Masked Interval Research Articles

Medical multivariate time series imputation and forecasting based on a recurrent conditional Wasserstein GAN and attention

Objective:In the fields of medical care and research as well as hospital management, time series are an important part of the overall data basis. To ensure high quality standards and enable suitable decisions, tools for precise and generic imputations and forecasts that integrate the temporal dynamics are of great importance. Since forecasting and imputation tasks involve an inherent uncertainty, the focus of our work lay on a probabilistic multivariate generative approach that samples infillings or forecasts from an analysable distribution rather than producing deterministic results. Materials and Methods:For this task, we developed a system based on generative adversarial networks that consist of recurrent encoders and decoders with attention mechanisms and can learn the distribution of intervals from multivariate time series conditioned on the periods before and, if available, periods after the values that are to be predicted. For training, validation and testing, a data set of jointly measured blood pressure series (ABP) and electrocardiograms (ECG) (length: 1,250=ˆ10s) was generated. For the imputation tasks, one interval of fixed length was masked randomly and independently in both channels of every sample. For the forecasting task, all masks were positioned at the end. Results:The models were trained on around 65,000 bivariate samples and tested against 14,000 series of different persons. For the evaluation, 50 samples were produced for every masked interval to estimate the range of the generated infillings or forecasts. The element-wise arithmetic average of these samples served as an estimator for the mean of the learned conditional distribution. The approach showed better results than a state-of-the-art probabilistic multivariate forecasting mechanism based on Gaussian copula transformation and recurrent neural networks. On the imputation task, the proposed method reached a mean squared error (MSE) of 0.057 on the ECG channel and an MSE of 28.30 on the ABP channel, while the baseline approach reached MSEs of 0.095 (ECG) and 229.1 (ABP). Moreover, on the forecasting task, the presented system achieved MSEs of 0.069 (ECG) and 33.73 (ABP), outperforming the recurrent copula approach, which reached MSEs of 0.082 (ECG) and 196.53 (ABP). Conclusion:The presented generative probabilistic system for the imputation and forecasting of (medical) time series features the flexibility to handle masks of different sizes and positions, the ability to quantify uncertainty due to its probabilistic predictions, and an adjustable trade-off between the goals of minimising errors in individual predictions and minimising the distance between the learned and the real conditional distribution of the infillings or forecasts.

Anticipatory memory for regular and random patterns.

Real-world scenes elicit anticipatory representation in long-term memory (LTM) and working memory (WM) resulting in boundary extension (BE). Would the same results hold for nonscene patterns of objects? In Experiment 1A (LTM-paradigm), 15 regular or 15 random object-patterns were sequentially presented (10 s each); a boundary-rating-task followed. Both pattern-types elicited BE. Surprisingly, regularity (greater pattern predictability) did not enhance BE. Instead, random patterns more consistently yielded BE. Perhaps random object-groupings enhance BE (e.g., more scene-like). Objects within regular patterns were colored to create object-groupings; this increased BE compared to regular patterns without color groups, but only in 1 of the 2 tests (Experiment 1B, LTM-paradigm). In Experiment 2 (WM-paradigm), a briefly presented pattern, regular or random, was followed by a 2-s masked interval and a boundary-rating-task on each trial. Here, results reversed: patterns now elicited boundary restriction (whether or not a concurrent task had been presented to prevent verbalization). Under time pressure, without real-world layout and meaning, pattern elements may not have cohered into a unified expanse, failing to elicit BE and resulting in loss of peripheral content. We propose 2 classes of visual content that may determine anticipatory representation: content that embellishes space versus content that defines space. (PsycInfo Database Record (c) 2020 APA, all rights reserved).

Statistical inference for masked interval data with Weibull distribution under simple step-stress test and tampered failure rate model

In this article, we consider the planning of simple step-stress life test in the presence of competing risks with the possibility of masked causes of failure and interval censoring. It is assumed t...

Does inversion affect boundary extension for briefly-presented views?

ABSTRACTInverting scenes interferes with visual perception and memory on many tasks. Might scene inversion eliminate boundary extension (BE) for briefly-presented photographs? In Experiment 1, an upright or inverted photograph (133, 258, or 383 ms) was followed by a 258 ms masked interval and a test photograph showing the identical view. Test photographs were rated as “same”, “closer”, or “farther away” (5-point scale). BE was just as great for inverted as upright views at the 133 and 383 ms durations, but surprisingly was greater for inverted views at the 258 ms duration. In Experiment 2, 258-ms views yielded greater BE when the study photographs were always tested in the opposite orientation, indicating that the difference in BE was related to encoding. Results suggest that scene construction beyond the view boundaries occurs rapidly and is not impeded by scene inversion, but that changes in the relative quality of visual details available for upright and inverted views may sometimes yield increased BE for inverted scenes.

Looking at scenes while searching for numbers: Dividing attention multiplies space

Observers tend to remember seeing a greater expanse of a scene than was shown (boundary extension [BE]). Is undivided visual attention necessary for BE? In Experiment 1, 108 observers viewed photographs with superimposed numerals (2s and 5s). Each appeared for 750 msec, followed by a masked interval and a test picture (same, closer up, or wider angled). Test pictures were rated as the same, closer, or wider angled on a 5-point scale. Visual attention was manipulated with a search task: The observers reported the number of 5s (zero, one, or two). The observers performed search only, picture rating only, or both (giving search priority). Search accuracy was unaffected by condition. BE occurred in both conditions but was greater with divided attention. The results were replicated using incidental BE tests (Experiments 2 and 3). We propose that anticipatory representation of layout occurs automatically during scene perception, with focal attention serving to constrain the boundary error.

A novel approach for telemonitoring the VDT workers’ activity

complex data-mining. The use of other instrumentations such as baro-podometric systems or GaitRite (GatRite,Jv,Fin), is not feasible for the costs and/or for the limitation in length (they usually are limited to 3–5 m in length). Even if wearable systems for the activity monitoring are very attractive, as it is well known, they are more suitable, for the assessment of a percentage of an activity (gait, sitting, lying etc.). Other solutions should be thus explored. The specific aim of the paper is thus design an equipment able to detect parameters relevant to the gait during the daily rehabilitation that can be integrated to a general rehabilitation walkway. Methods: The general idea was to design and construct an equipment based on the following components (Fig. 1): - A step counter based on the measurement of the expansion of the gastrocnemius muscle [2] - Detectors positioned at the initial and final position of the path, to detect the start and the end of the gait exercise - One central unit for the collection and elaboration of the telemetrically connected data Results and discussion: The equipment has been fully bench tested. The stress analysis (based on the data-sheets of the components) showed a foreseen life of 20 years. Furthermore all the bench tested components did not show any failure over 5000 trials. At the moment the equipment provides the following parameters: (a) the number of steps Nstep during the masked interval of time associated to the length of the rehabilitation walkway; (b) the exact mean velocity Vm, knowing the distance d and the

Looking at scenes while searching for numbers: dividing attention multiplies space.

Observers tend to remember seeing a greater expanse of a scene than was shown (boundary extension [BE]). Is undivided visual attention necessary for BE? In Experiment 1, 108 observers viewed photographs with superimposed numerals (2s and 5s). Each appeared for 750 msec, followed by a masked interval and a test picture (same, closer up, or wider angled). Test pictures were rated as the same, closer, or wider angled on a 5-point scale. Visual attention was manipulated with a search task: The observers reported the number of 5s (zero, one, or two). The observers performed search only, picture rating only, or both (giving search priority). Search accuracy was unaffected by condition. BE occurred in both conditions but was greater with divided attention. The results were replicated using incidental BE tests (Experiments 2 and 3). We propose that anticipatory representation of layout occurs automatically during scene perception, with focal attention serving to constrain the boundary error.

Transsaccadic representation of layout: What is the time course of boundary extension?

How rapidly does boundary extension occur? Across experiments, trials included a 3-scene sequence (325 ms/picture), masked interval, and repetition of 1 scene. The repetition was the same view or differed (more close-up or wide angle). Observers rated the repetition as same as, closer than, or more wide angle than the original view on a 5-point scale. Masked intervals were 100, 250, 625, or 1,000 ms in Experiment 1 and 42, 100, or 250 ms in Experiments 2 and 3. Boundary extension occurred in all cases: Identical views were rated as too "close-up," and distractor views elicited the rating asymmetry typical of boundary extension (wider angle distractors were rated as being more similar to the original than were closer up distractors). Most important, boundary extension was evident when only a 42-ms mask separated the original and test views. Experiments 1 and 3 included conditions eliciting a gaze shift prior to the rating test; this did not eliminate boundary extension. Results show that boundary extension is available soon enough and is robust enough to play an on-line role in view integration, perhaps supporting incorporation of views within a larger spatial framework.

Crossmodal change blindness between vision and touch

Change blindness is the name given to people’s inability to detect changes introduced between two consecutively-presented scenes when they are separated by a distractor that masks the transients that are typically associated with change. Change blindness has been reported within vision, audition, and touch, but has never before been investigated when successive patterns are presented to different sensory modalities. In the study reported here, we investigated change detection performance when the two to-be-compared stimulus patterns were presented in the same sensory modality (i.e., both visual or both tactile) and when one stimulus pattern was tactile while the other was presented visually or vice versa. The two to-be-compared patterns were presented consecutively, separated by an empty interval, or else separated by a masked interval. In the latter case, the masked interval could either be tactile or visual. The first experiment investigated visual–tactile and tactile–visual change detection performance. The results showed that in the absence of masking, participants detected changes in position accurately, despite the fact that the two to-be-compared displays were presented in different sensory modalities. Furthermore, when a mask was presented between the two to-be-compared displays, crossmodal change blindness was elicited no matter whether the mask was visual or tactile. The results of two further experiments showed that performance was better overall in the unimodal (visual or tactile) conditions than in the crossmodal conditions. These results suggest that certain of the processes underlying change blindness are multisensory in nature. We discuss these findings in relation to recent claims regarding the crossmodal nature of spatial attention.

More than meets the eye: The effect of planned fixations on scene representation

Scene memory frequently includes a swath of unseen layout beyond a photograph's boundaries (boundary extension [BE]; Intraub and Richardson, 1989). Might it be affected by the viewer's plan to shift fixation near a view boundary? When photographs were centrally fixated (500 msec), BE occurred following a 2-sec masked interval (Experiment 1). In Experiments 2-4, a cue during the first fixation signaled viewers to fixate an object near the left or right boundary. The picture was masked before the eyes landed. BE occurred on the cued side and on the top and bottom, but not on the uncued side. This relatively accurate performance on the uncued side suggests that inhibition of a movement to one side (in a competitive task) may also inhibit extrapolation of layout. BE on the to-be-fixated side, however, supports the idea that anticipatory representation of layout is an adaptive error that may aid the spatial integration of successive views.

The failure to detect tactile change: A tactile analogue of visual change blindness

A large body of empirical research now shows that people are surprisingly poor at detecting significant changes in visually presented scenes. This phenomenon is known as change blindness in vision. A similar phenomenon occurs in audition, but to date no such effect has been documented in touch. In the present study, we explored the ability of people to detect changes introduced between two consecutively presented vibrotactile patterns presented over the body surface. The patterns consisted of two or three vibrotactile stimuli presented for 200 msec. The position of one of the vibrotactile stimuli composing the display was repeatedly changed (alternating between two different positions) on 50% of the trials, but the same pattern was presented repeatedly on the remaining trials. Three conditions were investigated: No interval between the patterns, an empty interval between the patterns, and a masked interval between the patterns. Change detection was near perfect in the no-interval block. Performance deteriorated somewhat in the empty-interval block, but by far the worst change detection performance occurred in the masked-interval block. These results demonstrate that "change blindness" can also affect tactile perception.

Masked Interval Routing: A New Routing Scheme

We introduce the new Masked Interval Routing Scheme, MIRS for short, where a maskis added to each interval to indicate particular subsets of consecutive labels. Interval routing becomes more flexible, with the classical IRS scheme being a special case of MIRS. We then take two directions. First we show that the interval information stored in the network may be drastically reduced in the hard cases, proving that in globe graphs of O(n^2) vertices the number of intervals per edge goes down from Omega(n) to O(log n). The technique is then extended to globe graphs of arbitrary dimensions. Second we show that MIRS may be advantageously used in fault-tolerant networks, proving that optimal routing with one interval per edge is still possible in hypercubes with a harmless subset of faulty vertices. This work is aimed to introducing a new technique. Further research is needed in both the directions taken here. Still, the examples provided show that MIRS may be useful in practical applications.