervensystems (ZNS) schon länger bekannt. Bisher wurden viele unterschiedliche Ausprägungsformen und Effekte in Abhängigkeit verschiedener pathologischer Prozesse und der betroffenen histologischen Kompartimente beschrieben. Die übergeordneten Auswirkungen sowie Einflussfaktoren dieser multizellulären Reaktion sind nicht zuletzt aufgrund der hohen Komplexität bisher ungewiss. Durch die Erkenntnis der letzten Jahrzehnte über die essenziellen Funktionen von Gliazellen in ihren interzellulären Interaktionen aber auch für das ZNS als ganzheitliches Organ ist das Potenzial für therapeutische Einflussmöglichkeiten auf diese Zellpopulation deutlich geworden. Bislang fehlen allerdings nach wie vor ausreichend Daten zu verschiedenen pathologischen Einflüssen, um das generelle Zusammenspiel beteiligter molekularer Signalkaskaden und interzellulärer Interaktionen zur Orchestrierung einer reaktiven Gliose entschlüsseln und potenzielle Konsequenzen umfassend beurteilen zu können.
Angesichts der zentralen Rolle von Astrozyten für alle Ebenen der funktionellen Plastizität und Mikroglia als Immuneffektorzellen innerhalb des ZNS, habe ich in dieser Arbeit die regionsspezifische Dynamik der Aktivierung dieser Gliazellpopulationen nach thorakalen Rückenmarkstraumta untersucht. In dieser Arbeit konnte ich zeigen, dass auch Gliazellen des zerebralen Kortex auf Traumata des Rückenmarks reagieren. Diese Reaktion war darüber hinaus nicht nur auf invasive Rückenmarksverletzungen beschränkt, sondern konnte auch im Rahmen von Rückenoperationen in unmittelbarer Umgebung der Wirbelsäule beobachtet werden. Diese Ergebnisse unterstützen die Annahme, dass eine reaktive Gliose auch peripher zur Läsionstelle auftretten kann und die gliale Reaktion eine regionsspezifische Heterogenität aufweist. In der Konsequenz besteht aufgrund ausgelöster glialer Dysfunktionen das Potenzial für klinisch relevante längerfristige Folgen auf übergeordnete Funktionen des ZNS über die akuten neurologischen Symptomatiken hinaus
