In the study of various diseases, heterogeneity among patients usually leads to different progression patterns and may require types therapeutic intervention. Therefore, it is important patient subtyping, which grouping into disease characterizing subtypes. Subtyping from complex data challenging because information temporal dynamics. Long-Short Term Memory (LSTM) has been successfully used in many domains for processing sequential data, recently applied analyzing longitudinal records. The...

Many data mining applications involve a set of related learning tasks. Multi-task (MTL) is paradigm that improves generalization performance by transferring knowledge among those MTL has attracted so much attention in the community, and various algorithms have been successfully developed. Recently, distributed also studied for tasks whose across different geographical regions. One prominent challenge frameworks to maintain privacy data. The may contain sensitive private information such as...

Heterogeneous hyper-networks is used to represent multi-modal and composite interactions between data points. In such networks, several different types of nodes form a hyperedge. hyper-network embedding learns distributed node representation under complex while preserving the network structure. However, this challenging task due multiple modalities interactions. study, deep approach proposed embed heterogeneous attributed with complicated non-linear relationships. particular, fully-connected...

Many real-world machine learning applications involve several tasks which are inter-related. For example, in healthcare domain, we need to learn a predictive model of certain disease for many hospitals. The models each hospital may be different because the inherent differences distributions patient populations. However, also closely related nature modeling same disease. By simultaneously all tasks, multi-task (MTL) paradigm performs inductive knowledge transfer among improve generalization...

Sign languages are visual used as the primary communication medium for Deaf community. The signs comprise manual and non-manual articulators such hand shapes, upper body movement, facial expressions. Language Recognition (SLR) aims to learn spatial temporal representations from videos of signs. Most SLR studies focus on features often extracted shape dominant or entire frame. However, expressions combined with gestures may also play a significant role in discriminating context represented...

Early diagnosisof Alzheimer's disease plays a crucial role in treatment planning that might slow down the disease's progression. This problem is commonly posed as classification task performed by machine learning and deep techniques. Although data-driven techniques set state-of-the-art many domains, scale of available datasets research not sufficient to learn complex models from patient data. study proposes simple yet promising framework predict conversion Mild Cognitive Impairment (MCI)...

One major limitation of linear models is the lack capability to capture predictive information from interactions between features. While introducing high-order feature interaction terms can overcome this limitation, approach tremendously increases model complexity and imposes significant challenges in learning against overfitting. In paper, we proposed a novel Multi-Task Interaction Learning~(MTIL) framework exploit task relatedness interactions, which provides better generalization...

Delayed drug safety insights can impact patients, pharmaceutical companies, and the whole society. Post-market surveillance plays a critical role in providing insights, where real world evidence such as spontaneous reporting systems (SRS) series of disproportional analysis serve cornerstone proactive predictive surveillance. However, they still face several challenges including concomitant drugs confounders, rare adverse reaction (ADR) detection, data bias, under-reporting issue. In this...

Electronic health records (EHRs) capture comprehensive patient information in digital form from a variety of sources. Increasing availability EHRs has facilitated development data and visual analytic tools for healthcare analytics, such as clinical decision support care management systems. Many are used to investigate fundamental problems, study population, exploring complicated interactions among patients their medical histories, extracting structured phenotypes characterizing the...

Learning a robust and invariant representation of various unwanted factors in sign language recognition (SLR) applications is essential. One the that might degrade performance lack signer diversity training datasets, causing dependence on singer's identity during learning. Consequently, capturing signer-specific features hinders generalizability SLR systems. This study proposes feature disentanglement framework comprising convolutional neural network (CNN) long short-term memory (LSTM) based...

Over the past decade a wide spectrum of machine learning models have been developed to model neurodegenerative diseases, associating biomarkers, especially non-intrusive neuroimaging markers, with key clinical scores measuring cognitive status patients. Multi-task (MTL) has commonly utilized by these studies address high dimensionality and small cohort size challenges. However, most existing MTL approaches are based on linear suffer from two major limitations: 1) they cannot explicitly...

Principal component analysis (PCA) is a dimensionality reduction and data tool commonly used in many areas. The main idea of PCA to represent high-dimensional with few representative components that capture most the variance present data. However, there an obvious disadvantage traditional when it applied analyze where interpretability important. In applications, features have some physical meanings, we lose ability interpret principal extracted by conventional because each linear combination...

Adversarial Training (AT) aims to alleviate the vulnerability of deep neural networks adversarial perturbations. However, AT techniques struggle maintain performance on natural samples while improving model’s robustness. The absence perturbation diversity in generated during training degrades generalizability robust models, causing overfitting particular perturbations and a decrease performance. This study proposes an framework that augments directions from single-step attack address...

Alzheimer's disease is the most common cause of dementia that affects millions lives worldwide. Investigating underlying causes and risk factors essential to prevent its progression. Mild Cognitive Impairment (MCI) considered an intermediate stage before disease. Early prediction conversion from MCI crucial take necessary precautions for decelerating progression developing suitable treatments. In this study, we propose a deep learning framework discover variables which are identifiers...

Deep neural networks have been successful in various domains, such as computer vision and natural language processing. On the other hand, researchers discovered a vulnerability of convolutional to samples with imperceptible perturbations, also known as, adversarial perturbations. It has observed that perturbations can alter predictions deep model. One most common approaches increase robustness models is training. However, training often suffers from degradation generalization performance. In...

We study the problem of similarity learning and its application to image retrieval with large-scale data. The between pairs images can be measured by distances their high dimensional representations, appropriate is often addressed distance metric learning. However, requires learned a PSD matrix, which computational expensive not necessary for ranking problem. On other hand, bilinear model shown more flexible task, hence, we adopt it learn matrix estimating pairwise similarities under...

This work aims to classify the changes in head pose of a user sitting front screen by using estimated rotation. Considered classes include ∓15, ∓30, ve ∓ 45 degree pan, tilt and combinations these poses. SIFT flow algorithm is used for motion estimation. Two dimensional feature vectors are extracted calculating magnitude angle vectors. Classification has been performed Support Vector Machine Naive Bayesian classifiers. Test results reported on Pointing'04 database demonstrate that enable us...