Patterns and anomalies in big data can help businesses target likely customers, reveal fraud or even predict drug interactions. Unfortunately, these patterns are often not easily observable. To extract the needles of useful information out of haystacks of data, data scientists need increasingly powerful methods of machine learning.

Dr. Aria Nosratinia, the Erik Jonsson Distinguished Professor of electrical and computer engineering at The University of Texas at Dallas, has received two grants from the National Science Foundation totaling $749,492 to uncover relationships hiding in big data via machine learning and to develop methods to keep data communications safe.

“The contribution of my lab is to expand the universe of tools and techniques so we can discover new connections in the data,” said Nosratinia, who is associate department head of electrical and computer engineering in the Erik Jonsson School of Engineering and Computer Science.

Many machine learning and data mining algorithms use graphs, which are simply lists of connections between people, groups or objects. Examples include “friend,” “like” or “follow” relationships in social networks, or the list of videos streamed or marked as favorites in a streaming subscription service.

These mountains of data hide useful information whose extraction belongs to an area known as graph inference. Graph inference has many interesting and useful applications—for example, suggesting movies in a streaming service based on viewing history or purchasing suggestions in online shopping. It also can reveal patterns in the spread of epidemics, or provide insights into the folding of proteins, which is important in understanding how proteins function.

Nosratinia’s work for the first time proposes and analyzes techniques to improve graph inference by absorbing nongraph information, whose efficient blending with graph information was previously not well understood. Examples of non-graph information include a person’s age and residence ZIP code, which are individual attributes.

“In almost every practical application involving graphs, there exist nongraph data of great relevance,” Nosratinia said. “The kind of work we do is further upstream, developing the mathematical models, theory and techniques, but it has widespread applications.”

In several published works, Nosratinia describes the mathematical models he and members of his lab have developed that can improve the estimation of the information hidden in the graph with the aid of side information. Nosratinia and co-author Hussein Saad Ph.D.”19, now a senior engineer with Qualcomm Inc., recently analyzed how to identify a small cluster or community hidden in a large graph. Their latest work appeared in the December 2020 issue of the journal IEEE Transactions on Information Theory.

The second component of Nosratinia’s research addresses data security. His work harnesses the natural variations of wireless channels to provide layers of security for data transmission. This area of work, known as physical layer security, aims to leverage the imperfections of the communication channel as a tool for security. Part of this research is aimed at developing techniques for making the presence of electronic communication undetectable to cybercriminals.

“To give a simple example, a password works by leveraging the difference between what is known by a legitimate user versus cybercriminals who want to steal information,” Nosratinia said. “Our work creates, amplifies and analyzes statistical asymmetry of information against adversaries in ways that do not involve passwords or keys, and uses them for securing communications.”

The post Researcher develops better tools for understanding, protecting big data appeared first on Artificial Intelligence.

New Delhi: Scientists at the Institute of Advanced Study in Science and Technology (IASST), Guwahati, an autonomous institute of the Department of Science & Technology, Govt of India, have developed an artificial intelligence (AI) based algorithms as an aid to rapid diagnosis and prediction of oral squamous cell carcinoma.

The framework developed by the research group at the Central Computational and Numerical Sciences Division, IASST led by Dr. Lipi B Mahanta, will also help grading of oral squamous cell carcinoma.

An indigenous dataset was developed by the scientists through collaborations to make for the unavailability of any benchmark oral cancer dataset for the study. Exploring different state-of-the-art AI techniques and playing with their proposed method, the scientists have gained unprecedented accuracy in oral cancer grading. The study was conducted applying two approaches through the application of transfer learning using a pre-trained deep convolutional neural network (CNN).

Four candidate pre-trained models, namely Alexnet, VGG-16, VGG-19, and Resnet-50, were chosen to find the most suitable model for the classification problem, and a proposed CNN model developed to fit the problem. Although the highest classification accuracy of 92.15% was achieved by the Resnet-50 model, the experimental findings highlight that the proposed CNN model outperformed the transfer learning approaches displaying accuracy of 97.5%. The work has been published in the journal Neural Networks.

As of now, the group is set for converting the algorithm into proper software to move on to carry out field trials. This is the next challenge that the group is prepared to meet, considering the ever-present gap between the health and IT sectors. Dr. Mahanta aspires for all the advanced infrastructural support to meet these challenges and feels that the software needs to be actively tested in hospitals, to make it truly robust, more accurate, and real-time worthy.

Around 16.1% of all cancers amongst men and 10.4% amongst women are oral cancer, and the picture is all the more alarming in NE India. Oral cavity cancers are also known to have a high recurrence rate compared to other cancers due to the high consumption of betel nut and tobacco.

This cancer group is characterized by epithelial squamous tissue differentiation and aggressive tumour growth, disrupting the basement membrane of the inner cheek region and thus can be graded by Broder’s histopathological system as well-differentiated SCC (WDSCC), moderately differentiated SCC (MDSCC) and poorly differentiated SCC (PDSCC). The cellular morphometry highlighting the tumour growth displays a very minute histological difference separating the three classes, which are very hard to capture by the human eye. It has remained elusive due to its highly similar histological features, which even pathologists find difficult to classify.

The advent of deep learning in AI holds an extraordinary prospect in digital image analysis to serve as a computational aid in the diagnosis of cancer, thus providing help in timely and effective prognosis and multi-modal treatment protocols for cancer patients and reducing the operational workload of pathologists while enhancing management of the disease.

The post IASST develops an artificial intelligence-based computer diagnosis framework for rapid and accurate diagnosis of oral cancers appeared first on Artificial Intelligence.

Chatbots and virtual assistants that don’t provide the right answers to customer queries can result in a frustrating digital experiences for customers – particularly those chatbots programmed without deep learning capabilities.

Deep learning is essential for automated chatbots to understand natural language questions and to provide the right answers, which is something that AI-powered search firm Lucidworks has taken on board. The company recently launched Smart Answers for chatbots and virtual assistants – a solution the company says will help to avoid frustrating digital support experiences and long customer support wait times.

According to Lucidworks, companies rely on digital portals to provide information to users, whether digital commerce customers looking for product information before purchase, employees hunting for an HR document, or someone looking for an airline’s updated cancellation policies. Information is often scattered across disparate silos and is impossible for a user to locate using natural language questions.

Smart Answers on Lucidworks Fusion aims to help employees and customers resolve issues via chatbots or virtual assistants without the need for additional support or ‘irrelevant’ search results.

Lucidworks CEO Will Hayes explains that customers expect more conversational interactions with a chatbot or virtual assistant that understands natural language questions, and is quickly provides the correct answers. 

“In the current environment, it’s important to meet customers where they are, especially if they’re moving from a real life experience to digital,” he says.

“Our work over the years with Fusion has been focused on understanding a user’s intent. Digging for answers wears their patience, wastes their time, and can even motivate them to seek out a competitor. Being able to understand what your users are asking for in their own words and returning the best answer instantly, allows companies to provide a natural conversational experience digitally. With Smart Answers, we’re cutting down time-to-resolution, increasing customer retention, and powering conversational experiences for users.”

The company cites Red Hat as one of its customers.

“We conducted an A/B test where we introduced a self-solve based homepage to some customers. There was an increase in traffic that confirmed that customers are really motivated to self-solve and we saw a seven per cent decrease in support case creation for customers who were given the self-solve homepage,” comments Red Hat principal software engineer Manikandan Sivanesan.

Lucidworks states that Smart Answers on Lucidworks Fusion enhances conversational applications to help users help themselves by delivering immediate and contextual responses that drive engagement and satisfaction.

The post Lucidworks develops deep learning solution to make chatbots smarter appeared first on Artificial Intelligence.

It has taken the services through a proof of concept phase and is now planning to work with Microsoft and Amazon Web Services (AWS) to make them widely available over the next six months.

Chris Suter (pictured) head of digital platforms and innovation at NHSBSA, said it is also looking to further validate the findings and “take full advantage of this solution into other form types”.

The effort was launched early in 2019 as one of a series of concept projects aimed at understanding how artificial intelligence and machine learning can support the health service.

It was focused on prescriptions due the heavy cost of processing, which averages £20 each and amounts to £9.4 billion per year – approximately 10% of NHS costs.

NHSBSA has been working with Microsoft and AWS on the programme, with a focus on the reading accuracy of any solutions, looking for improvements in processes and identifying future use cases. It also placed an emphasis on complying with data governance guidance such as the Caldicott Principles and Code of Conduct for the use of AI in the NHS.

Several stages

So far it has produced microservices for a number of stages of handling prescriptions, including image reprocessing, the extraction of data through machine learning models, validation of the data captured, a data analytics module, and image processing and storage at scale.

This has helped to equip the organisation with a scanning facility, secure cloud platform and the skills to programme and train machine learning models.

Suter said this could be deployed in other processes that rely heavily on paper, including the management of pensions and student bursaries, and that it could be extended into other parts of the NHS.

“One of the key outcomes of this activity is the future potential of the solution we have created,” he said. “Not only will it solve current problems, but it also has been designed to be easily transferable to other paper based scenarios not only in the NHSBSA but potentially within the entire NHS family.”

The post NHSBSA develops machine learning microservices for prescriptions appeared first on Artificial Intelligence.

ABERDEEN PROVING GROUND, Md. — A big data approach to neuroscience promises to significantly improve our understanding of the relationship between brain activity and performance.

To date, there have been relatively few attempts to use a big-data approach within the emerging field of neurotechnology. In this field, the few attempts at meta-analysis (analysis across multiple studies) combine only the results from individual studies rather than the raw data. A new study is one of the first to combine data across a diverse set of experiments to identify patterns of brain activity that are common across tasks and people.

The Army in particular is interested in how the cognitive state of Soldiers can affect their performance during a mission. If you can understand the brain, you can predict and even enhance cognitive performance.

Researchers from the U.S. Army Combat Capabilities Development Command’s Army Research Laboratory teamed with the University of Texas at San Antonio and Intheon Labs to develop a first-of-its-kind mega-analysis of brain imaging data–in this case electroencephalography, or EEG.

In the two-part paper, they aggregate the raw data from 17 individual studies, collected at six different locations, into a single analytical framework, with their findings published in a series of two papers in the journal NeuroImage (see Related Links below). The individual studies included in this analysis encompass a diverse set of tasks such simulated driving and visual search.

“The vast majority of human neuroscientific studies use a very small number of participants employed in very specific tasks,” said Dr. Jonathan Touryan, an Army scientist and co-author of the paper. “This limits how well the results from any single study can be generalized to a broader population and a larger range of activities.”

Mega-analysis of EEG is extremely challenging due to the many types of hardware systems (properties and configuration of the electrodes), the diversity of tasks, how different datasets are annotated, and the intrinsic variability between individuals and within an individual over time, Touryan said.

These sources of variability make it difficult to find robust relationships between brain and behavior. Mega-analysis seeks to address this by aggregating large, heterogeneous datasets to identify universal features that link neural activity, cognitive state and task performance.

Next-generation neurotechnologies will require a thorough understanding of this relationship in order to mitigate deficits or augment performance of human operators. Ultimately, these neurotechnologies will enable autonomous systems to better understand the Soldier and facilitate communications within multi-domain operations, he said.

To combine the raw data from the collection of studies, the researchers developed Hierarchical Event Descriptors (HED tags) — a novel labeling ontology that captures the wide range of experimental events encountered in diverse datasets. This HED tag system was recently adopted into the Brain Imaging Data Structure international standard, one of the most common formats for organizing and analyzing brain data, Touryan said.

The research team also developed a fully automated processing pipeline to perform large-scale analysis of their high-dimensional time-series data–amounting to more than 1,000 recording sessions.

Much of this data was collected over the last 10 years through the U.S. Army’s Cognition and Neuroergonomics Collaborative Technology Alliance and is now available in an online repository for the scientific community (see Related Links below). The U.S. Army continues to use this data to develop human-autonomy adaptive systems for both the Next Generation Combat Vehicle and Soldier Lethality Cross-Functional Teams.

The post Army develops big data approach to neuroscience appeared first on Artificial Intelligence.

To help geoscientists take advantage of machine learning and deep learning technology, CGG GeoSoftware has developed a machine learning ecosystem. It provides open access to data within its geophysical and petrophysical applications. Python-scripted machine learning lets users get their hands dirty if they like to tinker under the hood, or users can select pre-built recipes. Many tasks can now be completed more quickly and with more detailed results; for example, well log editing, petrophysical analysis, facies classification, and reservoir property prediction. Meanwhile, deep neural networks provide benefits for tasks as varied as reservoir quality assessment and near-surface characterisation.

Even if geoscientists are not using machine learning personally, it is increasingly involved across various aspects of geoscience projects and workflows around them.

Before the industry gets to the point where it can truly benefit from big data analytics and take full advantage of machine learning, there is a need to reach a minimum common denominator in terms of the data itself. Recent efforts have seen the liberation of huge volumes of data from legacy formats, migrating to new data management platforms, including an increasing mix of cloud storage. CGG Smart Data Solutions help to ease this digital transition with end-to-end services, from expert upcycling of legacy data into the cloud to the deployment of their modern and flexible GeoTrove data management platform.

Integration and interoperability of geoscience data becomes important to really take advantage of data analytics and machine learning applications. CGG has spent the last few years gaining valuable experience while taking its geological library into the digital realm, using a proprietary taxonomy and ontology to create a unique framework for its GeoVerse data set. Meanwhile, its multi-client seismic library is now assessable through its new GeoStore portal, with controlled access to historical client entitlement data. Upload to the cloud is underway for the entire multi-client seismic library.

The cloud offers more than just data storage – cloud computing provides scalable and flexible solutions to compute-intensive reservoir characterisation workflows and very large projects. Through its technical collaboration with Microsoft, CGG’s latest GeoSoftware releases run seamlessly in the Microsoft Azure Cloud Environment, with other major cloud platforms soon to follow.

The post CGG GeoSoftware develops machine learning ecosystem appeared first on Artificial Intelligence.