9,819 research outputs found
Metadata Representations for Queryable ML Model Zoos
No full textMachine learning (ML) practitioners and organizations are building model zoos of pre-trained models, containing metadata describing properties of the ML models and datasets that are useful for reporting, auditing, reproducibility, and interpretability purposes. The metatada is currently not standardised; its expressivity is limited; and there is no interoperable way to store and query it. Consequently, model search, reuse, comparison, and composition are hindered. In this paper, we advocate for standardized ML model metadata representation and management, proposing a toolkit supported to help practitioners manage and query that metadata.Web Information SystemsHuman-Centred Artificial Intelligenc
A Manifesto of Nodalism
No full textThis paper proposes the notion of Nodalism as a means describing contemporary culture and of understanding my own creative practice in electronic music composition. It draws on theories and ideas from Kirby, Bauman, Bourriaud, Deleuze, Guatarri, and Gochenour, to demonstrate how networks of ideas or connectionist neural models of cognitive behaviour can be used to contextualize, understand and become a creative tool for the creation of contemporary electronic music
Optimizing ML Inference Queries Under Constraints
No full textThe proliferation of pre-trained ML models in public Web-based model zoos facilitates the engineering of ML pipelines to address complex inference queries over datasets and streams of unstructured content. Constructing optimal plan for a query is hard, especially when constraints (e.g. accuracy or execution time) must be taken into consideration, and the complexity of the inference query increases. To address this issue, we propose a method for optimizing ML inference queries that selects the most suitable ML models to use, as well as the order in which those models are executed. We formally define the constraint-based ML inference query optimization problem, formulate it as a Mixed Integer Programming (MIP) problem, and develop an optimizer that maximizes accuracy given constraints. This optimizer is capable of navigating a large search space to identify optimal query plans on various model zoos.Green Open Access added to TU Delft Institutional Repository ‘You share, we take care!’ – Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.Web Information SystemsHuman-Centred Artificial Intelligenc
Characterizing determinants of BK Polyomavirus-specific immune response
No full textBK polyomavirus (BKPyV) is one of now 13 human polyomavirus (HPyV) species detected in humans. BKPyV is only known to infect humans and seroprevalence rates of more than 90% have been reported in adult populations around the world. Following primary infection, BKPyV persists in the renourinary tract without causing any disease as evidenced by urinary shedding in 5% - 10% of healthy immunocompetent blood donors.
In immunocompromised persons, however, BKPyV can cause significant diseases whereby uncontrolled high-level replication may lead to organ invasive pathologies in kidneys, bladder, lungs, vasculature, and the central nervous system. The most consistently found diseases are BKPyV-associated hemorrhagic cystitis (BKPyVHC) in 5%-20% allogeneic hematopoietic stem cells transplant patients, and BKPyV-associated nephropathy (BKPyVAN) in 1%-15% of kidney transplant patients. BKPyVHC is highly symptomatic with pain, anemic bleeding, and increased mortality. BKPyVAN is asymptomatic except for progressive renal failure and premature return to dialysis. Both entities are characterized by high-level viral replication i.e. with urine BKPyV loads of 8-10 log10 Geq/mL, plasma BKPyV loads often above 4 log10 Geq/mL, and an allogeneic constellation between the virus-infected host cell and the available T-cell effectors. Despite these similarities, the clinical manifestations are strikingly different suggesting relevant, but experimentally undefined differences in pathogenesis. Thus, BKPyVHC typically occurs within 4 weeks after allogeneic HSCT and is confined to the bladder, and typically without kidney involvement. By contrast, BKPyVAN is diagnosed around 3-6 months after kidney transplantation and confined to the kidney allograft without causing cystitis. Although high-level BKPyV replication should be formally amenable to antiviral drug treatment, no effective and BKPyV-specific antiviral therapy is currently available. Therefore, a better understanding of the immune alteration in both diseases has been deemed essential to identify patients at risk and to develop prophylactic, preemptive and therapeutic strategies.
The currently recommended strategy for BKPyVAN is to screen kidney transplant patients for BKPyV replication and to promptly reduce immunosuppressive therapy in those with significant replication to facilitate mounting of BKPyV-specific T cell responses and thereby preventing progression to disease. This manoeuver has been linked to expanding BKPyV-specific T cell responses in the peripheral blood of kidney transplant patients. However, this approach may place patients at risk for acute rejection episodes that predispose equally well to premature kidney transplant failure. Although the clinical feasibility of reducing immunosuppression and curtailing BKPyV replication has been shown to be effective in prospective cohort studies for many, but not all of kidney transplant patients, this approach has not been possible in allogeneic HSCT patients because of concurrent or imminent graft-versus host disease. Thus, there are significant gaps in the current understanding of the BKPyV– host interaction in the normal host and in the allogeneic setting, which need to be investigated for a more effective and safer management of these significant viral complications.
In this thesis, the interaction of BKPyV and the immune response has been approached from two different angles. In the first project, potential mechanisms of BKPyV immune evasion were studied. Here, we focused on a small accessory protein called agnoprotein encoded as a leader protein in the late viral early region (LVGR). Although HPyV genomes overall show a very similar genome organization, agnoproteins are only found in the genomes of BKPyV and JCPyV that have a kidney tropisms, but not in any of the other 11 presumably non-renotropic HPyVs. We hypothesized that agnoprotein could play a role in immune evasion by downregulating HLA expression. The effects of agnoprotein were studied on HLA class I and II expression in vitro by flow cytometry following transfection of primary human renal tubular epithelial cells, which are the viral target of BKPyV-associated nephropathy. In addition, transfected human UTA-6 cells were studied as well as UTA-6 cells bearing a tetracycline-regulated agnoprotein. As control, the effects were compared with the ICP47 protein of Herpes simplex virus-1, which has been previously reported to effectively down-regulate HLA class I. Although both viral proteins share some similarities at the protein level, our results showed that BKPyV agnoprotein did not down-regulate HLA class I or class II molecules. Also, there was not inhibitory effect on the increase of HLA-class I or class-II surface expression following exposure to interferon-. By contrast, ICP47 reduced HLA class I surface expression, but not class II. We also evaluated effects of agnoprotein on virus epitope-specific T-cell killing by 51Chromium release assay, however no interference could be observed. We concluded that agnoprotein did not contribute to these types of HLA-dependent immune evasion processes. However, further investigations are needed to understand if agnoprotein could contribute to viral immune escape by other mechanisms.
In the second project, we aimed at better characterizing BKPyV-specific CD8 T cell immunity targeting epitopes encoded in the early viral gene region (EVGR). Selected coding sequences of the BKPyV EVGR were submitted to two web-based computer algorithms (SYFPEITHI, IEDB) in order to predict immunodominant 9mer epitopes presented by 14 frequent HLA-class I molecules. For an experimental confirmation, 97 different 9mer epitopes were chemically synthesized and tested in 42 healthy individuals. A total of 39 epitopes could be confirmed by interferon- ELISpot assay in at least 30% of healthy individuals. Interestingly, most of the 9mer epitopes appeared to cluster in short amino acid stretches, and some 9mer could be presented by more than one HLA class I allele as expected for immunodominant domains. HLA-specific presentation was demonstrated by 9mer- MHC-I streptamers for 21/39 (54%) epitopes. The 9mer dependent T-cell killing by 51Chromium release assay and the CD107a surface detection indicated that the 9mer epitopes could be recognized by cytotoxic T-cells. Moving to a clinically relevant situation, 13 9mer epitopes could be validated in 19 kidney transplant patients protected from, or recovering from, BKPyV viremia. The results suggest that, pending further corroboration in larger patient populations, novel 9mer epitopes can be identified, which are associated with CD8 T cell control of BKPyV replication. Thus the identified immunodominant 9mer T-cell epitopes could be further developed for clinical assays to better predict the risk and the recovery of BKPyV diseases, help guiding immunosuppression reduction, and to develop specific adoptive T-cell therapy or vaccine responses to prevent or treat BKPyV-associated disease
The evaluation of 24-hour spontaneous GH secretion in short children: relationship between mean concentration and pulsatile parameters
No full textIn 116 short children (height 3 ng/ml. MGHC was highly correlated (p < 0.001) with iMGHP (r = 0.80), nMGHP (r = 0.82), dMGHP (r = 0.59), MPA (r = 0.85), nMPA (r = 0.86), dMPA (r = 0.56), NP (r = 0.70), nNP (r = 0.68), dNP (r = 0.46). By the analysis of the regression equations, the values corresponding to 3 ng/ml for MGHC were 11.08 ng/ml for iMGHP, 11.66 ng/ml for nMGHP, 5.21 ng/ml for dMGHP, 7.29 ng/ml for MPA, 8.40 ng/ml for nMPA, 4.25 ng/ml for dMPA, 3.2 for NP, 2.41 for nNP and 0.78 for dNP. By using these values as cut-off points, the diagnostic accuracy yielded 83.6% for iMGHP, 84.5% for nMGHP, 69.8% for dMGHP, 92.2% for MPA, 90.5% for nMPA, 81.9% for dMPA, 80.2% for NP, 77.6% for nNP, 71.5% for dNP. In conclusion, we found a strong correlation between mean GH secretion over 24 h and the number or amplitude of pulses: particularly, nocturnal pulsatile GH parameters show a higher correlation in comparison with diurnal pulsatile GH parameters, so that the examination of GH values during nocturnal hours may be considered a reliable index of GH secretory status
Mental Retardation As Presenting Symptom Of Guanidinoacetate (Gaa) Metyltransferase (Gamt) Deficiency
No full text
Building a generalisable ML pipeline at ING
No full textAdvances in data science have caused an increase in the use of Artificial Intelligence (AI), specifically Machine Learning (ML), throughout various fields. Not only in research but in the industry as well, has ML been receiving increasing amounts of interest. Many companies rely on ML models to increase the efficiency of existing processes or offer new services and products. The industry, however, is facing several additional challenges compared to the academic context. One of those challenges is applying the Development Operations (DevOps) model to an ML application, also referred to as MLOps. This thesis sets out to find the specific challenges that practitioners encounter while operationalising ML models. To do so, we perform a single-case case study on an ML pipeline built by the Trade & Communication Surveillance team at the ING bank. This case study consists of conducting a set of interviews and performing a manual code inspection of the pipeline. The team faces challenges ranging from having insufficient time for operationalising each ML project individually to operating in the highlyregulated fintech context. Their pipeline is able to deploy a single ML model but it does not generalise well to other projects. We present the first version of an application that mitigates these challenges. The application is able to deploy ML models to the development environment at ING and can be operated by data scientists to reduce the effort of operationalising an ML model. Computer Science | Software Technolog
Fatal thrombosis of the portal vein following single-session percutaneous ethanol injection therapy of hepatocellular carcinoma
No full textTwo weeks after percutaneous ethanol injection therapy for hepatocellular carcinoma, performed by injecting 110 mL ethanol in a single session with general anesthesia, a 69-year-old woman with well-compensated liver cirrhosis developed an extensive thrombosis of the whole portal tree that caused severe uncorrectable ascites and progressive deterioration of her general condition, resulting in death 6 weeks after the procedur
'Project smells' - Experiences in Analysing the Software Quality of ML Projects with mllint
No full textMachine Learning (ML) projects incur novel challenges in their development and productionisation over traditional software applications, though established principles and best practices in ensuring the project's software quality still apply. While using static analysis to catch code smells has been shown to improve software quality attributes, it is only a small piece of the software quality puzzle, especially in the case of ML projects given their additional challenges and lower degree of Software Engineering (SE) experience in the data scientists that develop them. We introduce the novel concept of project smells which consider deficits in project management as a more holistic perspective on software quality in ML projects. An open-source static analysis tool mllint was also implemented to help detect and mitigate these. Our research evaluates this novel concept of project smells in the industrial context of ING, a global bank and large software- and data-intensive organisation. We also investigate the perceived importance of these project smells for proof-of-concept versus production-ready ML projects, as well as the perceived obstructions and benefits to using static analysis tools such as mllint. Our findings indicate a need for context-aware static analysis tools, that fit the needs of the project at its current stage of development, while requiring minimal configuration effort from the user. Green Open Access added to TU Delft Institutional Repository ‘You share, we take care!’ – Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.Software EngineeringSoftware Technolog
- …
