More Like this

1. PrimeNet: Pre-training for Irregular Multivariate Time Series

   Chowdhury, Ranak Roy; Li, Jiacheng; Zhang, Xiyuan; Hong, Dezhi; Gupta, Rajesh; Shang, Jingbo (January 2023, Proceedings of the AAAI Conference on Artificial Intelligence)

   Real-world applications often involve irregular time series, for which the time intervals between successive observations are non-uniform. Irregularity across multiple features in a multi-variate time series further results in a different subset of features at any given time (i.e., asynchronicity). Existing pre-training schemes for time-series, however, often assume regularity of time series and make no special treatment of irregularity. We argue that such irregularity offers insight about domain property of the data—for example, frequency of hospital visits may signal patient health condition—that can guide representation learning. In this work, we propose PrimeNet to learn a self-supervised representation for irregular multivariate time-series. Specifically, we design a timesensitive contrastive learning and data reconstruction task to pre-train a model. Irregular time-series exhibits considerable variations in sampling density over time. Hence, our triplet generation strategy follows the density of the original data points, preserving its native irregularity. Moreover, the sampling density variation over time makes data reconstruction difficult for different regions. Therefore, we design a data masking technique that always masks a constant time duration to accommodate reconstruction for regions of different sampling density. We learn with these tasks using unlabeled data to build a pre-trained model and fine-tune on a downstream task with limited labeled data, in contrast with existing fully supervised approach for irregular time-series, requiring large amounts of labeled data. Experiment results show that PrimeNet significantly outperforms state-of-the-art methods on naturally irregular and asynchronous data from Healthcare and IoT applications for several downstream tasks, including classification, interpolation, and regression. 

   more » « less
2. Effects of grid spacing on high-frequency precipitation variance in coupled high-resolution global ocean–atmosphere models

   https://doi.org/10.1007/s00382-022-06257-6  

   Light, Charles X.; Arbic, Brian K.; Martin, Paige E.; Brodeau, Laurent; Farrar, J. Thomas; Griffies, Stephen M.; Kirtman, Ben P.; Laurindo, Lucas C.; Menemenlis, Dimitris; Molod, Andrea; et al (March 2022, Climate Dynamics)

   Abstract High-frequency precipitation variance is calculated in 12 different free-running (non-data-assimilative) coupled high resolution atmosphere–ocean model simulations, an assimilative coupled atmosphere–ocean weather forecast model, and an assimilative reanalysis. The results are compared with results from satellite estimates of precipitation and rain gauge observations. An analysis of irregular sub-daily fluctuations, which was applied by Covey et al. (Geophys Res Lett 45:12514–12522, 2018.https://doi.org/10.1029/2018GL078926) to satellite products and low-resolution climate models, is applied here to rain gauges and higher-resolution models. In contrast to lower-resolution climate simulations, which Covey et al. (2018) found to be lacking with respect to variance in irregular sub-daily fluctuations, the highest-resolution simulations examined here display an irregular sub-daily fluctuation variance that lies closer to that found in satellite products. Most of the simulations used here cannot be analyzed via the Covey et al. (2018) technique, because they do not output precipitation at sub-daily intervals. Thus the remainder of the paper focuses on frequency power spectral density of precipitation and on cumulative distribution functions over time scales (2–100 days) that are still relatively “high-frequency” in the context of climate modeling. Refined atmospheric or oceanic model grid spacing is generally found to increase high-frequency precipitation variance in simulations, approaching the values derived from observations. Mesoscale-eddy-rich ocean simulations significantly increase precipitation variance only when the atmosphere grid spacing is sufficiently fine (< 0.5°). Despite the improvements noted above, all of the simulations examined here suffer from the “drizzle effect”, in which precipitation is not temporally intermittent to the extent found in observations. 

   more » « less
3. Conformal irregular-grid metasurfaces comprising a dense array of free-standing dipoles

   Kwon, Do-Hoon (July 2022, 2022 IEEE International Symposium on Antennas and Propagation)

   For large or infinite metasurfaces, a design tech- nique for a dense array of subwavelength resonators on an irregular grid is presented. For a given incident wave, the desired induced dipole moment distribution determines the local electric field that excites individual meta-atoms. The interaction field that accounts for mutual coupling is evaluated via a combination of discrete coupling from nearby resonators and continuous sheet current coupling from far-separated resonators. Meta-atoms placed on an irregular grid can be treated, greatly enhancing the flexibility in surface profile in practical conformal metasurfaces. 

   more » « less
4. Loaded Microstrip Network Transmision-Line Metamaterials on an Irregular Grid

   https://doi.org/10.1109/AP-S/USNC-URSI47032.2022.9886547  

   Maurer, Tina E.; Kwon, Do-Hoon (July 2022, 2022 IEEE International Symposium on Antennas and Propagation)

   A planar loaded microstrip network metamaterial on an irregular grid for modeling 2-D electromagnetic environment in TE polarization is presented. The technique is applied to a Luneburg lens example toward printed circuit implementation and verification. 

   more » « less
5. Wave-by-Wave Control of a Wave Energy Converter with Deterministic Wave Prediction

   Korde, Umesh A. (September 2017, European Wave and Tidal Energy Conference)

   This paper discusses wave-by-wave near-optimal control of a wave energy device in irregular waves. A deterministic propagation model is used to predict the wave elevation several seconds into the future at the device location. Two prediction approaches are considered. The first is based on a time series being measured over an advancing time window at a particular up-wave location. This approach is here utilized in long-crested irregular waves. The second approach uses successive snapshots of wave elevation measurements over an up-wave area. This approach is found more convenient for multi-directional waves, and is here applied in a bi-directional wave irregular wave field. A small, heaving vertical cylinder reacting against a deeply submerged (i.e. assumed to undergo negligible oscillations) mass is studied under wave-by-wave control. The non-causal feedforward control force required for optimum velocity under a swept-volume constraint is based on the past, current, and predicted wave elevation at the device. Results for time-averaged converted power and displacement/force maxima are obtained for a range of irregular wave conditions. Also presented in addition are energy conversion results with a feedback-alone control force using a multi-resonant control technique. 

   more » « less