Argus Prime is a series of experiments using the Argus Simulated Task Environment that are focused on determining how humans operate within a dynamic environment. The series of studies include dual-task and interruption conditions while an operator classifies the threat-level of incoming aircraft (see Argus ).

Few tasks are so new as to require the invention of strategies that have never been used by the task performer. Hence, in many situations, settling on a strategy or set of strategies for performing a task is not so much a matter of learning new strategies as it is learning which strategy, out of a set of already acquired strategies, is best adapted to the current environment. Blocks World is a simple task that has been used to study the tradeoff bnetween interaction-intensive and memory-intensive strategies.

Best path Length On a Semantic Self-Organizing Map (BLOSSOM) is an experimental, unsupervised text-clustering technique with applications in computational linguistics and data mining. This project involves the development and refinement of BLOSSOM as a Measure of Semantic Relatedness and the exploration of its benefits.

Humans routinely interact with simple information devices, such as watches, PDAs, cell phones, and both physical and electronic calendars. Although infrequently studied, the design and operation of these devices can have significant impacts on how we operate in the world. The Calendar Study is designed to investigate the effects of small costs on performance, as well as the stability of strategy selection during routine, everyday tasks.

Configural Memory with Network Reinforcement Learning (CMNRL, pronounced Sea Mineral) is a categorization-based cognitive architecture and an autonomous agent. It is an unsupervised incremental neural network with two main components. The first component, configural memory, is similar to the configural approaches of Gluck & Bower (1988) and Heydemann (1995). Configural approaches have been used to model a wide variety of psychological data (e.g. Pearce, 1994). The second component of CMNRL, Network Reinforcement Learning (NRL) extends traditional reinforcement learning (Sutton & Barto, 1998) by allowing for simultaneous updates of multiple state-action pairs. Just as configural memory, reinforcement learning has been affirmed as a psychologically and biologically plausible mechanism (e.g. Holroyd & Coles, 2002).

Over the last two decades attempts to quantify decision-making have established that, under a wide range of conditions, people trade-off effectiveness for efficiency in the strategies they adopt. However, as interesting, significant, and influential as this research has been, its scope is limited by three factors; the coarseness of how effort was measured, the confounding of the costs of steps in the decision-making algorithm with the costs of steps in a given task environment, and the static nature of the decision tasks studied. Across a series of experiments, we embed decision-making tasks into dynamic task environments and vary the cost required for various steps. Across studies, small changes in the cost of interactive behavior leads to changes in the strategy adopted for decision-making as well as to differences in how a step in the same strategy is implemented. Work is proceeding to construct a family of ACT-R models, simBorgs, that perform aspects of the DMAP tasks in the same way as humans.

Measures of Semantic Relatedness (MSRs) are computational means for calculating the association strength between terms. MSRs have been used to produce models of human web-browsing behavior (Pirolli, 2005), augmented search engine technology (Dumais, 2003), essay-grading algorithms for ETS (Landauer, Foltz, & Laham, 1998), and could be useful for any cognitive models or AI agents that have to deal with text.

http://cwl-projects.cogsci.rpi.edu/msr

Over the years, many researchers have found that humans often exhibit stable, suboptimal behavior. This project is designed to further examine the factors influencing strategy selection in a simple paradigm involving repeatedly choosing between two options.

The ObViS (pronounced like obvious) research project involves the development and validation of measures of visual similarity. We predict that when searching for a particular target object, the similarity of low-level visual features of any given object to the features of the target object will be a better predictor of visual attention than the saliency of the object.

Saccadic selectivity refers to the systematic selection of some visual locations rather than others due to one of three sources: stimulus-driven processes, soft constraints operating through acquired expected utilities, or deliberately adopted goal-driven strategies. In previous research, we manipulated the global configuration of a visual display to study its influence on the initial fixation in a search task. We also manipulated cognitive load. Across three experiments we found a systematic influence of global configuration on saccadic selectivity. In the second experiment we found that performing a secondary task increases the influence of our global configuration on saccadic selectivity. Experiment 3 pushed our paradigm to its limit to reveal intriguing data regarding the time course of the tradeoff between stimulus--driven processes and soft constraints.

‘Simon’ uses a simple memory game to study the role of visual, auditory and spatial sensory modalities in the cognitive processes of memory and attention.

StreamScout is an extension for the Firefox browser which applies measures of semantic relatedness to assess the relevancy of webpages to others in the user's clickstream.

Tardast is a new and intriguing paradigm to investigate aspects of human multitasking behavior, complex system management and supervisory control. We replicate and extend the original Tardast study (Shakeri, 2003, Shakeri & Funk, in press) that assesses operators' learning curve and explains performance gains in terms of increased sensitivity to task parameters and a superior ability of better operators to prioritze tasks.

See http://www.cogsci.rpi.edu/cogworks/tardast/ for additional information.

Apart from being a game that we all know and love, Tetris is also a relatively complex cognitive task that requires dynamic task processing, advanced perceptual capabilities, and involves both top-down and bottom-up strategies. Given the lack of cognitive models designed for such dynamic tasks, we explore the Tetris environment as a way to get at human perception, learning and memory, categorization, attention, and procedure/strategy selection.

more to come soon

Visual search takes place whenever we are looking for something. But when the location of a search target has been encoded on a previous occasion, memory processes can supplement or compete with eye movements during search. The goal of this project is to illuminate the interactions of visual attention and memory by assessing how humans adapt their search strategies to the cost structure of a task environment.

Space Fortress is an action video game that requires constant shifts of attention, memory retrievals, visual tracking, fine motor control, and dynamic decision making. We are creating a hybrid cognitive model to play the game, part of a larger effort in studying skill transfer.

Argus is a radar simulation that containins approaching airships. Each airship varies across seven attributes, such as speed, altitude, distance from ownship, etc. The primary task is to classify the threat of each airship. Argus was developed to support research in measuring and modeling cognitive workload. Argus can be used in single-subject and team modes.

CTK is a DTO sponsored project that seeks to support interface design for advanced visualization and interaction techniques. We achieve this using the VIA architecture as a testbed for software applications, performing detailed user analysis (including ProtoMatch eye-data analysis), visual and semantic saliency analyses (Visual Saliency Maps, Measures of Semantic Distance, Information Foraging), developing high-fidelity cognitive models for robust and exhaustive interface testing (simBorgs), and creating methods for assessing dynamic changes in cognitive workload (Cognitive Metrics Profiling).

The goal of MultiWorld is to further integrate cognitive models with the real world. Traditionally, psychologists studying task switching have studied simple laboratory tasks in isolation. This approach has yielded decades of data, but little relevance to the sorts of tasks humans perform routinely. How are we able to perform multiple, complex tasks simultaneously? How do we decide to switch between two tasks? What is the mechanism enabling task switching? MultiWorld seeks to answer these questions by providing a framework for studying multitasking in complex environments. By running tasks and cognitive models on seperate computers, MultiWorld ensures that the dividing line between the two is made explicit. Thus the model sees only what its human counterpart sees.

ProtoMatch is a software tool designed for exploratory data analyses on high-density behavioral data. It provides basic protocol analyses and a means of computing the similarity between two or more sequences of temporally ordered data. ProtoMatch is modularized software that integrates both eye gaze and cursor protocols into a unified stream of high-density, sequential, data.

The CogWorks Visualization-Interaction Architecture is a flexible software system designed to facilitate R&D for advanced, dynamic, highly visual interfaces such as those produced by University of Maryland (UMd) and by RPI’s RAIR Lab. Although our configuration uses one PC and one Mac, VIA is platform independent. As all communication between computers occurs over TCP/IP, the key component of VIA, the handler (see Figure) can be written for any platform (Unix, Windows XP, Mac OS 10, etc) and in any language. VIA release 0.1 was used in Sept 2005 with two visualization tools developed by UMd – TreePlus and GraphPlus. The current C# handler can interact with any program that uses the standard C# GUI library (WinForm) and that is written using standard object-oriented design techniques. In the near future we expect to develop handlers for Piccolo™, LispWorks™, and possibly, Java™.

CASS (the Cognitive Affective State System) aims at developing a model of the mutual effects of affect on cognition and of cognition on affect. In particular we work towards a deep and fundamental integration of emotion with low-level features of the human cognitive architecture.

Together, fNIR and computational cognitive modeling promise to have a synergistic relationship that should affect both basic and applied research. For example, recently fMRI has been used to associate different modules of the ACT-R architecture of cognition to different brain areas ([1]). Given this association, a trace of the model can be used to predict brain activity. In turn, brain activity is used to validate distinctions made by the cognitive theory that underlies the model. We propose to take this relationship between models and brain activity one step further. Our work focuses on the brain and model activity that reflects changes in cognitive workload.

TRACS is a 'Tool for Research on Adaptive Cognitive Strategies'. It is a simple card game developed by Kevin Burns, and can be played online at http://tracsgame.com. (Additional background information is available at http://mentalmodels.mitre.org.) As in the original TRACS studies, we are using the game to assess people's abilities to keep track of the changing odds throughout the course of a game. By equipping our application with eye-tracking and various optional enhancements we are able to run subtle variations and gather a full range of experimental data. Consistent with our "to understand it, build it" approach, we use computational cognitive modeling techniques to test our theories as to the cognitive mechanisms involved.
Our findings have direct implications for research in memory, embodied cognition, and interactive behavior. In addition, our application will be used in combination with the CogWorks MultiWorld for further research on the effects of cognitive workload.

The error-prone task of programming a VCR is representative of a growing number of end-user programmable devices. To help understand this task a display-based model of VCR programming was developed and implemented as a computational cognitive model. The model accurately predicted the vast majority of correct and error recovery keypresses collected from human subjects (Gray, 1995). Further VCR studies were done to compare strategies given additional costs of unnatural interface design.
Currently VCR programming data are being analyzed for memory errors and memory strategies. Computation cognitive models are constantly being developed to match and explain empirical data.