Search for: All records

Clustering is a machine learning paradigm of dividing sample subjects into a number of groups such that subjects in the same groups are more similar to those in other groups. With advances in information acquisition technologies, samples can frequently be viewed from different angles or in different modalities, generating multi-view data. Multi-view clustering, that clusters subjects into subgroups using multi-view data, has attracted more and more attentions. Although MVC methods have been developed rapidly, there has not been enough survey to summarize and analyze the current progress. Therefore, we propose a novel taxonomy of the MVC approaches. Similar with machine learning methods, we categorize them into generative and discriminative classes. In discriminative class, based on the way to integrate multiple views, we split it further into five groups: Common Eigenvector Matrix, Common Coefficient Matrix, Common Indicator Matrix, Direct Combination and Combination After Projection. Furthermore, we discuss the relationships between MVC and some related topics: multi-view representation, ensemble clustering, multi-task clustering, multi-view supervised and semi-supervised learning. Several representative real-world applications are elaborated for practitioners. Some commonly used multi-view datasets are introduced and several representative MVC algorithms from each group are run to conduct the comparison to analyze how and why they perform on those datasets. To promote future development of MVC approaches, we point out several open problems that may require further investigation and thorough examination. 

Abstract The Electron Loss and Fields Investigation with a Spatio-Temporal Ambiguity-Resolving option (ELFIN-STAR, or heretoforth simply: ELFIN) mission comprises two identical 3-Unit (3U) CubeSats on a polar (∼93 ∘ inclination), nearly circular, low-Earth (∼450 km altitude) orbit. Launched on September 15, 2018, ELFIN is expected to have a >2.5 year lifetime. Its primary science objective is to resolve the mechanism of storm-time relativistic electron precipitation, for which electromagnetic ion cyclotron (EMIC) waves are a prime candidate. From its ionospheric vantage point, ELFIN uses its unique pitch-angle-resolving capability to determine whether measured relativistic electron pitch-angle and energy spectra within the loss cone bear the characteristic signatures of scattering by EMIC waves or whether such scattering may be due to other processes. Pairing identical ELFIN satellites with slowly-variable along-track separation allows disambiguation of spatial and temporal evolution of the precipitation over minutes-to-tens-of-minutes timescales, faster than the orbit period of a single low-altitude satellite (T orbit ∼ 90 min). Each satellite carries an energetic particle detector for electrons (EPDE) that measures 50 keV to 5 MeV electrons with $\Delta $ Δ E/E < 40% and a fluxgate magnetometer (FGM) on a ∼72 cm boom that measures magnetic field waves (e.g., EMIC waves) in the range from DC to 5 Hz Nyquist (nominally) with <0.3 nT/sqrt(Hz) noise at 1 Hz. The spinning satellites (T spin $\,\sim $ ∼ 3 s) are equipped with magnetorquers (air coils) that permit spin-up or -down and reorientation maneuvers. Using those, the spin axis is placed normal to the orbit plane (nominally), allowing full pitch-angle resolution twice per spin. An energetic particle detector for ions (EPDI) measures 250 keV – 5 MeV ions, addressing secondary science. Funded initially by CalSpace and the University Nanosat Program, ELFIN was selected for flight with joint support from NSF and NASA between 2014 and 2018 and launched by the ELaNa XVIII program on a Delta II rocket (with IceSatII as the primary). Mission operations are currently funded by NASA. Working under experienced UCLA mentors, with advice from The Aerospace Corporation and NASA personnel, more than 250 undergraduates have matured the ELFIN implementation strategy; developed the instruments, satellite, and ground systems and operate the two satellites. ELFIN’s already high potential for cutting-edge science return is compounded by concurrent equatorial Heliophysics missions (THEMIS, Arase, Van Allen Probes, MMS) and ground stations. ELFIN’s integrated data analysis approach, rapid dissemination strategies via the SPace Environment Data Analysis System (SPEDAS), and data coordination with the Heliophysics/Geospace System Observatory (H/GSO) optimize science yield, enabling the widest community benefits. Several storm-time events have already been captured and are presented herein to demonstrate ELFIN’s data analysis methods and potential. These form the basis of on-going studies to resolve the primary mission science objective. Broad energy precipitation events, precipitation bands, and microbursts, clearly seen both at dawn and dusk, extend from tens of keV to >1 MeV. This broad energy range of precipitation indicates that multiple waves are providing scattering concurrently. Many observed events show significant backscattered fluxes, which in the past were hard to resolve by equatorial spacecraft or non-pitch-angle-resolving ionospheric missions. These observations suggest that the ionosphere plays a significant role in modifying magnetospheric electron fluxes and wave-particle interactions. Routine data captures starting in February 2020 and lasting for at least another year, approximately the remainder of the mission lifetime, are expected to provide a very rich dataset to address questions even beyond the primary mission science objective.