Search NASA⌕ Search

SEARCH · Search NASA

Results for “ATTENTION”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 73 records · Page 4
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Attention-Augmented Parametric Kernel Graph Neural Network (APKGNN) for Node Classification

We present a new graph neural network, the Attention-based Parametric-Kernel augmented Graph Neural Network (APKGNN), developed for node classification tasks. Despite extensive work on modeling multi-faceted relationships between connected nodes of a graph, the effect of attention on edge features mapped to relationships has not yet been analyzed through learning representation. This study derives such an attention vector by first calculating node features corresponding to endpoints of an edge and then aggregating these with extracted local intrinsic patches of a given graph to generate augmented local patch vectors. This process uses a parametric kernel based on Gaussian mixture models (GMMs) to embed local neighborhoods of the graph in local patches. The patch vectors then convolve with the above node features to produce an updated node representation. We show that this new learning representation (APKGNN) achieves higher node classification accuracy on tasks - both standard benchmarks (Cora, PubMed, Citeseer) and new experimental short text corpora where nodes correspond to text documents and words. This implementation of the GNN convolution layer outperforms state-of-the-art (SOTA) algorithms, achieving higher training, validation, and test accuracy by a significant margin on three standard benchmark data sets under both SOTA experimental settings and those for new testbeds.

Bose, Avishek↗
DOE OSTI2025

Attention to quantum complexity

The imminent era of error-corrected quantum computing demands robust methods to characterize quantum state complexity from limited, noisy measurements. We introduce the Quantum Attention Network (QuAN), a classical artificial intelligence (AI) framework leveraging attention mechanisms tailored for learning quantum complexity. Inspired by large language models, QuAN treats measurement snapshots as tokens while respecting permutation invariance. Combined with our parameter-efficient miniset self-attention block, this enables QuAN to access high-order moments of bit-string distributions and preferentially attend to less noisy snapshots. We test QuAN across three quantum simulation settings: driven hard-core Bose-Hubbard model, random quantum circuits, and toric code under coherent and incoherent noise. QuAN directly learns entanglement and state complexity growth from experimental computational basis measurements, including complexity growth in random circuits from noisy data. In regimes inaccessible to existing theory, QuAN unveils the complete phase diagram for noisy toric code data as a function of both noise types, highlighting AI’s transformative potential for assisting quantum hardware.

Kim, Hyejin [Cornell Univ., Ithaca, NY (United Sta↗
DOE OSTI2022

A Length Adaptive Algorithm-Hardware Co-design of Transformer on FPGA Through Sparse Attention and Dynamic Pipelining

Transformers are considered one of the most important deep learning models since 2018, in part because it establishes state-of-the-art (SOTA) records and could potentially replace existing Deep Neural Networks (DNNs). Despite the remarkable triumphs, the prolonged turnaround time of Transformer models is a widely recognized roadblock. The variety of sequence lengths imposes additional computing overhead where inputs need to be zero-padded to the maximum sentence length in the batch to accommodate the parallel computing platforms. This paper targets the field-programmable gate array (FPGA) and proposes a coherent sequence length adaptive algorithm–hardware co-design for Transformer acceleration. Particularly, we develop a hardware-friendly sparse attention operator and a length-aware hardware resource scheduling algorithm. The proposed sparse attention operator brings the complexity of attention-based models down to linear complexity and alleviates the off-chip memory traffic. The proposed length-aware resource hardware scheduling algorithm dynamically allocates the hardware resources to fill up the pipeline slots and eliminates bubbles for NLP tasks. Experiments show that our design has very small accuracy loss and has 80.2 × and 2.6 × speedup compared to CPU and GPU implementation, and 4 × higher energy efficiency than state-of-the-art GPU accelerator optimized via CUBLAS GEMM.

Peng, Hongwu↗
DOE OSTI2022

Attention-based Aspect Reasoning for Knowledge Base Question Answering on Clinical Notes

Question Answering (QA) in clinical notes has gained a lot of attention in the past few years. Existing machine reading comprehension approaches in clinical domain can only handle questions about a single block of clinical texts and fail to retrieve information about different patients and clinical notes. To handle more complex questions, we aim at creating knowledge base from clinical notes to link different patients and clinical notes, and performing knowledge base question answering (KBQA). Based on the expert annotations in n2c2, we first created the ClinicalKBQA dataset that includes 8,952 QA pairs and covers questions about seven medical topics through 322 question templates. Then, we proposed an attention-based aspect reasoning (AAR) method for KBQA and investigated the impact of different aspects of answers (e.g., entity, type, path, and context) for prediction. The AAR method achieves better performance due to the well-designed encoder and attention mechanism. In the experiments, we find that both aspects, type and path, enable the model to identify answers satisfying the general conditions and produce lower precision and higher recall. On the other hand, the aspects, entity and context, limit the answers by node-specific information and lead to higher precision and lower recall.

Wang, Ping↗
DOE OSTI2022

DISTEMA: distance map-based estimation of single protein model accuracy with attentive 2D convolutional neural network

Abstract Background Estimation of the accuracy (quality) of protein structural models is important for both prediction and use of protein structural models. Deep learning methods have been used to integrate protein structure features to predict the quality of protein models. Inter-residue distances are key information for predicting protein’s tertiary structures and therefore have good potentials to predict the quality of protein structural models. However, few methods have been developed to fully take advantage of predicted inter-residue distance maps to estimate the accuracy of a single protein structural model. Result We developed an attentive 2D convolutional neural network (CNN) with channel-wise attention to take only a raw difference map between the inter-residue distance map calculated from a single protein model and the distance map predicted from the protein sequence as input to predict the quality of the model. The network comprises multiple convolutional layers, batch normalization layers, dense layers, and Squeeze-and-Excitation blocks with attention to automatically extract features relevant to protein model quality from the raw input without using any expert-curated features. We evaluated DISTEMA’s capability of selecting the best models for CASP13 targets in terms of ranking loss of GDT-TS score. The ranking loss of DISTEMA is 0.079, lower than several state-of-the-art single-model quality assessment methods. Conclusion This work demonstrates that using raw inter-residue distance information with deep learning can predict the quality of protein structural models reasonably well. DISTEMA is freely at https://github.com/jianlin-cheng/DISTEMA

59 BASIC BIOLOGICAL SCIENCES↗
NASA NTRS1977

Visual evoked potentials and selective attention to points in space

Visual evoked potentials (VEPs) were recorded to sequences of flashes delivered to the right and left visual fields while subjects responded promptly to designated stimuli in one field at a time (focused attention), in both fields at once (divided attention), or to neither field (passive). Three stimulus schedules were used: the first was a replication of a previous study (Eason, Harter, and White, 1969) where left- and right-field flashes were delivered quasi-independently, while in the other two the flashes were delivered to the two fields in random order (Bernoulli sequence). VEPs to attended-field stimuli were enhanced at both occipital (O2) and central (Cz) recording sites under all stimulus sequences, but different components were affected at the two scalp sites. It was suggested that the VEP at O2 may reflect modality-specific processing events, while the response at Cz, like its auditory homologue, may index more general aspects of selective attention.

Van Voorhis, S.↗
NASA NTRS1977

Required attention for synthesized speech perception for three levels of linguistic redundancy

The study evaluates the attention required for synthesized speech perception with reference to three levels of linguistic redundancy. Twelve commercial airline pilots were individually tested for 16 cockpit warning messages eight of which consisted of two monosyllabic key words and eight of which consisted of two polysyllabic key words. Three levels of linguistic redundancy were identified: monosyllabic words, polysyllabic words, and sentences. The experiment contained a message familiarization phase and a message recognition phase. It was found that: (1) when the messages are part of a previously learned and recently heard set, and the subject is familiar with the phrasing, the attention needed to recognize the message is not a function of the level of linguistic redundancy, and (2) there is a quantitative and qualitative difference between recognition and comprehension processes; only in the case of active comprehension does additional redundancy reduce attention requirements.

Simpson, C. A.↗
NASA NTRS1977

Error rate information in attention allocation pilot models

The Northrop urgency decision pilot model was used in a command tracking task to compare the optimized performance of multiaxis attention allocation pilot models whose urgency functions were (1) based on tracking error alone, and (2) based on both tracking error and error rate. A matrix of system dynamics and command inputs was employed, to create both symmetric and asymmetric two axis compensatory tracking tasks. All tasks were single loop on each axis. Analysis showed that a model that allocates control attention through nonlinear urgency functions using only error information could not achieve performance of the full model whose attention shifting algorithm included both error and error rate terms. Subsequent to this analysis, tracking performance predictions for the full model were verified by piloted flight simulation. Complete model and simulation data are presented.

Faulkner, W. H.↗
NASA NTRS1977

Prediction of pilot reserve attention capacity during air-to-air target tracking

Reserve attention capacity of a pilot was calculated using a pilot model that allocates exclusive model attention according to the ranking of task urgency functions whose variables are tracking error and error rate. The modeled task consisted of tracking a maneuvering target aircraft both vertically and horizontally, and when possible, performing a diverting side task which was simulated by the precise positioning of an electrical stylus and modeled as a task of constant urgency in the attention allocation algorithm. The urgency of the single loop vertical task is simply the magnitude of the vertical tracking error, while the multiloop horizontal task requires a nonlinear urgency measure of error and error rate terms. Comparison of model results with flight simulation data verified the computed model statistics of tracking error of both axes, lateral and longitudinal stick amplitude and rate, and side task episodes. Full data for the simulation tracking statistics as well as the explicit equations and structure of the urgency function multiaxis pilot model are presented.

Onstott, E. D.↗
NASA NTRS1986

The effects of stress on attentional resources

A new perspective is presented from which to view the action of stress on human behavior. At a behavioral level, the action of stress is related to notions of human attention and an indication of an isomorphic relationship between modes of control at a physiological and behavioral level is presented. Examples of this phenomenon are extracted from performance under heat stress, since this is one of the most simple stress circumstances. It is suggested that stress sufficient to overcome adaptive capability, that is efficient homeostasis, acts to drain attentional resources. The manner in which such resources fail approximates that function typical of a positive feedback system, which also characterizes the breakdown of physiological response under severe environmental stress. The end point of this draining sequence is the absence of all attentional resources, which is taken to be unconsciousness, to be rapidly followed by the failure of physiological adaptability upon which life sustaining functions depend. This overall picture preserves the inverted-U shaped relationship between stress and performance, yet is in distinct contrast to the traditional arousal account of such behavior. The theoretical and practical ramifications of these observations are explored.

Hancock, P. A.↗
NASA NTRS1993

Attentional limitations with Head-Up Displays

Recent models of visual information processing suggest that visual attention can be focussed on either Head-Up Displays (HUD) or on the world beyond them, but not on both simultaneously. This hypothesis was tested in a part-task simulation in which subjects viewed a simulated approach to a runway with a HUD superimposed. An alphanumeric cue ('IFR' or 'VFR') appeared on either the HUD or the runway and was followed by two sets of three geometric forms; one set on the HUD and one set on the runway. Each set contained one potential target, either a stop sign or a diamond. If the cue spelled 'IFR', subjects made a speeded response based on the identity of the HUD target; if the cue spelled 'VFR', subjects made a speeded response based on the identity of the runway target. Regardless of cue location (HUD or Runway), responses were faster when the cue and the relevant target were part of the same perceptual group (i.e., both on the HUD or both on the runway) than when they were part of different perceptual groups. These results, as well as others, suggest that attentional constraints place severe limits on the ability of pilots to process HUD-referenced information and world-referenced information simultaneously. In addition, they provide direct evidence that transitioning from processing HUD information to processing world information requires an attention shift. Implications for HUD design are considered.

Mccann, Robert S.↗
NASA NTRS1994

Can Attention be Divided Between Perceptual Groups?

Previous work using Head-Up Displays (HUDs) suggests that the visual system parses the HUD and the outside world into distinct perceptual groups, with attention deployed sequentially to first one group and then the other. New experiments show that both groups can be processed in parallel in a divided attention search task, even though subjects have just processed a stimulus in one perceptual group or the other. Implications for models of visual attention will be discussed.

McCann, Robert S.↗
NASA NTRS1994

Spatial Attention Enhances Perceptual Processing of Single-Element Displays

Shiu and Pashler (1993) reported that precueing masked, single-element displays had negligible effects on identification accuracy. They argued that spatial attention does not actually enhance stimulus perceptibility, but only reduces decision noise. Alternatively, such negative results may arise if cues are sub-optimal, or if masks place an insufficient premium on timely deployment of attention. We report results showing that valid cueing enhances processing of even single-element displays. Spatial attention does indeed enhance perceptual processes.

Bacon, William↗
NASA NTRS2004

Head-Up Displays and Attention Capture

The primary role of head-up displays (HUDs) is to provide primary flight, navigation, and guidance information to the pilot in a forward field-of-view on a head-up transparent screen. Therefore, this theoretically allows for optimal control of an aircraft through the simultaneous scanning of both instrument data and the out-the-window scene. However, despite significant aviation safety benefits afforded by HUDs, a number of accidents have shown that their use does not come without costs. The human factors community has identified significant issues related to the pilot distribution of near and far domain attentional resources because of the compellingness of symbology elements on the HUD; a concern termed, attention or cognitive capture. The paper describes the phenomena of attention capture and presents a selected survey of the literature on the etiology and potential prescriptions.

Prinzel, Lawrence J., III↗
NASA NTRS2002

Attentional sensitivity and asymmetries of vertical saccade generation in monkey

The first goal of this study was to systematically document asymmetries in vertical saccade generation. We found that visually guided upward saccades have not only shorter latencies, but higher peak velocities, shorter durations and smaller errors. The second goal was to identify possible mechanisms underlying the asymmetry in vertical saccade latencies. Based on a recent model of saccade generation, three stages of saccade generation were investigated using specific behavioral paradigms: attention shift to a visual target (CUED paradigm), initiation of saccade generation (GAP paradigm) and release of the motor command to execute the saccade (DELAY paradigm). Our results suggest that initiation of a saccade (or "ocular disengagement") and its motor release contribute little to the asymmetry in vertical saccade latency. However, analysis of saccades made in the CUED paradigm indicated that it took less time to shift attention to a target in the upper visual field than to a target in the lower visual field. These data suggest that higher attentional sensitivity to targets in the upper visual field may contribute to shorter latencies of upward saccades.

NASA Discipline Neuroscience↗
NASA NTRS1998

A common network of functional areas for attention and eye movements

Functional magnetic resonance imaging (fMRI) and surface-based representations of brain activity were used to compare the functional anatomy of two tasks, one involving covert shifts of attention to peripheral visual stimuli, the other involving both attentional and saccadic shifts to the same stimuli. Overlapping regional networks in parietal, frontal, and temporal lobes were active in both tasks. This anatomical overlap is consistent with the hypothesis that attentional and oculomotor processes are tightly integrated at the neural level.

Non-NASA Center↗
NASA NTRS2015

A Metric to Quantify Shared Visual Attention in Two-Person Teams

Introduction: Critical tasks in high-risk environments are often performed by teams, the members of which must work together efficiently. In some situations, the team members may have to work together to solve a particular problem, while in others it may be better for them to divide the work into separate tasks that can be completed in parallel. We hypothesize that these two team strategies can be differentiated on the basis of shared visual attention, measured by gaze tracking. 2) Methods: Gaze recordings were obtained for two-person flight crews flying a high-fidelity simulator (Gontar, Hoermann, 2014). Gaze was categorized with respect to 12 areas of interest (AOIs). We used these data to construct time series of 12 dimensional vectors, with each vector component representing one of the AOIs. At each time step, each vector component was set to 0, except for the one corresponding to the currently fixated AOI, which was set to 1. This time series could then be averaged in time, with the averaging window time (t) as a variable parameter. For example, when we average with a t of one minute, each vector component represents the proportion of time that the corresponding AOI was fixated within the corresponding one minute interval. We then computed the Pearson product-moment correlation coefficient between the gaze proportion vectors for each of the two crew members, at each point in time, resulting in a signal representing the time-varying correlation between gaze behaviors. We determined criteria for concluding correlated gaze behavior using two methods: first, a permutation test was applied to the subjects' data. When one crew member's gaze proportion vector is correlated with a random time sample from the other crewmember's data, a distribution of correlation values is obtained that differs markedly from the distribution obtained from temporally aligned samples. In addition to validating that the gaze tracker was functioning reasonably well, this also allows us to compute probabilities of coordinated behavior for each value of the correlation. As an alternative, we also tabulated distributions of correlation coefficients for synthetic data sets, in which the behavior was modeled as a first-order Markov process, and compared correlation distributions for identical processes with those for disparate processes, allowing us to choose criteria and estimate error rates. 3) Discussion: Our method of gaze correlation is able to measure shared visual attention, and can distinguish between activities involving different instruments. We plan to analyze whether pilots strategies of sharing visual attention can predict performance. Possible measurements of performance include expert ratings from instructors, fuel consumption, total task time, and failure rate. While developed for two-person crews, our approach can be applied to larger groups, using intra-class correlation coefficients instead of the Pearson product-moment correlation.

Gontar, Patrick↗
NASA NTRS2023

A Meta-Analytic Approach to Investigating the Relationship Between Human-Automation Trust and Attention Allocation

Trust and attention allocation are pivotal determinants in human-automation interaction. However, there are scarce empirical findings regarding the relationship between trust and attention allocation. Observations from our previous work suggested there may be a negative correlation between trust in automation and eye movement towards automation, though no formal analysis of these data had been conducted to quantify this relationship. The present meta-analysis examined the relationship between three dimensions of trust in automation (performance, process, and purpose) and visual attention allocation to the automation. Specifically, we applied Cumming’s (2014) meta-analysis technique to combine evidence across three experiments. Results indicated a negative correlation between trust in automation and visual sampling of the automated system monitoring task for performance-based trust, but not for process- or purpose-based trust. These findings suggest that operators scanned the automation’s behavior less frequently when indicating higher performance-based trust towards the automation.

Human-Autonomy Teaming (HAT)↗