skip to main content

Search for: All records

Creators/Authors contains: "Lee, K."

Note: When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external site maintained by the publisher. Some full text articles may not yet be available without a charge during the embargo (administrative interval).
What is a DOI Number?

Some links on this page may take you to non-federal websites. Their policies may differ from this site.

  1. Teachable object recognizers provide a solution for a very practical need for blind people – instance level object recognition. They assume one can visually inspect the photos they provide for training, a critical and inaccessible step for those who are blind. In this work, we engineer data descriptors that address this challenge. They indicate in real time whether the object in the photo is cropped or too small, a hand is included, the photos is blurred, and how much photos vary from each other. Our descriptors are built into open source testbed iOS app, called MYCam. In a remote user study in (N = 12) blind participants’ homes, we show how descriptors, even when error-prone, support experimentation and have a positive impact in the quality of training set that can translate to model performance though this gain is not uniform. Participants found the app simple to use indicating that they could effectively train it and that the descriptors were useful. However, many found the training being tedious, opening discussions around the need for balance between information, time, and cognitive load.
    Free, publicly-accessible full text available October 1, 2023
  2. Abstract The heaviest elements in the universe are synthesized through rapid neutron capture ( r -process) in extremely neutron-rich outflows. Neutron star mergers were established as an important r -process source through the multimessenger observation of GW170817. Collapsars were also proposed as a potentially major source of heavy elements; however, this is difficult to probe through optical observations due to contamination by other emission mechanisms. Here we present observational constraints on r -process nucleosynthesis by collapsars based on radio follow-up observations of nearby long gamma-ray bursts (GRBs). We make the hypothesis that late-time radio emission arises from the collapsar wind ejecta responsible for forging r -process elements, and consider the constraints that can be set on this scenario using radio observations of a sample of Swift/Burst Alert Telescope GRBs located within 2 Gpc. No radio counterpart was identified in excess of the radio afterglow of the GRBs in our sample. This gives the strictest limit to the collapsar r -process contribution of ≲0.2 M ⊙ for GRB 060505 and GRB 05826, under the models we considered. Our results additionally constrain energy injection by a long-lived neutron star remnant in some of the considered GRBs. While our results are in tensionmore »with collapsars being the majority of r -process production sites, the ejecta mass and velocity profile of collapsar winds, and the emission parameters, are not yet well modeled. As such, our results are currently subject to large uncertainties, but further theoretical work could greatly improve them.« less
    Free, publicly-accessible full text available July 1, 2023
  3. Free, publicly-accessible full text available October 1, 2023
  4. Abstract We present extensive optical photometry of the afterglow of GRB 221009A. Our data cover 0.9–59.9 days from the time of Swift and Fermi gamma-ray burst (GRB) detections. Photometry in rizy -band filters was collected primarily with Pan-STARRS and supplemented by multiple 1–4 m imaging facilities. We analyzed the Swift X-ray data of the afterglow and found a single decline rate power law f ( t ) ∝ t −1.556±0.002 best describes the light curve. In addition to the high foreground Milky Way dust extinction along this line of sight, the data favor additional extinction to consistently model the optical to X-ray flux with optically thin synchrotron emission. We fit the X-ray-derived power law to the optical light curve and find good agreement with the measured data up to 5−6 days. Thereafter we find a flux excess in the riy bands that peaks in the observer frame at ∼20 days. This excess shares similar light-curve profiles to the Type Ic broad-lined supernovae SN 2016jca and SN 2017iuk once corrected for the GRB redshift of z = 0.151 and arbitrarily scaled. This may be representative of an SN emerging from the declining afterglow. We measure rest-frame absolute peak AB magnitudes ofmore »M g = −19.8 ± 0.6 and M r = − 19.4 ± 0.3 and M z = −20.1 ± 0.3. If this is an SN component, then Bayesian modeling of the excess flux would imply explosion parameters of M ej = 7.1 − 1.7 + 2.4 M ⊙ , M Ni = 1.0 − 0.4 + 0.6 M ⊙ , and v ej = 33,900 − 5700 + 5900 km s −1 , for the ejecta mass, nickel mass, and ejecta velocity respectively, inferring an explosion energy of E kin ≃ 2.6–9.0 × 10 52 erg.« less
    Free, publicly-accessible full text available March 1, 2024
  5. Distributed model training suffers from communication bottlenecks due to frequent model updates transmitted across compute nodes. To alleviate these bottlenecks, practitioners use gradient compression techniques like sparsification, quantization, or low-rank updates. The techniques usually require choosing a static compression ratio, often requiring users to balance the trade-off between model accuracy and per-iteration speedup. In this work, we show that such performance degradation due to choosing a high compression ratio is not fundamental. An adaptive compression strategy can reduce communication while maintaining final test accuracy. Inspired by recent findings on critical learning regimes, in which small gradient errors can have irrecoverable impact on model performance, we propose Accordion a simple yet effective adaptive compression algorithm. While Accordion maintains a high enough compression rate on average, it avoids over-compressing gradients whenever in critical learning regimes, detected by a simple gradient-norm based criterion. Our extensive experimental study over a number of machine learning tasks in distributed environments indicates that Accordion, maintains similar model accuracy to uncompressed training, yet achieves up to 5.5x better compression and up to 4.1x end-to-end speedup over static approaches. We show that Accordion also works for adjusting the batch size, another popular strategy for alleviating communication bottlenecks.