Enriching representation learning using 53 million patient notes through human phenotype ontology embedding.

The Human Phenotype Ontology (HPO) is a dictionary of >15,000 clinical phenotypic terms with defined semantic relationships, developed to standardize phenotypic analysis. Over the last decade, the HPO has been used to accelerate the implementation of precision medicine into clinical practice. In addition, recent research in representation learning, specifically in graph embedding, has led to notable progress in automated prediction via learned features. Here, we present a novel approach to phenotype representation by incorporating phenotypic frequencies based on 53 million full-text health care notes from >1.5 million individuals. We demonstrate the efficacy of our proposed phenotype embedding technique by comparing our work to existing phenotypic similarity-measuring methods. Using phenotype frequencies in our embedding technique, we are able to identify phenotypic similarities that surpass current computational models. Furthermore, our embedding technique exhibits a high degree of agreement with domain experts' judgment. By transforming complex and multidimensional phenotypes from the HPO format into vectors, our proposed method enables efficient representation of these phenotypes for downstream tasks that require deep phenotyping. This is demonstrated in a patient similarity analysis and can further be applied to disease trajectory and risk prediction.

Interactive Grounding and Inference in Learning by Instruction.

Learning by instruction is one of the most common forms of learning, and a number of research efforts have modeled the cognitive process of instruction following, with many successes. However, most computational models remain brittle with respect to the given instructions, and they lack the ability to adapt dynamically to variants of the instructions. This paper aims to illustrate modeling constructs designed to make instruction following more robust, including (1) more flexible grounding of language to execution, (2) processing of instructions that allows for inference of implicit instruction knowledge, and (3) dynamic, interactive clarification of instructions during both the learning and execution stages. Examples in the context of a paired-associates task and a visual-search task are discussed.

Integration and reuse in cognitive skill acquisition.

Previous accounts of cognitive skill acquisition have demonstrated how procedural knowledge can be obtained and transformed over time into skilled task performance. This article focuses on a complementary aspect of skill acquisition, namely the integration and reuse of previously known component skills. The article posits that, in addition to mechanisms that proceduralize knowledge into more efficient forms, skill acquisition requires tight integration of newly acquired knowledge and previously learned knowledge. Skill acquisition also benefits from reuse of existing knowledge across disparate task domains, relying on indexicals to reference and share necessary information across knowledge components. To demonstrate these ideas, the article proposes a computational model of skill acquisition from instructions focused on integration and reuse, and applies this model to account for behavior across seven task domains.

Toward a unified view of cognitive control.

Allen Newell (1973) once observed that psychology researchers were playing "twenty questions with nature," carving up human cognition into hundreds of individual phenomena but shying away from the difficult task of integrating these phenomena with unifying theories. We argue that research on cognitive control has followed a similar path, and that the best approach toward unifying theories of cognitive control is that proposed by Newell, namely developing theories in computational cognitive architectures. Threaded cognition, a recent theory developed within the ACT-R cognitive architecture, offers promise as a unifying theory of cognitive control that addresses multitasking phenomena for both laboratory and applied task domains.

Effects of memory rehearsal on driver performance: experiment and theoretical account.

OBJECTIVE: We report an experiment and a theoretical analysis concerning the effects of an exclusively cognitive task, specifically a memory rehearsal task, on driver performance. BACKGROUND: Although recent work on driver distraction has elucidated the sometimes significant effects of cognitive processing on driver performance, these studies have typically mixed cognitive with perceptual and motor processing, making it difficult to isolate the effects of cognitive processing alone. METHOD: We asked participants to drive in a driving simulator during only the rehearsal stage of a serial-recall memory task while we measured their ability to maintain a central lane position and respond to the illumination of a lead vehicle's brake lights. RESULTS: Memory rehearsal significantly affected drivers' steering performance as measured by lateral deviation from lane center, and it also significantly affected drivers' response time to the braking stimulus for the higher load memory task. CONCLUSION: These results lend support to a theoretical account of cognitive distraction provided by threaded cognition theory in terms of a cognitive bottleneck in procedural processing, and they also suggest that consideration of task urgency may be important in accounting for performance trade-offs among concurrent tasks. APPLICATION: The experiment augments the current understanding of cognitive driver distraction and suggests that even exclusively cognitive secondary tasks may sometimes affect driver performance.

Threaded cognition: an integrated theory of concurrent multitasking.

The authors propose the idea of threaded cognition, an integrated theory of concurrent multitasking--that is, performing 2 or more tasks at once. Threaded cognition posits that streams of thought can be represented as threads of processing coordinated by a serial procedural resource and executed across other available resources (e.g., perceptual and motor resources). The theory specifies a parsimonious mechanism that allows for concurrent execution, resource acquisition, and resolution of resource conflicts, without the need for specialized executive processes. By instantiating this mechanism as a computational model, threaded cognition provides explicit predictions of how multitasking behavior can result in interference, or lack thereof, for a given set of tasks. The authors illustrate the theory in model simulations of several representative domains ranging from simple laboratory tasks such as dual-choice tasks to complex real-world domains such as driving and driver distraction.

Lane-change detection using a computational driver model.

OBJECTIVE: This paper introduces a robust, real-time system for detecting driver lane changes. BACKGROUND: As intelligent transportation systems evolve to assist drivers in their intended behaviors, the systems have demonstrated a need for methods of inferring driver intentions and detecting intended maneuvers. METHOD: Using a "model tracing" methodology, our system simulates a set of possible driver intentions and their resulting behaviors using a simplification of a previously validated computational model of driver behavior. The system compares the model's simulated behavior with a driver's actual observed behavior and thus continually infers the driver's unobservable intentions from her or his observable actions. RESULTS: For data collected in a driving simulator, the system detects 82% of lane changes within 0.5 s of maneuver onset (assuming a 5% false alarm rate), 93% within 1 s, and 95% before the vehicle moves one fourth of the lane width laterally. For data collected from an instrumented vehicle, the system detects 61% within 0.5 s, 77% within 1 s, and 84% before the vehicle moves one-fourth of the lane width laterally. CONCLUSION: The model-tracing system is the first system to demonstrate high sample-by-sample accuracy at low false alarm rates as well as high accuracy over the course of a lane change with respect to time and lateral movement. APPLICATION: By providing robust real-time detection of driver lane changes, the system shows good promise for incorporation into the next generation of intelligent transportation systems.

Modeling driver behavior in a cognitive architecture.

OBJECTIVE: This paper explores the development of a rigorous computational model of driver behavior in a cognitive architecture--a computational framework with underlying psychological theories that incorporate basic properties and limitations of the human system. BACKGROUND: Computational modeling has emerged as a powerful tool for studying the complex task of driving, allowing researchers to simulate driver behavior and explore the parameters and constraints of this behavior. METHOD: An integrated driver model developed in the ACT-R (Adaptive Control of Thought-Rational) cognitive architecture is described that focuses on the component processes of control, monitoring, and decision making in a multilane highway environment. RESULTS: This model accounts for the steering profiles, lateral position profiles, and gaze distributions of human drivers during lane keeping, curve negotiation, and lane changing. CONCLUSION: The model demonstrates how cognitive architectures facilitate understanding of driver behavior in the context of general human abilities and constraints and how the driving domain benefits cognitive architectures by pushing model development toward more complex, realistic tasks. APPLICATION: The model can also serve as a core computational engine for practical applications that predict and recognize driver behavior and distraction.

A multitasking general executive for compound continuous tasks.

As cognitive architectures move to account for increasingly complex real-world tasks, one of the most pressing challenges involves understanding and modeling human multitasking. Although a number of existing models now perform multitasking in real-world scenarios, these models typically employ customized executives that schedule tasks for the particular domain but do not generalize easily to other domains. This article outlines a general executive for the Adaptive Control of Thought-Rational (ACT-R) cognitive architecture that, given independent models of individual tasks, schedules and interleaves the models' behavior into integrated multitasking behavior. To demonstrate the power of the proposed approach, the article describes an application to the domain of driving, showing how the general executive can interleave component subtasks of the driving task (namely, control and monitoring) and interleave driving with in-vehicle secondary tasks (radio tuning and phone dialing).

A two-point visual control model of steering.

When steering down a winding road, drivers have been shown to use both near and far regions of the road for guidance during steering. We propose a model of steering that explicitly embodies this idea, using both a 'near point' to maintain a central lane position and a 'far point' to account for the upcoming roadway. Unlike control models that integrate near and far information to compute curvature or more complex features, our model relies solely on one perceptually plausible feature of the near and far points, namely the visual direction to each point. The resulting parsimonious model can be run in simulation within a realistic highway environment to facilitate direct comparison between model and human behavior. Using such simulations, we demonstrate that the proposed two-point model is able to account for four interesting aspects of steering behavior: curve negotiation with occluded visual regions, corrective steering after a lateral drift, lane changing, and individual differences.