An introduction to distributed training of deep neural networks for segmentation tasks with large seismic datasets

Deep learning applications are drastically progressing in seismic processing and interpretation tasks . The majority of approaches subsample data and restrict model sizes to minimise computational requirements . Subsampling the data risks losing vital spatio-temporal information which couldaid training whilst restricting model sizes can impact model performance .…

Distributionally Robust Federated Averaging

In this paper, we study communication efficient distributed algorithms fordistributionally robust federated learning via periodic averaging with adaptivesampling . We propose a Distributionally Robust Federated Federated Averaging(DRFA) algorithm that employs a novel snapshotting scheme to approximate theaccumulation of history gradients of the mixing parameter .…

ISALT Inference based schemes adaptive to large time stepping for locally Lipschitz ergodic systems

Efficient simulation of SDEs is essential in many applications, particularly for ergodic systems that demand efficient simulation of both short-timedynamics and large-time statistics . We introduce a framework to constructinference-based schemes adaptive to large time-steps (ISALT) from data . We explore the use of numerical schemes (such as theEuler-Maruyama, a hybrid RK4, and an implicit scheme) to derive informed basisfunctions, leading to a parameter inference problem .…

BGK models for inert mixtures comparison and applications

Consistent BGK models for inert mixtures are compared to hydrodynamic limits that can be derived from indifferent collision-dominated regimes . Application to the plane shock wave in a binary mixture of noble gases is also presented . The comparison is carried out bothanalytically and numerically, for the latter using an asymptotic preservingsemi-Lagrangian scheme .…

Batched Neural Bandits

The BatchNeuralUCB algorithm combines neuralnetworks with optimism to address the exploration-exploitation tradeoff while keeping the total number of batches limited . We prove that it achieves the same regret as the fully sequential version while reducing the number of policy updates considerably .…

Diffusion Earth Mover s Distance and Distribution Embeddings

Diffusion Earth Mover’s Distance(EMD) is more accurate than similarly fast algorithms such astree-based EMDs . Diffusion EMD is fully differentiable, making itenable to future uses in gradient-descent frameworks such as deep neuralnetworks . The method is applicable to all datasets that are massivelycollected in parallel in many medical and biological systems .…

On a Network SIS Epidemic Model with Cooperative and Antagonistic Opinion Dynamics

We propose a mathematical model to study coupled epidemic and opiniondynamics in a network of communities . Our model captures SIS epidemic dynamics whose evolution is dependent on the opinions of the communities toward the epidemic . We propose an Opinion-DependentReproduction Number to characterize the mutual influence between epidemicspreading and opinion dissemination over the networks .…

Metal Oxide Sensor Array for Selective Gas Detection in Mixtures

We present a monolithic, microfabricated, metal-oxide semiconductor (MOS) sensor array in conjunction with a machine learning algorithm to determine unique fingerprints of individual gases within homogenous mixtures . The arraycomprises four different metal oxides, and is engineered for independenttemperature control and readout from each individual pixel in a multiplexed fashion .…

Stochastic Aggregation in Graph Neural Networks

Graph neural networks (GNNs) manifest pathologies including over-smoothing and limited discriminating power as a result of suboptimally expressiveaggregating mechanisms . We present a unifying framework for stochasticaggregation (STAG) in GNNs . Noise is (adaptively) injected into theaggregation process from the neighborhood to form node embeddings .…

FASA Feature Augmentation and Sampling Adaptation for Long Tailed Instance Segmentation

Feature Augmentation and Sampling Adaptation (FASA) is a fast, generic method that can be easily plugged into standard or long-tailed segmentationframeworks, with consistent performance gains and little added cost . FASAdoes not require any elaborate loss design, and removes the need forinter-class transfer learning that often involves large cost andmanually-defined head/tail class groups .…

Lie Group integrators for mechanical systems

Lie group integrators have become amethod of choice in many application areas . They include multibody dynamics, shape analysis, data science, image registration and biophysical simulations . The theory is illustrated by applying the methods totwo nontrivial applications in mechanics .…

Automatic Story Generation Challenges and Attempts

The scope of this survey paper is to explore the challenges in automaticstory generation . We hope to contribute in the following ways: Explore howprevious research in story generation addressed those challenges . Discuss future research directions and new technologies that may aid more advancements .…

A flapping feathered wing powered aerial vehicle

An aerial vehicle powered by flapping feathered wings was designed, developed and fabricated . Different from legacy flapping-wing aerial vehicles withmembrane wings, the new design uses authentic bird feathers to fabricate wings . In field tests, a radio-controlled electric-powered aerial vehicle successfully took off, flew up to 63.88 s and landedsafely .…

On continual single index learning

In this paper, we generalize the problem of single index model to the context of continual learning in which a learner is challenged with a sequence of tasks one by one and the dataset of each task is revealed in an online fashion .…

SCD A Stacked Carton Dataset for Detection and Segmentation

Carton detection is an important technique in the automatic logistics system . Images are collected from the internet and several warehourses, and objects are labeled using per-instancesegmentation for precise localization . There are totally 250,000 instance masksfrom 16,136 images . The improvement of AP on MS COCO and PASCAL VOC is 1.8% – 2.2% and 3.4% – 4.3% respectively.…

Research on False Data Injection Attacks in VSC HVDC Systems

The false data injection (FDI) attack is a crucial form of cyber-physical security problems facing cyber power systems . There is noresearch revealing the problem of FDI attacks facing voltage source converterbased high voltage direct current transmission (VSC-HVDC) systems . And finally, the modified IEEE-14 bus system is used to demonstrate that attackers are capable of disrupting the operation security of converter stations in VSC- HVDC systems by FDI attack strategies .…

Lossless Compression of Efficient Private Local Randomizers

Locally Differentially Private (LDP) Reports are commonly used for collection of statistics and machine learning in the federated setting . LDP reports are known to have relatively little information about the user’s data due to randomization . Several schemes are known that exploit this fact to design low-communication versions of LDP algorithms but all of them do so at the expense of a significant loss in utility .…

Adversarial Robustness with Non uniform Perturbations

Robustness of machine learning models is critical for security related applications . We propose using characteristics of the empirical datadistribution, both on correlations between the features and the importance of the features themselves . The key idea of our proposed approach is to enable non-uniformperturbations that can adequately represent these feature dependencies during training .…

The non positive circuit weight problem in parametric graphs a fast solution based on dioid theory

In this paper, we design an algorithm thatsolves the Non-positive Circuit weight Problem (NCP) on this class ofparametric graphs . The proposed algorithm isbased on max-plus algebra and formal languages and runs faster than otherexisting approaches . It achieves strongly polynomial time complexity$\mathcal{O}(n^4)$ (where $n$ is the number of nodes in the graph) The proposed algorithms are based on max plus algebra, and run faster than existing approaches .…

Attestation Infrastructures for Private Wallets

In this paper we focus on one part of the trust infrastructures needed for the future virtual assets industry . Our focus is on regulated private wallets utilizing trusted hardware, and the capability of the wallet toyield attestation evidence suitable to address requirements in severaluse-cases, such as asset insurance and regulatory compliance .…

Decentralized conjugate gradients with finite step convergence

The decentralized solution of linear systems of equations arises as asubproblem in optimization over networks . Typical examples include the KKTsystem corresponding to equality constrained quadratic programs in distributedoptimization algorithms or in active set methods . This note presents a tailoredstructure-exploiting decentralized variant of the conjugate gradient method .…

Robust SleepNets

State-of-the-art convolutional neural networks excel in machine learning tasks such as face recognition, but suffers significantly when adversarial attacks are present . In this study, we investigate eyeclosedness detection to prevent vehicle accidents related to driverdisengagements and driver drowsiness . We develop two models to detect eye closedness:first model on eye images and a second model on face images .…

No Regret Algorithms for Private Gaussian Process Bandit Optimization

The widespread proliferation of data-driven decision-making has ushered in interest in the design of privacy-preserving algorithms . We propose a solution for differentially private GP bandit optimization that combines a uniform kernelapproximator with random perturbations . For twospecific DP settings – joint and local differential privacy, we provide algorithms based on efficient quadrature Fourier feature approximators that are computationally efficient and provably no-regret for popular stationary kernel functions .…