Incorporating higher-order optimization functions, such as Levenberg-Marquardt (LM) have revealed better generalizable solutions for deep learning problems. However, these higher-order optimization functions suffer from very large processing time and training complexity especially as training datasets become large, such as in multi-view classification problems, where finding global optima is a very costly problem. To solve this issue, we develop a solution for LM-enabled classification with, to the best of knowledge first-time implementation of hinge loss, for multiview classification. Hinge loss allows the neural network to converge faster and perform better than other loss functions such as logistic or square loss rates. We prove our method by experimenting with various multiclass classification challenges of varying complexity and training data size. The empirical results show the training time and accuracy rates achieved, highlighting how our method outperforms in all cases, especially when training time is limited. Our paper presents important results in the relationship between optimization and loss functions and how these can impact deep learning problems.

With advances in deep learning, exponential data growth and increasing model complexity, developing efficient optimization methods are attracting much research attention. Several implementations favor the use of Conjugate Gradient (CG) and Stochastic Gradient Descent (SGD) as being practical and elegant solutions to achieve quick convergence, however, these optimization processes also present many limitations in learning across deep learning applications. Recent research is exploring higher-order optimization functions as better approaches, but these present very complex computational challenges for practical use. Comparing first and higher-order optimization functions, in this paper, our experiments reveal that Levemberg-Marquardt (LM) significantly supersedes optimal convergence but suffers from very large processing time increasing the training complexity of both, classification and reinforcement learning problems. Our experiments compare off-the-shelf optimization functions(CG, SGD, LM and L-BFGS) in standard CIFAR, MNIST, CartPole and FlappyBird experiments.The paper presents arguments on which optimization functions to use and further, which functions would benefit from parallelization efforts to improve pretraining time and learning rate convergence.

In the past few years there has been a growing desire to provide more built in functionality to protect user communications from eavesdropping. An example of this is DNS over HTTPS (DoH) which can be used to protect user privacy, confidentiality and against spoofing attacks. Since its first popularity in 2018 as used in browsers, there is much further study to test the effectiveness of DoH in protection schemes and whether it is possible to detect the protocol over the web. Detecting DoH traffic among normal web traffic is also a major challenge for network admins to allow filtering of malicious traffic flows. In this paper, we investigate machine learning classification to study the detection of DoH traffic and further analyze the key feature characteristics in the protocol behavior to help researchers build credibility in the DoH protocol detection. Our study reveals key features and statistical relationships among DoH test runs on the Alexa-recommended 100 most-used websites using three different DoH servers, showing up to 98% test accuracy in our built classifier.

With advent of quantum internet, it becomes crucial to find novel ways to connect distributed quantum testbeds and develop novel technologies and research that extend innovations in managing the qubit performance. Numerous emerging technologies are focused on quantum repeaters and specialized hardware to extend the quantum distance over special-purpose channels. However, there is little work that utilizes current network technology, invested in optic technologies, to merge with quantum technologies. In this paper we argue for an AI-enabled control that allows optimized and efficient conversion between qubit and photon energies, to enable optic and quantum devices to work together. Our approach integrates AI techniques, such as deep reinforcement learning algorithms, with physical quantum transducer to inform real-time conversion between the two wavelengths. Learning from simulated environment, the trained AI-enabled transducer will lead to optimal quantum transduction to maximize the qubit lifetime.

Network performance monitoring collects heterogeneous data suchas network flow data to give an overview of network performance,and other metrics, necessary for diagnosing and optimizing servicequality. However, due to disparate and heterogeneity, to obtainmetrics and visualize entire data from several devices, engineershave to log into multiple dashboards.In this paper we present NetGraf, a complete end-to-end networkmonitoring stack, that uses open-source network monitoring toolsand collects, aggregates, and visualizes network measurements on asingle easy-to-use real-time Grafana dashboard. We develop a novelNetGraf architecture and can deploy it on any network testbed suchas Chameleon Cloud by single easy-to-use script for a full view ofnetwork performance in one dashboard.This paper contributes to the theme of automating open-sourcenetwork monitoring tools software setups and their usability forresearchers looking to deploy an end-to-end monitoring stack ontheir own testbeds.

FLAME allows complex models to be automatically parallelised on High Performance Computing (HPC) grids enabling large number of agents to be simulated over short periods of time. Modellers are hindered by complexities of porting models on parallel platforms and time taken to run large simulations on a single machine, which FLAME overcomes. Three case studies from different disciplines were modelled using FLAME, and are presented along with their performance results on a grid.

Wide area networking infrastructures (WANs), particularly science and research WANs, are the backbone for moving large volumes of scientific data between experimental facilities and data centers. With demands growing at exponential rates, these networks are struggling to cope with large data volumes, real-time responses, and overall network performance. Network operators are increasingly looking for innovative ways to manage the limited underlying network resources. Forecasting network traffic is a critical capability for proactive resource management, congestion mitigation, and dedicated transfer provisioning. To this end, we propose a nonautoregressive graph-based neural network for multistep network traffic forecasting. Specifically, we develop a dynamic variant of diffusion convolutional recurrent neural networks to forecast traffic in research WANs. We evaluate the efficacy of our approach on real traffic from ESnet, the U.S. Department of Energy's dedicated science network. Our results show that compared to classical forecasting methods, our approach explicitly learns the dynamic nature of spatiotemporal traffic patterns, showing significant improvements in forecasting accuracy. Our technique can surpass existing statistical and deep learning approaches by achieving approximately 20% mean absolute percentage error for multiple hours of forecasts despite dynamic network traffic settings.

Understanding flow traffic patterns in networks, such as the Internet or service provider networks, is crucial to improving their design and building them robustly. However, as networks grow and become more complex, it is increasingly cumbersome and challenging to study how the many flow patterns, sizes and the continually changing source-destination pairs in the network evolve with time. We present Netostat, a visualization-based network analysis tool that uses visual representation and a mathematics framework to study and capture flow patterns, using graph theoretical methods such as clustering, similarity and difference measures. Netostat generates an interactive graph of all traffic patterns in the network, to isolate key elements that can provide insights for traffic engineering. We present results for U.S. and European research networks, ESnet and GEANT, demonstrating network state changes, to identify major flow trends, potential points of failure, and bottlenecks.

Sensor and smart phone technologies present opportunities for data explosion, streaming and collecting from heterogeneous devices every second. Analyzing these large datasets can unlock multiple behaviors previously unknown, and help optimize approaches to city wide applications or societal use cases. However, collecting and handling of these massive datasets presents challenges in how to perform optimized online data analysis 'on-the-fly', as current approaches are often limited by capability, expense and resources. This presents a need for developing new methods for data management particularly using public clouds to minimize cost, network resources and on-demand availability. This paper presents an implementation of the lambda architecture design pattern to construct a data-handling backend on Amazon EC2, providing high throughput, dense and intense data demand delivered as services, minimizing the cost of the network maintenance. This paper combines ideas from database management, cost models, query management and cloud computing to present a general architecture that could be applied in any given scenario where affordable online data processing of Big Datasets is needed. The results are presented with a case study of processing router sensor data on the current ESnet network data as a working example of the approach. The results showcase a reduction in cost and argue benefits for performing online analysis and anomaly detection for sensor data.

Globally distributed scientific experiments involve movement of massive data volumes and many collaborators performing distributed data analysis. With complex workloads and heterogeneous resources, each user may desire certain behavior characteristics for their network paths. In this paper, we present the iNDIRA tool, which interacts with SDN north-bound interfaces to enable intent-based networking. It provides reliable, simple, and technology-agnostic communication between users and networks. Focusing particularly on science applications, iNDIRA uses natural language processing to construct semantic RDF graphs to understand, interact, and create the required network services. The technical challenges addressed by iNDIRA are: (1) development of a high-level descriptive language to query network-application requirements, (2) provides keyword identification and condition checking based on user profiles and topology details, (3) allows user negotiation based on the current network state, and (4) integrates network provisioning and service tools used by the application. iNDIRA is implemented on the ESnet network, where it interacts with OpenNSA (aka the NSI client) and Globus data transfer tools, to build complex cross-domain network paths for heterogeneous science applications, and perform secure data transfer. We argue that iNDIRA's approach presents users with an alternative approach to interact and communicate their network demands, allowing seamless network service integration.