Opportunities

With apologies and respect to our valued colleagues of other nationalities, only U.S. citizens and permanent legal residents of the United States are eligible for these positions.

We have a variety of research positions available for talented cognitive, computational, and computer scientists interested in working with the U.S. Air Force Research Laboratory's Performance and Learning Models (PALM) Team on basic and applied cognitive science research. Full-time, paid positions range from undergraduate and graduate-level internships and research assistantships, to post-doctoral research appointments, to visiting faculty appointments. Salaries are commensurate with experience.

The PALM research portfolio continues to expand and evolve. We use a combination of empirical human-subjects studies and formal, rigorous, computational and mathematical modeling and simulation methods to understand, replicate, and predict human performance and learning, and to create new cognitive science-based technology options. Currently there are research efforts underway in all of the following areas (with associated PIs):

Basic research

  • Efficient Constraint-Based Search Mechanisms in a Cognitive Domain Ontology (Douglass)
  • Representations and Processes of Spatial Visualization (Gunzelmann)
  • Modeling the Relationships between Fatigue and Cognitive Processes (Gunzelmann)
  • Strategic Adaptation in Dynamic Non-stationary Environments (Myers)

Applied Research

  • Generating Depictive and Diagrammatic Representations of Meaning from Linguistic Input (Ball)
  • Heuristics for Robustness and Trust in Integrated Human-Machine Decision Systems (Gluck)
  • Model Exploration and Optimization Using Distributed and High Performance Computing (Harris)
  • Mathematical Modeling for Performance Prediction: Mechanism Development to Account for Effects of Cognitive Moderation (Jastrzembski)

Brief elaborations of each area can be found below.

Brief Overview

Anyone interested in working with us on one or more of the research efforts listed above is encouraged to contact the PI for that particular research area as soon as possible.

Generating Depictive and Diagrammatic Representations of Meaning from Linguistic Input

Ball, Jerry

Phone: (937) 938-4065 Email: jerry.ballwpafb.af.mil

Current representations of meaning in Natural Language Processing (NLP) systems typically rely on the use of uppercase words like PILOT to represent the concept of a pilot or word senses like pilot_n_1 to represent the first noun sense of pilot (e.g. aircraft pilot, not movie pilot or the action of piloting). These representations are really just alternative language-based forms of representation which fail to adequately capture meaning (Kilgariff, 1997). Until we are able to ground linguistic expressions in representations of perceptually based experience, we will not be able to adequately represent meaning (Barsalou, 1999). As an initial step in this direction, this research will lead to the computational generation of depictive (Lathrop & Laird, 2009) and diagrammatic (Chandrasekaran, 2006) representations of meaning from linguistic input, starting with linguistic expressions that describe common objects and their spatial relationships, in an aviation scenario. There are two main thrusts for this research: 1) determine how to compose depictive representations from linguistic expressions like "asphalt runway" or "concrete runway" without having to pre-store all possible images; and 2) determine how to diagrammatically represent spatial relationships between the depictive representation of objects.

A follow on goal is to integrate the depictive and diagrammatic representations into the situation model component of the synthetic teammate (Rodgers et al., 2011) to provide grounding for the current propositional representations of objects and situations, and to support reasoning over the depictive and diagrammatic representations.

References:

  • Barsalou, L. (1999). Perceptual symbol systems. Behavioral and Brain Sciences, 22, 577-609.
  • Chandrasekaran, B. (2006). Multimodal Cognitive Architecture: Making Perception More Central to Intelligent Behavior. AAAI National Conference on Artificial Intelligence, Boston, MA.
  • Kilgariff, A. (1997). "I don't believe in word senses". Computers and Humanities 31 (2), pp. 91-113.
  • Lathrop, S. & Laird, J., (2009). Extending Cognitive Architectures with Mental Imagery. Proceedings of the Second Conference on Artificial General Intelligence. Arlington, VA.
  • Rodgers, S., Myers, C., Ball, J. & Freiman, M. (2011). The Situation Model in the Synthetic Teammate Project. Proceedings of the 20th Annual Conference on Behavior Representation in Modeling and Simulation.

Efficient Constraint-Based Search Mechanisms in a Cognitive Domain Ontology

Douglass, Scott

Phone: (937) 938-4057 Email: scott.douglasswpafb.af.mil

AFRL/RHAC cognitive scientists are researching ways to increase the autonomy of cognitive models and agents. One approach to increasing autonomy involves specifying agents with formal representations of themselves, their knowledge, and the affordances of the situations in which they are acting. Researchers leading this approach are developing these representations or Cognitive Domain Ontologies (CDO) using System Entity Structure (SES) theory. SESs are founded on set theory. CDOs are used by autonomous agents to generate effective actions according to the contingencies and affordances presented by the environments they are situated in. These contingencies and affordances are made available as 'constraints' in a CDO. CDO also contains an agent's behavior repertoirre that gets soft assembled per these dynamic constraints. A CDO is the knowledge-base of the agent and contains representations of elements such as the environment, resources, goals, behaviors, etc. The elements of a CDO are linked together by these constaints. RHAC researchers are looking for efficient constraint-based search mechanisms to limit the combinatorics of CDO search. The search algorithms will be grounded in AI-based methodologies & Set theory and must be executable on parallel/distributed high performance systems. A key features of the proposed algorithms must be scalability. The algorithms must be able to complete the search process within 0.3-1.0 sec wall-clock time. Performance analayis of algorithms will therefore be a critical aspect of the research. The successful execution of the algorithm will result in a set of cognitive behaviors within the CDO which will prescribe effective action in the situated environment.

References:

  • Zeigler, B., & Hammonds, P. (2007). Modeling & Simulation-Based Data Engineering: Introducing Pragmatics into Ontologies for Net-centric Information Exchange. Academic Press.

Heuristics for Robustness and Trust in Integrated Human-Machine Decision Systems

Gluck, Kevin

Phone: (937) 938-3552 Email: kevin.gluckwpafb.af.mil

The idea of integrated human-machine decision systems is rapidly progressing from a vision of a possible future to a picture of our present reality. We need a better understanding of the basic science of mixed human - machine decision making, and we need to make use of this science to develop increasingly robust, automated knowledge-extraction tools and intelligent, trusted machine-based decision aids that improve and adaptively adjust inference, prediction, and decision processes. We are interested in new metrics, models and methods for objective, rigorous assessment of robustness and trust, as well as mathematical and computational models of heuristic-based decision processes that are demonstrably robust and trusted in dynamic, uncertain, non-stationary environments.

References:

  • Gluck, K. A., & Gunzelmann, G. (in press). Computational process modeling and cognitive stressors: Background and prospects for application in cognitive engineering. In J. D. Lee & A. Kirlik (Eds.) The Oxford Handbook of Cognitive Engineering.
  • Gluck, K. A., McNamara, J. M., Brighton, H., Dayan, P., Kareev, Y., Krause, J., Kurzban, R., Selten, R., Stevens, J. R., Voelkl, B., & Wimsatt, W. C. (2012). Robustness in a variable environment. In Stevens, J. R. & P. Hammerstein (Eds.) Evolution and the Mechanisms of Decision Making. Strüngmann Forum Report, vol. 11, J. Lupp, series ed. Cambridge, MA: MIT Press.
  • Gluck, K. A. (2010). Cognitive architectures for human factors in aviation. In E. Salas & D. Maurino (Eds.) Human Factors in Aviation, 2nd Edition (pp.375-400). New York, NY: Elsevier.
  • Gluck, K. A., Stanley, C. T., Moore, L. R., Reitter, D., & Halbrugge, M. (2010). Exploration for understanding in cognitive modeling. Journal of Artificial General Intelligence, 2(2), 88-107.

Representations and Processes of Spatial Visualization

Gunzelmann, Glenn

Phone: (937) 938-3554 Email: glenn.gunzelmannwpafb.af.mil

Cognitive architectures have been developed as general theories of cognition, based primarily on the results of laboratory studies. However, there are many components of cognition where cognitive architectures lack explanatory mechanisms. The Air Force Research Laboratory's Cognitive Models and Agents Branch is conducting research to expand the breadth of architectural accounts of human cognition by developing cognitively valid mechanisms for human spatial competence (e.g., Gunzelmann & Lyon, 2011). This research involves a combination of empirical investigation of human performance and computational modeling to account for the capacities and limitations of human spatial processing. Specific research efforts are focused on (1) spatial visualization and (2) encoding allocentric spatial information from visual stimuli.

References:

  • Gunzelmann, G., & Lyon, D. R. (2011). Representations and processes of human spatial competence. Topics in Cognitive Science, 3(4) 741-759.

Modeling the Relationships between Fatigue and Cognitive Processes

Gunzelmann, Glenn

Phone: (937) 938-3554 Email: glenn.gunzelmannwpafb.af.mil

Fatigue has long been recognized as a significant moderator of cognitive performance that has led to devastating consequences in real-world environments. Despite more than a century of research, however, a quantitative account of fatigue that provides the capacity for making performance predictions in real-world contexts has not emerged. In this research, we are brining together mathematical models of fatigue dynamics with unified theories of human cognition to account for the impact of fatigue on components of cognitive processing (e.g., Gunzelmann, Gross, Gluck, & Dinges, 2009; Gunzelmann, Moore, Salvucci, & Gluck, 2011). The research is focused on developing computational mechanisms to account for the deleterious effects of fatigue. We also profit from collaborations with international leaders in sleep research to collect data on relevant tasks in carefully controlled studies of sleep deprivation and restriction. The research is leading to an integrated theory of how time awake, circadian rhythms, and time on task impact the effectiveness and efficiency of human cognitive processing.

References:

  • Gunzelmann, G., Gross, J. B., Gluck, K. A., & Dinges, D. F. (2009). Sleep deprivation and sustained attention performance: Integrating mathematical and cognitive modeling. Cognitive Science, 33(5), 880-910.
  • Gunzelmann, G., Moore, L. R., Salvucci, D. D., & Gluck, K. A. (2011). Sleep loss and driver performance: Quantitative predictions with zero free parameters. Cognitive Systems Research, 12(2), 154-163.

Model Exploration and Optimization Using Distributed and High Performance Computing

Harris, Jack

Phone: (937) 938-4044 Email: jack.harriswpafb.af.mil

Computational complexity grows quickly with increases in the granularity of models, the fidelity of the models' operating environment, and the time scales across which these models are used in simulations. We must find ways to deal with the computational demands of large-scale basic and applied cognitive modeling. One approach is to acquire more computational horsepower, such as through high performance computing (HPC) clusters, volunteer computing, or cloud computing. Another approach is to reduce the size of the required computational space through predictive analytics and parallelized exploration and optimization algorithms. Our view is that it is only through the combined use of these approaches that we can meet our far-term scientific and technological objectives, both as a research team and as a broader research community.

Keywords: high performance computing (HPC), intelligent search algorithms (ISAs), computational mathematics

Mathematical Modeling for Performance Prediction: Mechanism Development to Account for Effects of Cognitive Moderation

Jastrzembski, Tiffany

Phone: (937) 938-4055 Email: tiffany.jastrzembskiwpafb.af.mil

Researchers at the 711th HPW/RHAC have developed, matured, and made more robust a mathematical model for performance prediction, known as the Predictive Performance Equation (PPE) (see Jastrzembski, Gluck, & Gunzelmann, 2006; Jastrzembski, Gluck, & Rodgers, 2009). This model has been carefully validated across a variety of domains and contexts - scaling from laboratory experimental data available in the psychological literature to increasingly complex and militarily relevant team and pilot data measured in the Air Force Research Laboratory's Distributed Missions Operations testbed (Schreiber, Stock, & Bennett, 2006). The predictive model functions by capturing learning signatures and mathematical regularities from the human memory system through calibration of learning and decay parameters using historical performance data, and extrapolates those unique learning signatures to make predictions of performance at specific later dates in time. The model critically extends that previous research by additionally accounting for the effects of temporal distribution of training on learning - a well-documented phenomenon known as the spacing effect - which reveals that given two training regimens of equal length and equal amounts of training opportunities, learning is more stable when practice events are spaced further apart in time. This research seeks to extend the model even further, by incorporating mechanisms that explicitly attenuate performance through effects of cognitive moderation (i.e., enhancement of performance from brain stimulation or caffeine; decrement of performance from fatigue or excess workload), so that a more complete picture may be gleaned regarding the range of likely performance effectiveness under known conditions.

References:

  • Anderson, J. R., & Schunn, C. D. (2000). Implications of the ACT-R learning theory: No magic bullets. In R. Glaser (Ed.), Advances in instructional psychology: Educational design and cognitive science, Vol. 5. Mahwah, NJ: Erlbaum.
  • Jastrzembski, T. S., Gluck, K. A., & Gunzelmann, G. (2006). Knowledge tracing and prediction of future trainee performance. I/ITSEC annual meetings, Orlando.
  • Jastrzembski, T. S., Gluck, K. A., & Rodgers, S. (2009). Improving military readiness: A state-of-the-art cognitive tool to predict performance optimize training effectiveness. I/ITSEC annual meetings, Orlando.
  • Schreiber, B. T., Stock, W. A., & Bennett, W. (2006). Distributed mission operations within-simulator training effectiveness baseline study: Metric development and objectively quantifying the degree of learning. AFRL-HE-AZ-TR-2006-0015-Vol II. Available online at: www.dtic.mil.

Strategic Adaptation in Dynamic Non-stationary Environments

Myers, Christopher

Phone: (937) 938-4044 Email: christopher.myers2wpafb.af.mil

Dr. Chris Myers is conducting research that uses machine learning and control theory techniques to model human performance and decision making within dynamic, non-stationary environments. The work involves both human experiments and computational modeling, with the aim of understanding how humans adapt to changes in complex, dynamic, and collaborative task environments. We are looking for a postdoctoral research associate interested in developing optimal models of visual search, multitasking, and dyadic collaboration. The ideal candidate will have a background in machine learning, cognitive science/cognitive psychology/mathematical psychology, and/or control theory, and will have programming experience with R and/or Python. Candidates must be a U.S. citizen or permanent resident.

Computational cognitive models are frequently developed to account for data gathered from experiments designed to isolate distinct cognitive functions (e.g., memory) from other functions (e.g., visual search, reasoning, decision-making, etc.). The promise of this approach is that veridical models of distinct functions will eventually be integrated to produce more complex processes. While this approach has proven beneficial to isolating, studying, and understanding distinct cognitive processes, the resulting models are typically brittle, engineered, short-lived and tailored to specific experimental psychology paradigms. These limitations are barriers to the development of models which require strategic adaptation to dynamic, non-stationary, increasingly complex environments which persist over long periods of time.