The current archive of the LAMOST telescope contains about 4 million of astronomical spectra. The goal of the thesis is to identify objects with interesting spectral line profiles (namely emission line stars) based on a small set of examples.
- Make a survey of semi-supervised training methods and define the optimal strategy of learning, taking into consideration massively parallel solutions (e.g., SPARK, GPUs...).
- Select spectra of known emission-line stars from archives of the Ondřejov 2m telescope.
- Make a positional cross-match of objects from 2) with LAMOST survey to get a sample for training.
- If the sample obtained in 3) is insufficient, simulate the expected appearance of spectra from 2) in the LAMOST spectrograph to get a training sample.
- Identify new objects of the same class in the whole LAMOST survey and make visual confirmation.
- Discuss the success rate and computing performance of your strategy. Try to integrate your solution into the system VO-CLOUD.