Seminar 1: Hierarchical Classification of Variable Stars Using Deep Convolutional and Recurrent Neural Networks

Mahdi Abdollahi

Department of Physics, Sharif University of Technology

Seminar 2: Accretion disk dynamics and luminosity

Amirhossein Mohammadi

Department of Physics, Sharif University of Technology

 

Abstract 1: The importance of using fast and automatic methods to classify variable stars for a large amount of data is undeniable. There have been many attempts at classifying Variable Stars by traditional algorithms, which require long pre-processing times. In recent years, neural networks as classifiers have come to notice. This paper proposes the Hierarchical Classification technique, which contains several models with the same network structure. We use two pre-processing methods, which produce input data by using light curves and the period. We use the OGLE variable stars’ database to train and test the performance with different models based on the Hierarchical Classification technique. Further, we use Convolutional Neural Networks and Recurrent Neural Networks in the network structure. We see these neural networks work faster than traditional methods and have more accurate predictions. We obtain the best accuracy of 98% for class classification and 92% for subclasses classification.

 

Abstract 2: Accretion disks are disk-like flows of materials that are orbiting a central object in a gravitational potential. We can see these disks as a fluid and discuss their dynamics which we will do in steady and decaying states. The central body of the accretion disks can be stars which, we will discuss the effect of the accretion disk on the luminosity of the star in the steady-state.

 

یکشنبه 24 مرداد 1400، ساعت 19:00

Sunday 15 August 2021 – 19:00 Tehran Time

اتاق سمینار مجازی –Virtual Seminar Room

https://vc.sharif.edu/ch/cosmology

گزینه ورود به صورت مهمان – Enter as a Guest