Joint Physics Dept.–Inst. for Space Weather Sci. Colloquium

November 6, Thursday (** SPECIAL DAY**)

Ten Thousand Filaments, Manually Annotated: Can AI Tell Their Chirality?

Prof. Azim Ahmadzadeh

Univ. of Missouri - St. Louis

(Solar Physics, Host: Haimin Wang)

Room: ECE 202

**Special Time: 1pm - 2pm with 12:45 pm teatime

**ZOOM Meeting ID for those who cannot attend in-person: 955 9399 6954

(APPROVAL by Prof Ahn REQUIRED for APPH/MTSE PhD Students to attend online)

*Password: check email or request from kenahn@njit.edu

The automation of detection and analysis of solar filaments has been a critical research-objective of the community for over two decades. Several major leaps have been demonstrated so far, however, the contribution of the AI-era algorithms (i.e., deep neural networks; DNN) has not been as impressive as expected. This is primarily due to the absence of large, gold-standard datasets of filaments. In this talk, I introduce MAGFiLO (https://www.mlecofi.net/magfilo), the largest manually-annotated dataset of solar filaments, which contains spatiotemporal information about over 10,000 filaments, as well as their magnetic-field chirality. MAGFiLO opens the door for the true power of supervised DNN algorithms to be explored. I will also showcase the first attempt towards automatic identification of filaments’ chirality, the true performance, and the challenges involved.