Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Detecting and Tracking Communal Bird Roosts in Weather Radar Data
Authors: Zezhou Cheng, Saadia Gabriel, Pankaj Bhambhani, Daniel Sheldon, Subhransu Maji, Andrew Laughlin, David Winkler378-385
AAAI 2020 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | This paper describes a machine learning system to detect and track roost signatures in weather radar data. We contribute a latent-variable model and EM algorithm to learn a detection model together with models of labeling styles for individual annotators. We divided the 88972 radar scans from the manually labeled dataset (Sec. 2) into training, validation, and test sets. Tab. 1 shows the performance of various detectors across radar stations. |
| Researcher Affiliation | Academia | Zezhou Cheng UMass Amherst EMAIL Saadia Gabriel University of Washington EMAIL Pankaj Bhambhani UMass Amherst EMAIL Daniel Sheldon UMass Amherst EMAIL Subhransu Maji UMass Amherst EMAIL Andrew Laughlin UNC Asheville EMAIL David Winkler Cornell University EMAIL |
| Pseudocode | No | The paper does not contain any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper mentions "See the supplementary material on the project page for details on these baseline detection models." and includes a footnote "1See: https://people.cs.umass.edu/ zezhoucheng/roosts". This is a personal homepage and not an explicit statement of code release to a repository for the methodology described. |
| Open Datasets | Yes | We obtained a data set of manually annotated roosts collected for prior ecological research (Laughlin et al. 2016). |
| Dataset Splits | Yes | We divided the 88972 radar scans from the manually labeled dataset (Sec. 2) into training, validation, and test sets. The validation set (not shown) is roughly half the size of the test set and was used to set the hyper-parameters of the detector and the tracker. |
| Hardware Specification | No | The paper mentions "This research was supported in part by NSF #1749833, #1749854, #1661259 and the Mass Tech Collaborative for funding the UMass GPU cluster." This indicates GPU usage but lacks specific model numbers or detailed hardware specifications. |
| Software Dependencies | No | The paper mentions using "Faster R-CNN" and "VGG-M network" but does not provide specific version numbers for software dependencies like Python, PyTorch, TensorFlow, or other libraries. |
| Experiment Setup | Yes | We initialize the Faster R-CNN parameters θ by training for 50K iterations starting from the Image Net pretrained VGG-M model using the original uncorrected labels. The optimization is performed separately to determine the reverse scaling factor φu for each user using Brent s method with search boundary [0.1, 2] and black-box access to Lcnn. then update θ by training Faster R-CNN for 50K iterations using the resampled annotations. |