Search for: All records

Machine learning has been successfully applied to big data analytics across various disciplines. However, as data is collected from diverse sectors, much of it is private and confidential. At the same time, one of the major challenges in machine learning is the slow training speed of large models, which often requires high-performance servers or cloud services. To protect data privacy while still allowing model training on such servers, privacy-preserving machine learning using Fully Homomorphic Encryption (FHE) has gained significant attention. However, its widespread adoption is hindered by performance degradation. This paper presents our experiments on training models over encrypted data using FHE. The results show that while FHE ensures privacy, it can significantly degrade performance, requiring complex tuning to optimize. 

Abstract Background and objectivesThe optimal iron hypothesis (OIH) posits that risk for infection is lowest at a mild level of iron deficiency. The extent to which this protection results from arms race dynamics in the evolution of iron acquisition and sequestration mechanisms is unclear. We evaluated the OIH with regard to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), an emerging infectious agent. MethodologyWe tested 304 healthcare workers at baseline for iron deficiency (zinc protoporphyrin:heme), anemia (hemoglobin), and SARS-CoV-2 (salivary PCR), and followed them for ~3 months with biweekly SARS-CoV-2 tests. We fit logistic regression models based on Akaike Information Criterion. ResultsAdequate data were available for 199 participants. Iron replete (OR: 2.87, 95% CI: 0.85, 9.75) and anemia (OR: 2.48; 95% CI: 0.82, 7.85) were associated with higher risk for SARS-CoV-2 infection after control for covariates. Logistic regression and Cox proportional hazards models of the SARS-CoV-2 outcome were similar. Anemia (OR: 1.81; 95% CI: 0.88, 3.71) was associated with respiratory symptoms regardless of SARS-CoV-2 infection. Conclusions and implicationsThese findings provide partial support for the OIH: SARS-CoV-2 infection risk was elevated at the high end of the range of iron availability; however, the elevated risk among those with anemia was not, as expected, specific to severe iron deficiency. Narrowly, for COVID-19 epidemiology, these findings accord with evidence that SARS-CoV-2’s ability to establish infection is enhanced by access to iron. More broadly, these findings suggest that the OIH does not hinge on a long history of evolutionary arms race dynamics in access to host iron. 

Online action detection is the task of predicting the action as soon as it happens in a streaming video. A major challenge is that the model does not have access to the future and has to solely rely on the history, i.e., the frames observed so far, to make predictions. It is therefore important to accentuate parts of the history that are more informative to the prediction of the current frame. We present GateHUB, Gated History Unit with Background Suppression, that comprises a novel position-guided gated cross-attention mechanism to enhance or suppress parts of the history as per how informative they are for current frame prediction. GateHUB further proposes Future-augmented History (FaH) to make history features more informative by using subsequently observed frames when available. In a single unified framework, GateHUB integrates the transformer's ability of long-range temporal modeling and the recurrent model's capacity to selectively encode relevant information. GateHUB also introduces a background suppression objective to further mitigate false positive background frames that closely resemble the action frames. Extensive validation on three benchmark datasets, THUMOS, TVSeries, and HDD, demonstrates that GateHUB significantly outperforms all existing methods and is also more efficient than the existing best work. Furthermore, a flow-free version of GateHUB is able to achieve higher or close accuracy at 2.8x higher frame rate compared to all existing methods that require both RGB and optical flow information for prediction. 

Calcific nodules form in the fibrosa layer of the aortic valve in calcific aortic valve disease (CAVD). Glycosaminoglycans (GAGs), which are normally found in the valve spongiosa, are located local to calcific nodules. Previous work suggests that GAGs induce endothelial to mesenchymal transformation (EndMT), a phenomenon described by endothelial cells’ loss of the endothelial markers, gaining of migratory properties, and expression of mesenchymal markers such as alpha smooth muscle actin (α-SMA). EndMT is known to play roles in valvulogenesis and may provide a source of activated fibroblast with a potential role in CAVD progression. In this study, a 3D collagen hydrogel co-culture model of the aortic valve fibrosa was created to study the role of EndMT-derived activated valvular interstitial cell behavior in CAVD progression. Porcine aortic valve interstitial cells (PAVIC) and porcine aortic valve endothelial cells (PAVEC) were cultured within collagen I hydrogels containing the GAGs chondroitin sulfate (CS) or hyaluronic acid (HA). The model was used to study alkaline phosphatase (ALP) enzyme activity, cellular proliferation and matrix invasion, protein expression, and calcific nodule formation of the resident cell populations. CS and HA were found to alter ALP activity and increase cell proliferation. CS increased the formation of calcified nodules without the addition of osteogenic culture medium. This model has applications in the improvement of bioprosthetic valves by making replacements more micro-compositionally dynamic, as well as providing a platform for testing new pharmaceutical treatments of CAVD. 

Calcific aortic valve disease (CAVD) is an active pathobiological process leading to severe aortic stenosis, where the only treatment is valve replacement. Late-stage CAVD is characterized by calcification, disorganization of collagen, and deposition of glycosaminoglycans, such as chondroitin sulfate (CS), in the fibrosa. We developed a three-dimensional microfluidic device of the aortic valve fibrosa to study the effects of shear stress (1 or 20 dyne per cm 2 ), CS (1 or 20 mg mL −1 ), and endothelial cell presence on calcification. CAVD chips consisted of a collagen I hydrogel, where porcine aortic valve interstitial cells were embedded within and porcine aortic valve endothelial cells were seeded on top of the matrix for up to 21 days. Here, we show that this CAVD-on-a-chip is the first to develop human-like calcified nodules varying in calcium phosphate mineralization maturity resulting from high shear and endothelial cells, specifically di- and octa-calcium phosphates. Long-term co-culture microfluidic studies confirmed cell viability and calcium phosphate formations throughout 21 days. Given that CAVD has no targeted therapies, the creation of a physiologically relevant test-bed of the aortic valve could lead to advances in preclinical studies. 

We analyze submissions for homework assignments of 527 students in an upper-level database course offered at the University of Illinois at Urbana-Champaign. The ability to query databases is becoming a crucial skill for technology professionals and academics. Although we observe a large demand for teaching database skills, there is little research on database education. Also, despite the industry's continued demand for NoSQL databases, we have virtually no research on the matter of how students learn NoSQL databases, such as MongoDB. In this paper, we offer an in-depth analysis of errors committed by students working on MongoDB homework assignments over the course of two semesters. We show that as students use more advanced MongoDB operators, they make more Reference errors. Additionally, when students face a new functionality of MongoDB operators, such as texttt$group operator, they usually take time to understand it but do not make the same errors again in later problems. Finally, our analysis suggests that students struggle with advanced concepts for a comparable amount of time. Our results suggest that instructors should allocate more time and effort for the discussed topics in our paper. 

As data grow both in size and in connectivity, the interest to use graph databases in the industry has been proliferating. However, there has been little research on graph database education. In response to the need to introduce college students to graph databases, this paper is the first to analyze students' errors in homework submissions of queries written in Cypher, the query language for Neo4j---the most prominent graph database. Based on 40,093 student submissions from homework assignments in an upper-level computer science database course at one university, this paper provides a quantitative analysis of students' learning when solving graph database problems. The data shows that students struggle the most to correctly use Cypher's WITH clause to define variable names before referencing in the WHERE clause and these errors persist over multiple homework problems requiring the same techniques, and we suggest a further improvement on the classification of syntactic errors. 

Introduction: Calcific aortic valve disease (CAVD) is an active pathological process leading to severe valve calcification. Late-stage CAVD is characterized by increased leaflet stiffness, disorganized collagen bundles and the deposition of glycosaminoglycans, such as chondroitin sulfate (CS), in the fibrosa layer. However, many details of the cellular pathological cascade remain unknown. Animal models such as mice, rabbits, and pigs are used in understanding human CAVD, but mice do not have similar anatomy, rabbits cannot spontaneously develop atherosclerotic lesions, and pigs require long, expensive and complex studies. Here we utilize microfluidic devices of the aortic valve fibrosa to model late-stage CAVD. Hypothesis: We assessed the hypothesis that microfluidic calcification will increase with increased shear rates and CS content. Methods: Valve-on-a-chip devices contained a hydrogel of 1.5 mg/mL collagen I-only healthy controls or 1.5 mg/mL collagen I with 1 mg/mL or 20 mg/mL CS. Porcine aortic valve interstitial cells (PAVIC) were embedded within and endothelial cells (PAVEC) were seeded onto the matrix. Steady shear stress at 1 dyne/cm 2 and 20 dyne/cm 2 were applied using a peristaltic pump for 14 days. Alizarin Red S (ARS), an assay to assess calcium deposition, was used to quantify calcific nodule formation. Scanning electron microscopy with energy dispersive x-ray (SEM/EDX) was used to further analyze sample mineralization. Results: Co-cultures in the presence of increasing shear stress and CS exhibit increased calcific nodule formation compared to static controls, both qualitatively and quantitatively (n≥3). SEM revealed the microstructure of calcified nodules and EDX confirmed calcium phosphate mineralization with physiologically-relevant calcium to phosphorous ratios (Ca/P= 0.88 - 1.4). Conclusions: These results show that in vitro calcification is driven by shear stress in the presence of PAVEC and CS. As seen in ex vivo studies of human calcification, these microfluidic-derived nodules are similarly composed of a range of naturally-occurring calcium phosphates. Given that CAVD has no targeted therapy, the creation of a physiologically relevant model of the aortic valve can provide a test bed for novel therapeutic interventions.