U.S. Department of Energy

Pacific Northwest National Laboratory

Seminar Series

Ontologies for the Modern Age

Dr. Deborah McGuinness Tetherless World Senior Constellation Chair and Professor of Computer, Cognitive, and Web Sciences
Rensselaer Polytechnic Institute
Dr. Deborah McGuinness Tetherless World Senior Constellation Chair and Professor of Computer, Cognitive, and Web Sciences
Rensselaer Polytechnic Institute
Wednesday, June 7, 2017

Ontologies are seeing a resurgence of interest and usage as big data proliferates, machine learning advances, and integration of data becomes more paramount. The previous models of sometimes labor-intensive, centralized ontology construction and maintenance do not mesh well in today’s interdisciplinary world that is in the midst of a big data, information extraction, and machine learning explosion.   

Learning Nanomaterial Processes from Big Image Data

Dr. Chiwoo Park Assistant Professor, Department of Industrial and Manufacturing Engineering
Florida State University
Dr. Chiwoo Park Assistant Professor, Department of Industrial and Manufacturing Engineering
Florida State University
Friday, May 19, 2017

In this talk, we discuss related research issues and present a big data framework that addresses the issues, including in-line image understanding, real-time event tracking, and longitudinal and cross sectional analysis of the events over time.

Applying Topology In Visualization: From Atmospheric Phenomena to Battery Materials

Dr. Bei Wang Assistant Professor, School of Computing
University of Utah
Dr. Bei Wang Assistant Professor, School of Computing
University of Utah
Wednesday, May 17, 2017

In this talk, I will describe a few recent research activities involving the application of topology in data analysis and visualization.

Improving Reinforcement Learning with Human Input

Matthew Taylor
Washington State University
Matthew Taylor
Washington State University
Monday, April 17, 2017

This talk will discuss a selection of our recent work that improves reinforcement learning by leveraging demonstrations and reward feedback from imperfect users, with an emphasis on how interactive machine learning can be extended to best leverage the unique abilities of both computers and humans.

Engineering Swarms and Beyond

Dr. Sanza Kazadi Director of Student Inquiry and Research
Illinois Mathematics and Science Academy
Dr. Sanza Kazadi Director of Student Inquiry and Research
Illinois Mathematics and Science Academy
Wednesday, March 15, 2017

Swarm engineering is a discipline that has largely focused on the description by biologists of swarm dynamics in nature and on development of computational and robotic swarms by engineers. In both cases, swarms are generally developed in an ad hoc manner, requiring a guess-and-check set of steps. We discuss the Hamiltonian method of swarm design, which generates swarm design requirements as functions of locally measurable and manipulable quantities. 

Unfolding Statistical Inference Algorithms into Better Deep Networks

Robert Kosara Senior Research Scientist
Tableau
Robert Kosara Senior Research Scientist
Tableau
Friday, March 10, 2017

Data visualization is still a young field. It’s also a hugely complicated field that tries to combine the rigor of statistics and computer science with the complications of cognitive psychology and vision science – and that’s not even considering application fields. While we know some things about visualization, we don’t know others – and some of the things we thought we knew have also turned out to not be quite that clear-cut.

Better Machine Learning Through Data

Dr. Saleema Amershi Researcher, Machine Teaching Group
Microsoft Research
Dr. Saleema Amershi Researcher, Machine Teaching Group
Microsoft Research
Friday, March 3, 2017

Machine learning is the product of both an algorithm and data. While machine learning research tends to focus on algorithmic advances, taking the data as given, machine learning practice is quite the opposite. In this talk, I will present tools and techniques we have been developing in the Machine Teaching Group at Microsoft Research to support the model building process. I will discuss some of the open challenges and opportunities in improving the practice of machine learning.

Unfolding Statistical Inference Algorithms into Better Deep Networks

Scott Wisdom Ph.D. student
University of Washington
Scott Wisdom Ph.D. student
University of Washington
Thursday, March 2, 2017

The core idea of my research is "deep unfolding," which can be used to construct and explain deep network architectures with inference in statistical models. Deep unfolding yields principled initializations for training deep networks, provides insight into the effectiveness of deep networks, and assists with interpretation of what these networks learn.

Investigating Human-Machine and Human-Human Trust: Observations and Findings from a Human Factors Psychologist

Dr. Corey Fallon Independent Consultant and Visiting Professor
University of Cincinnati
Dr. Corey Fallon Independent Consultant and Visiting Professor
University of Cincinnati
Thursday, January 5, 2017

Dr. Fallon’s presentation will review several key topics related to the study of trust in human-machine and human-human teams. The discussion will be supported by Dr. Fallon’s own research and thinking, as well as the ideas and findings of other trusted researchers in the field. Several common misconceptions of trust will be exposed and discussed.

Sensemaking: From crime scenes to incident response. What is sense and how is it made?

Dr. Chris Baber Professor
University of Birmingham
Dr. Chris Baber Professor
University of Birmingham
Tuesday, December 6, 2016

This talk considers the ways in which our interactions with our environment, our data and each other create opportunities for sensemaking. The talk will present some of my team's work on crime scene examination and incident response. I will develop and compare two notions of sensemaking that we call semantic and pragmatic and look at ways in which these notions complement and challenge each other in the ways that people respond to complex, uncertain situations.

Unfetter: Discover Gaps in Your Security Posture

Matt Davis Information Assurance
Matt Davis Information Assurance
Thursday, September 29, 2016

Unfetter is a community-driven suite of open source tools to help cyber security professionals explore and analyze gaps in their security posture. Two members from the Unfetter development team will be joining us to present on topics related to AIM research into Insider Threat Detection in Cloud Computing environments.

Deep Learning for Unsupervised Anomaly Detection in Streaming Cybersecurity Data

Dr. Brian Hutchinson Assistant Professor, Department of Computer Science
Western Washington University
Dr. Brian Hutchinson Assistant Professor, Department of Computer Science
Western Washington University
Thursday, August 11, 2016

In this talk I will discuss our approach to insider threat detection that addresses these challenges. We use a novel deep learning architecture to model the day-to-day dynamics of "normal" user behavior so that we can automatically identify and flag anomalous behavior, under the assumption that threat behavior will be anomalous. Our system is explicitly designed for a real-time, multi-user, streaming environment.

Human-Machine-Interaction: The Best of Both Worlds

Eli Brown Assistant Professor, College of Computing and Digital Media
DePaul University
Eli Brown Assistant Professor, College of Computing and Digital Media
DePaul University
Tuesday, August 2, 2016

There are an increasing number of applications of machine learning in a wide variety of areas, and plenty of examples where a human is in-the-loop. If we believe that human and machine are better together than either alone, we must strive to demonstrate this principle on systems that solve real problems with collaboration between HCI and machine learning communities.

Gaussian Processes in the Study of Cognition

Joseph Houpt Assistant Professor
Wright State University
Joseph Houpt Assistant Professor
Wright State University
Tuesday, July 26, 2016

Many empirical studies of cognition focus on comparing estimates of single points (e.g., mean response times, choice probabilities, test scores).  This is due, at least in part, to constraints imposed by using traditional statistical analyses such as t-tests, ANOVAs and regression. Even in areas in which multidimensional dependent variables are used, analysis is usually confined to relatively small dimensional vectors. 

Enabling Discovery through Interactive Visual Data Analysis

Dr. Alexander Lex Assistant Professor of Computer Science at the Scientific Computing and Imaging Institute and the School of Computing
University of Utah
Dr. Alexander Lex Assistant Professor of Computer Science at the Scientific Computing and Imaging Institute and the School of Computing
University of Utah
Tuesday, July 19, 2016

Scientific discovery today is increasingly data driven and requires computational support. However, there is an important class of problems for which purely automatic approaches do not suffice since they require human reasoning and decision making, and can benefit from contextual knowledge humans possess.

Machine Learning Counterclockwise: Tips From the Trenches

Mr. Shawn Rutledge Principal Scientist
KFold Enterprises
Mr. Shawn Rutledge Principal Scientist
KFold Enterprises
Thursday, June 16, 2016

"Ship early, ship often." Continuous integration. Test-driven development. As software engineering has matured in industry, these and other patterns/ anti-patterns have helped avoid the pitfalls common to moving from code to working software systems. As machine learning grows in practice and pervasiveness in industry, similar patterns can help the move from models into working  machine learning pipelines and systems.

Utilizing Topological Visual Analytics for Persistent Human Performance Monitoring

Dr. Ryan Kramer Air Force Officer and Civilian
Air Force Research Laboratory
Dr. Ryan Kramer Air Force Officer and Civilian
Air Force Research Laboratory
Monday, May 23, 2016

The application of machine learning and deep learning approaches to multiple data types is providing increased insights into multivariate and multimodal data. Although inclusion of these approaches has dramatically enhanced the speed of data to decision processes, there are multiple drawbacks that include “black box” and “hidden layers” that obfuscate interrelationships between multivariate data and their conclusions.

Quantitative Models for User-Centered Visualization Systems

Dr. Lane Harrison Assistant Professor, Department of Computer Science
Worcester Polytechnic Institute
Dr. Lane Harrison Assistant Professor, Department of Computer Science
Worcester Polytechnic Institute
Tuesday, April 19, 2016

In a world that is becoming increasingly data-driven, people rely on visualizations to make high-impact and even life-critical decisions. Because of this, there is a growing need to ensure that the information displayed in visualizations is perceived accurately and precisely.  Although many off-the-shelf visualization tools exist, to date we have lacked a robust understanding of how much information the users of these tools can perceive and process.

A New Territory of Machine Translation

Dr. Kyunghyun Cho Assistant Professor of Computer Science and Data Science
New York University
Dr. Kyunghyun Cho Assistant Professor of Computer Science and Data Science
New York University
Wednesday, March 2, 2016

In this talk, instead of telling you how neural machine translation has been successful at the existing setting of machine translation, I describe new opportunities in machine translation that have become possible by introducing deep learning to machine translation. These opportunities include sub-word/character-level translation, multilingual translation, and larger-context modelling.

Security and Privacy for Robots and Brains

Dr. Tamara Bonaci
University of Washington
Dr. Tamara Bonaci
University of Washington
Tuesday, February 9, 2016

Tamara will present her recent work on security and privacy of biomedical cyber-physical systems. She will focus on two questions: (1) how do human idiosyncrasies open these systems for exploitations and attacks, and (2) how can we leverage human uniqueness to personalize these systems, but also to make them more security – and privacy-preserving and enhancing.

How to Take Pictures of Molecules with a Two-Mile Long Camera

Dr. TJ Lane
Dr. TJ Lane
Wednesday, December 2, 2015

The world’s first such source, The Linac Coherent Light Source (LCLS), is located at SLAC. I will discuss three projects the LCLS has enabled: generating a molecular movie of a chemical reaction, understanding protein motion in the crystalline state, and the possibility of taking pictures of single molecules or virus particles. Emphasis will be placed on current bottlenecks in these projects that could be alleviated by algorithmic or software advances.

Tensor Deep Stacking Networks

Dr. Brian Hutchinson
Dr. Brian Hutchinson
Wednesday, November 11, 2015

The Tensor Deep Stacking Network is a deep learning architecture that has demonstrated strong results on speech and vision tasks. The model consists of a set of stacked blocks, where the output predictions of each block are fed into the following block. Unlike deep neural networks, model performance steadily improves when trained with larger batch sizes.

The Challenges of Forecasting and Measuring a Complex Networked World

Dr. Bruno Ribeiro
Dr. Bruno Ribeiro
Friday, November 6, 2015

A new era of data analytics for online social networks promises tremendous high-impact societal, business, and healthcare applications. As more users join online social networks, the data available for analysis and forecast of human social and collective behavior grows at an incredible pace.

What can (large scale) machine learning do for you?

Dr. Abhinav Vishnu
Dr. Abhinav Vishnu
Tuesday, October 20, 2015

In this talk, we will present a set of steps which domain scientists may use for large scale data analysis. We will provide a brief introduction to these steps, and demonstrate with one or more use cases. We expect that many data producing domains will benefit significantly from this talk, with a potential for longer term collaboration.

Stream Data Mining and Applications: A Big Data Perspective

Dr. Latifur Khan
Dr. Latifur Khan
Tuesday, September 15, 2015

In this talk we demonstrate how to make fast and correct classification decisions under this constraint. Furthermore, we will present a semi supervised framework which exploits change detection on classifier confidence values to update the classifier intelligently with limited labeled training data. We present a number of stream classification applications such as website fingerprinting, real time monitoring, evolving insider threat detection, and textual stream classification.

Learning Intelligent Assistants for Understanding Behavior

Dr. Dragos Margineantu
Dr. Dragos Margineantu
Thursday, August 27, 2015

This talk will analyze how, by formulating adequately our research questions and by employing models that are capable of capturing domain interaction, learning can be made scalable via expert interaction. Next, we will discuss how our techniques can be employed for assisting experts in surveillance tasks such as intent recognition and detecting abnormal behavior. Finally, we will outline open research questions for usable expert-interactive learning.

The Evolving Landscape for Ontologies and Semantic Technologies

Dr. Deborah McGuinness
Dr. Deborah McGuinness
Tuesday, June 23, 2015

Open source ontologies exist in most domains, often developed and maintained by communities, and are in broad usage in a wide range of domains and applications. In this talk, trends in ontologies, their usage, and their supporting semantic environments will be introduced, as well as describing successes and challenges through the use of examples.

Modeling Workload Capacity

Dr. Leslie Blaha
Dr. Leslie Blaha
Thursday, June 18, 2015

Workload capacity: ability of our cognitive processing mechanisms to respond to the changes in task demands that influence the workload. These include changes to the number of subtasks within a task, items in a visual or memory search display, and manipulated features in a visual processing task. Blaha will discuss a human information processing modeling perspective on workload capacity, from which we measure mental work with hazard functions of response time data.

Acquisition and Analysis of Functional Brain Signals

Dr. Dianne Patterson
Dr. Dianne Patterson
Friday, April 24, 2015

Functional Magnetic Resonance Imaging (fMRI) has provided significant insight into the function of the human brain over the last two decades; as such, brain imaging with an emphasis on fMRI will be discussed. In the last five years computational power has improved to the point where we can move away from the traditional but limited general linear model analyses to more robust and sensitive techniques like independent component analysis, network analyses, and machine learning.

Towards Automated Feature Engineering for Transactional Data

Rob Jasper
Rob Jasper
Monday, April 6, 2015

This talk presents an automated method for constructing features from transactional data based on parameterized feature templates. Templates can represent a near infinite number of features compactly. Previous experiments have shown this approach to be highly competitive. A framework for generalizing this approach is proposed.

Big Graph Search and Analytics: A Journey of Usability and Scalability

Dr. Yinghui Wu
Dr. Yinghui Wu
Wednesday, April 1, 2015

Real-life graphs are heterogeneous and huge. These bring two challenges to emerging graph data applications: How to make big graphs usable and useful? And how to adapt the data processing scale to big graphs? In this talk, Dr. Wu will share his experience improving the usability and scalability for big graphs, focusing on graph querying.

Stream Processing at Facebook

Blake Matheny
Blake Matheny
Thursday, March 19, 2015

Facebook continuously processes an incredible amount of data in real-time, near-time, and batch. In this talk, through the use of case studies, we'll examine the systems that support this infrastructure. We will start by examining the systems responsible for data collection, processing, storage, and analysis. We will also, through the use of case studies, take a look at some of the interesting outcomes from recent analysis, mostly on the infrastructure (non-user) side of things.

Corvus: An Intelligent Workflow Tool for Computational Materials Science and Spectroscopy

Shauna Story
Shauna Story
Wednesday, March 18, 2015

Shauna Story will explain how during the past decade, her working group has increased its use of third-party codes to obtain properties that are unavailable in the code-base developed in-house. Until now, most development has relied on the creation of interfaces between pairs of codes, using a plethora of languages and coding strategies.

Data Cartography: Using Maps to Navigate Knowledge Networks

Dr. Jevin West
Dr. Jevin West
Thursday, January 15, 2015

In this presentation, Dr. West will talk about the methods for mapping knowledge networks and provide examples on how these maps can be used to identify innovative and influential ideas, people, and institutions.    

I-VEST: Intelligent Visual Email Search and Triage

Enrico Bertini
Enrico Bertini
Wednesday, October 29, 2014

Lawyers and investigators are often presented with large email datasets that contain emails that are not all relevant to any given investigative search. They must often manually comb through information contained within these large datasets in order to find the information they need, expending large amounts of time and money in the process. Our work offers an interactive visual analytic alternative to current methodology.

Quantitative Metabolomics Using NMR

Dr. David Chang
Dr. David Chang
Monday, September 15, 2014

The field of metabolomics started by studying small molecule data from nuclear magnetic resonance (NMR) and mass spectrometry (MS) experiments. This presentation will address specifically the use of NMR and the requirements for accurate quantitation, as well as address some published work on known variations in these data.

Joining Forces from Data Mining and Visual Analytics for Large-scale High-dimensional Data

Dr. Jaegul Choo
Dr. Jaegul Choo
Monday, June 16, 2014

Visual analytics, which leverages human exploration via interactive visualization in data analyses, has recently gained popularity. Data mining methods often play a crucial role in visual analytics by providing an important insight about data. In this talk, I will present both fundamental approaches and visual analytics systems that join forces from data mining and visual analytics.

Analyzing User Interactions for Data and User Modeling

Dr. Remco Chang
Dr. Remco Chang
Monday, June 16, 2014

User interactions with a visualization system reflect a great deal of the user's reasoning process and personality. In this talk, Remco will present techniques that he developed to analyze the user's interactions in order to (a) model the data in the form of metric learning that reflect a user's understanding of high-dimensional data, and (b) model the user and learn the user's individual differences and analysis behavior.

Human learning processes for sensorimotor control

Dr. Loes van Dam
Dr. Loes van Dam
Tuesday, May 27, 2014

Dr. Loes van Dam is a cognitive neuroscientist / psychophysicist with expertise in the areas of human multi-sensory perception and action. Her interests include determining how the human brain transforms the sensory information it receives into adequate perceptual interpretations and goal-oriented behavior.

| Pacific Northwest National Laboratory