Despite advances in deep learning, neural networks can only learn multiple tasks when trained on them jointly. When arrive sequentially, they lose performance previously learnt tasks. This phenomenon called catastrophic forgetting is a fundamental challenge to overcome before continually from incoming data. In this work, we derive inspiration human memory develop an architecture capable of learning continuously sequentially tasks, while averting forgetting. Specifically, our contributions...

Emotions are physiological states generated in humans reaction to internal or external events. They complex and studied across numerous fields including computer science. As humans, on reading "Why don't you ever text me!" we can either interpret it as a sad angry emotion the same ambiguity exists for machines. Lack of facial expressions voice modulations make detecting emotions from challenging problem. However, increasingly communicate using messaging applications, digital agents gain...

Federated learning (FL) enables distributed computation of machine models over various disparate, remote data sources, without requiring to transfer any individual a centralized location. This results in an improved generalizability and efficient scaling as more sources larger datasets are added the federation. Nevertheless, recent membership attacks show that private or sensitive personal can sometimes be leaked inferred when model parameters summary statistics shared with central site,...

Deep Learning for neuroimaging data is a promising but challenging direction. The high dimensionality of 3D MRI scans makes this endeavor compute and data-intensive. Most conventional methods use 3D-CNN-based architectures with large number parameters require more time to train. Recently, 2D-slice-based models have received increasing attention as they fewer may samples achieve comparable performance. In paper, we propose new architecture BrainAGE prediction. proposed works by encoding each...

A wealth of algorithms centered around (integer) linear programming have been proposed to compute equilibrium strategies in security games with discrete states and actions. However, practice many domains possess continuous state action spaces. In this paper, we consider a space game model infinite-size sets for players present novel deep learning based approach extend the existing toolkit solving games. Specifically, (i) OptGradFP, general algorithm that searches optimal defender strategy...

The widespread use of Artificial Intelligence (AI) in consequential domains, such as health-care and parole decision-making systems, has drawn intense scrutiny on the fairness these methods. However, ensuring is often insufficient rationale for a contentious decision needs to be audited, understood, defended. We propose that attention mechanism can used ensure fair outcomes while simultaneously providing feature attributions account how was made. Toward this goal, we design an...

Umang Gupta, Jwala Dhamala, Varun Kumar, Apurv Verma, Yada Pruksachatkun, Satyapriya Krishna, Rahul Kai-Wei Chang, Greg Ver Steeg, Aram Galstyan. Findings of the Association for Computational Linguistics: ACL 2022.

The amount of biomedical data continues to grow rapidly. However, collecting from multiple sites for joint analysis remains challenging due security, privacy, and regulatory concerns. To overcome this challenge, we use federated learning, which enables distributed training neural network models over sources without sharing data. Each site trains the its private some time then shares parameters (i.e., weights and/or gradients) with a federation controller, in turn aggregates local sends...

Controlling bias in training datasets is vital for ensuring equal treatment, or parity, between different groups downstream applications. A naive solution to transform the data so that it statistically independent of group membership, but this may throw away too much information when a reasonable compromise fairness and accuracy desired. Another common approach limit ability particular adversary who seeks maximize parity. Unfortunately, representations produced by adversarial approaches...

Purely, data-driven large scale image classification has been achieved using various feature descriptors like SIFT, HOG etc. Major milestone in this regards is Convolutional Neural Networks (CNN) based methods which learn optimal as filters. Little attention given to the use of domain knowledge. Ontology plays an important role learning categorize images into abstract classes where there may not be a clear visual connect between category and image, for example identifying mood - happy, sad...

Prediction of human behavior from his/her traits has long been sought by cognitive scientists. Human are often embedded in one's writings. Although some work done on identification essays, very little can be found extracting personality written texts. Psychological studies suggest that extraction and prediction rules a data pursued, several methods have proposed. In the present we used Rough sets to extract for traits. Set is comparatively recent method effective various fields such as...

Over the ages technology has been occupying every field including agriculture. Precision agriculture and Visual data mining uses to apply specific principles of interpret details like when how much fertilizers be used in a particular area (land). Data is process detecting patterns clustering form (certain chunk information) get more precise accurate information. Not only it provides facilitating results but also improves efficiency farmer's productivity, helped qualitative improvement...

Transfer learning has remarkably improved computer vision. These advances also promise improvements in neuroimaging, where training set sizes are often small. However, various difficulties arise directly applying models pretrained on natural images to radiologic images, such as MRIs. In particular, a mismatch the input space (2D vs. 3D MRIs) restricts direct transfer of models, forcing us consider only few MRI slices input. To this end, we leverage 2D-Slice-CNN architecture Gupta et al....

The Kepler satellite has provided photometric timeseries data of unprecedented length, duty cycle and precision. To fully analyse these for the tens thousands stars observed by Kepler, automated methods are a prerequisite. Here we present an procedure to determine period spacing gravity modes in red-giant ascending branch. reside cavity deep interior provide information on conditions stellar core. However, red giants not directly observable surface, hence this method is based...

Deep learning methods trained on brain MRI data from one scanner or imaging protocol can fail catastrophically when tested other sites protocols - a problem known as domain shift. To address this, here we propose adaptation method that trains 3D CycleGAN (cycle-consistent generative adversarial network) to harmonize 5 diverse sources (ADNI, WHIMS, OASIS, AIBL, and the UK Biobank- total of N=4,941 MRIs, age range: 46-96 years). The approach uses 2 generators discriminators generate an image...

Ensuring the privacy of research participants is vital, even more so in healthcare environments. Deep learning approaches to neuroimaging require large datasets, and this often necessitates sharing data between multiple sites, which antithetical objectives. Federated a commonly proposed solution problem. It circumvents need for by parameters during training process. However, we demonstrate that allowing access may leak private information if never directly shared. In particular, show it...

Today, the Internet traffic is crowded by emerging bandwidth hungry multimedia services. These services required dynamic which may be high, low or moderate. For establishing such a connection demand, we propose novel routing and assignment (RBA) algorithm. In order to efficiently utilize network resources reduce blocking probability, various ordering policies based RBA algorithms. RBA, two constraints, wavelength continuity contiguity constraint must satisfied, different complex with...