Supernova Recognition using Support Vector Machines
We introduce a novel application of Support Vector Machines(SVMs) to the problem of identifying potential supernovae usingphotometric and geometric features computed from astronomical imagery.The challenges of this supervised learning application are significant:1) noisy and corrupt imagery resulting in high levels of featureuncertainty,2) features with heavy-tailed, peaked distributions,3)extremely imbalanced and overlapping positiveand negative data sets, and4) the need to reach high positive classification rates, i.e. to find allpotential supernovae, while reducing the burdensome workload of manuallyexamining false positives. High accuracy is achieved viaasign-preserving, shifted log transform applied to features with peaked,heavy-tailed distributions. The imbalanced data problem is handled byoversampling positive examples,selectively sampling misclassifiednegative examples,and iteratively training multiple SVMs for improvedsupernovarecognition on unseen test data. We present crossvalidationresults and demonstrate the impact on a largescale supernova survey thatcurrently uses the SVM decision value to rank-order 600,000 potentialsupernovae each night.
- Research Organization:
- Ernest Orlando Lawrence Berkeley NationalLaboratory, Berkeley, CA (US)
- Sponsoring Organization:
- USDOE Director. Office of Science. Advanced ScientificComputing Research
- DOE Contract Number:
- AC02-05CH11231
- OSTI ID:
- 920165
- Report Number(s):
- LBNL--61192; BnR: YN0100000
- Country of Publication:
- United States
- Language:
- English
Similar Records
Discriminating Tailing Piles using VNIR Imagery and SVMs
Aerial Scene Recognition using Efficient Sparse Representation