Search

Large language models (LLMs) have revolutionized how we interact with technology, but their personalization to individual user preferences remains a significant challenge, particularly in on-device applications. Traditional methods often depend heavily on labeled datasets and can be resource-intensive. To address these issues, we present Adaptive Self-Supervised Learning Strategies (ASLS), which utilizes self-supervised learning techniques to personalize LLMs dynamically. The framework comprises a user profiling layer for collecting interaction data and a neural adaptation layer for real-time model fine-tuning. This innovative approach enables continuous learning from user feedback, allowing the model to generate responses that align closely with user-specific contexts. The adaptive mechanisms of ASLS minimize computational demands and enhance personalization efficiency. Experimental results across various user scenarios illustrate the superior performance of ASLS in boosting user engagement and satisfaction, highlighting its potential to redefine LLMs as highly responsive and context-aware systems on-device., Comment: First ASLS

This work introduces a new solution-transfer process for slab-based space-time finite element methods. The new transfer process is based on Hsieh-Clough-Tocher (HCT) splines and satisfies the following requirements: (i) it maintains high-order accuracy up to 4th order, (ii) it preserves a discrete maximum principle, (iii) it asymptotically enforces mass conservation, and (iv) it constructs a smooth, continuous surrogate solution in between space-time slabs. While many existing transfer methods meet the first three requirements, the fourth requirement is crucial for enabling visualization and boundary condition enforcement for space-time applications. In this paper, we derive an error bound for our HCT spline-based transfer process. Additionally, we conduct numerical experiments quantifying the conservative nature and order of accuracy of the transfer process. Lastly, we present a qualitative evaluation of the visualization properties of the smooth surrogate solution., Comment: 36 pages, 17 figures, 3 tables

Large language models (LLMs) are increasingly integrated into real-time machine learning applications, where safeguarding user privacy is paramount. Traditional differential privacy mechanisms often struggle to balance privacy and accuracy, particularly in fast-changing environments with continuously flowing data. To address these issues, we introduce Scalable Differential Privacy (SDP), a framework tailored for real-time machine learning that emphasizes both robust privacy guarantees and enhanced model performance. SDP employs a hierarchical architecture to facilitate efficient noise aggregation across various learning agents. By integrating adaptive noise scheduling and gradient compression methods, our approach minimizes performance degradation while ensuring significant privacy protection. Extensive experiments on diverse datasets reveal that SDP maintains high accuracy levels while applying differential privacy effectively, showcasing its suitability for deployment in sensitive domains. This advancement points towards the potential for widespread adoption of privacy-preserving techniques in machine learning workflows., Comment: First v of SDP

The ideal magnetohydrodynamic torus instability can drive the eruption of coronal mass ejections. The critical threshold of magnetic field strength decay for the onset of the torus instability occurs at different heights in different solar active regions, and understanding this variation could therefore improve space weather prediction. In this work, we aim to find out how the critical torus instability height evolves throughout the solar activity cycle. We study a significant subset of HMI and MDI Space-Weather HMI Active Region Patches (SHARPs and SMARPs) from 1996-2023, totalling 21584 magnetograms from 4436 unique active region patches. For each magnetogram, we compute the critical height averaged across the main polarity inversion line, the total unsigned magnetic flux and the separation between the positive and negative magnetic polarities. We find the critical height in active regions varies with the solar cycle, with higher (lower) average critical heights observed around solar maximum (minimum). We conclude this is because the critical height is proportional to the separation between opposite magnetic polarities, which in turn is proportional to the total magnetic flux in a region, and more magnetic regions with larger fluxes and larger sizes are observed at solar maximum. This result is noteworthy because, despite the higher critical heights, more CMEs are observed around solar maximum than at solar minimum., Comment: Accepted to ApJ. 14 pages, 6 Figures, 1 Table

We calculate the minimal attached primes of the local cohomology modules of the binomial edge ideals of block graphs. In particular, we obtain a combinatorial characterisation of which of these modules are non-vanishing. We also show that the main result of this paper follows from a recent result of Lax, Rinaldo, and Romeo (arXiv:2405.08671, Theorem 3.2), which was published independently during the writing of this paper. This provides a short alternative proof of our result., Comment: 16 pages, v2 replaces what was Section 5 in v1 to take into account recent work of Lax, Rinaldo, and Romeo (arXiv:2405.08671), in which they had in fact already proved a result we stated as a conjecture. Many thanks to Ernesto Lax for bringing this to our attention

Since volcanic activity was first discovered on Io from Voyager images in 1979, changes on Io's surface have been monitored from both spacecraft and ground-based telescopes. Here, we present the highest spatial resolution images of Io ever obtained from a ground-based telescope. These images, acquired by the SHARK-VIS instrument on the Large Binocular Telescope, show evidence of a major resurfacing event on Io's trailing hemisphere. When compared to the most recent spacecraft images, the SHARK-VIS images show that a plume deposit from a powerful eruption at Pillan Patera has covered part of the long-lived Pele plume deposit. Although this type of resurfacing event may be common on Io, few have been detected due to the rarity of spacecraft visits and the previously low spatial resolution available from Earth-based telescopes. The SHARK-VIS instrument ushers in a new era of high resolution imaging of Io's surface using adaptive optics at visible wavelengths., Comment: 15 pages, 4 figures

In this work, a family of symmetric interpolation points are generated on the four-dimensional simplex (i.e. the pentatope). These points are optimized in order to minimize the Lebesgue constant. The process of generating these points closely follows that outlined by Warburton in "An explicit construction of interpolation nodes on the simplex," Journal of Engineering Mathematics, 2006. Here, Warburton generated optimal interpolation points on the triangle and tetrahedron by formulating explicit geometric warping and blending functions, and applying these functions to equidistant nodal distributions. The locations of the resulting points were Lebesgue-optimized. In our work, we extend this procedure to four dimensions, and construct interpolation points on the pentatope up to order ten. The Lebesgue constants of our nodal sets are calculated, and are shown to outperform those of equidistant nodal distributions.

If a reduced bivariate polynomial is quasi-homogeneous, then its discriminant is a monomial. Over fields of characteristic $0$, we show that if one adds another simple condition, this becomes an equivalence. We also give a third equivalent condition that is stated geometrically., Comment: Fixed minor mistakes, reordered the introduction, streamlined some terminology and extended the result to large positive characteristic

We present a diffusion-based, generative model for conformer generation. Our model is focused on the reproduction of bonded structure and is constructed from the associated terms traditionally found in classical force fields to ensure a physically relevant representation. Techniques in deep learning are used to infer atom typing and geometric parameters from a training set. Conformer sampling is achieved by taking advantage of recent advancements in diffusion-based generation. By training on large, synthetic data sets of diverse, drug-like molecules optimized with the semiempirical GFN2-xTB method, high accuracy is achieved for bonded parameters, exceeding that of conventional, knowledge-based methods. Results are also compared to experimental structures from the Protein Databank (PDB) and Cambridge Structural Database (CSD)., Comment: To appear in the Journal of Chemical Information and Modeling

Versatile mixed finite element methods were originally developed by Chen and Williams for isothermal incompressible flows in "Versatile mixed methods for the incompressible Navier-Stokes equations," Computers & Mathematics with Applications, Volume 80, 2020. Thereafter, these methods were extended by Miller, Chen, and Williams to non-isothermal incompressible flows in "Versatile mixed methods for non-isothermal incompressible flows," Computers & Mathematics with Applications, Volume 125, 2022. The main advantage of these methods lies in their flexibility. Unlike traditional mixed methods, they retain the divergence terms in the momentum and temperature equations. As a result, the favorable properties of the schemes are maintained even in the presence of non-zero divergence. This makes them an ideal candidate for an extension to compressible flows, in which the divergence does not generally vanish. In the present article, we finally construct the fully-compressible extension of the methods. In addition, we demonstrate the excellent performance of the resulting methods for weakly-compressible flows that arise near the incompressible limit, as well as more strongly-compressible flows that arise near Mach 0.5., Comment: 28 pages, 4 figures, 4 tables

Knowing when a trained segmentation model is encountering data that is different to its training data is important. Understanding and mitigating the effects of this play an important part in their application from a performance and assurance perspective - this being a safety concern in applications such as autonomous vehicles (AVs). This work presents a segmentation network that can detect errors caused by challenging test domains without any additional annotation in a single forward pass. As annotation costs limit the diversity of labelled datasets, we use easy-to-obtain, uncurated and unlabelled data to learn to perform uncertainty estimation by selectively enforcing consistency over data augmentation. To this end, a novel segmentation benchmark based on the SAX Dataset is used, which includes labelled test data spanning three autonomous-driving domains, ranging in appearance from dense urban to off-road. The proposed method, named Gamma-SSL, consistently outperforms uncertainty estimation and Out-of-Distribution (OoD) techniques on this difficult benchmark - by up to 10.7% in area under the receiver operating characteristic (ROC) curve and 19.2% in area under the precision-recall (PR) curve in the most challenging of the three scenarios., Comment: Accepted for publication in IEEE Transactions on Robotics (T-RO)

This work proposes a semantic segmentation network that produces high-quality uncertainty estimates in a single forward pass. We exploit general representations from foundation models and unlabelled datasets through a Masked Image Modeling (MIM) approach, which is robust to augmentation hyper-parameters and simpler than previous techniques. For neural networks used in safety-critical applications, bias in the training data can lead to errors; therefore it is crucial to understand a network's limitations at run time and act accordingly. To this end, we test our proposed method on a number of test domains including the SAX Segmentation benchmark, which includes labelled test data from dense urban, rural and off-road driving domains. The proposed method consistently outperforms uncertainty estimation and Out-of-Distribution (OoD) techniques on this difficult benchmark., Comment: Accepted for publication at 2024 IEEE International Conference on Robotics and Automation (ICRA)

This paper provides a comprehensive guide to generating unconstrained, simplicial, four-dimensional (4D), hypervolume meshes for space-time applications. While several universal procedures for constructing unconstrained, d-dimensional, anisotropic Delaunay meshes are already known, many of the explicit implementation details are missing from the relevant literature for cases in which d >= 4. As a result, the purpose of this paper is to provide explicit descriptions of the key components in the 4D meshing algorithm: namely, the point-insertion process, geometric predicates, element quality heuristics, and bistellar flips. This paper represents a natural continuation of the work which was pioneered by Anderson et al. in "Surface and hypersurface meshing techniques for space-time finite element methods", Computer-Aided Design, 2023. In this previous paper, hypersurface meshes were generated using a novel, trajectory-tracking procedure. In the current paper, we are interested in generating coarse, 4D hypervolume meshes (boundary meshes) which are formed by sequentially inserting points from an existing hypersurface mesh. In the latter portion of this paper, we present numerical experiments which demonstrate the viability of this approach for a simple, convex domain. Although, our main focus is on the generation of hypervolume boundary meshes, the techniques described in this paper are broadly applicable to a much wider range of 4D meshing methods. We note that the more complex topics of constrained hypervolume meshing, and boundary recovery for non-convex domains will be covered in a companion paper., Comment: 65 pages, 33 figures, and 11 tables

We calculate the local cohomology modules of the binomial edge ideals of the complements of connected graphs of girth at least 5 using the tools introduced by \`Alvarez Montaner in arXiv:1901.08645. We then use this calculation to compute the depth, dimension, and regularity of these binomial edge ideals., Comment: 15 pages, v2 includes small changes to several proofs for clarity, and reorganises section 2 into subsections (whilst some numbering has changed, no new results are introduced)

We present here broadband radio observations of the CSM interacting SN2019oys. SN2019oys was first detected in the optical and was classified as a Type Ib SN. Then, about $\sim 100$ days after discovery, it showed an optical rebrightening and a spectral transition to a spectrum dominated by strong narrow emission lines, which suggests strong interaction with a distant, dense, CSM shell. We modeled the broadband, multi-epoch, radio spectra, covering 2.2 to 36 GHz and spanning from 22 to 1425 days after optical discovery, as a synchrotron emitting source. Using this modeling we characterized the shockwave and the mass-loss rate of the progenitor. Our broadband radio observations show strong synchrotron emission. This emission, as observed 201 and 221 days after optical discovery, exhibits signs of free-free absorption from the material in front of the shock traveling in the CSM. In addition, the steep power law of the optically thin regime points towards synchrotron cooling of the radiating electrons. Analyzing these spectra in the context of the SN-CSM interaction model gives a shock velocity of 14,000 $\rm km \, s^{-1}$, and an electron number density of $2.6 \times 10^5 \, \rm cm^{-3}$ at a distance of $2.6 \times 10^{16}$ cm. This translates to a high mass-loss rate from the progenitor massive star of $6.7 \times 10^{-4} \, \rm M_{\odot} yr^{-1}$ for an assumed wind of 100 $\rm km s^{-1}$ (assuming constant mass-loss rate in steady winds). The late-time radio spectra, 392 and 557 days after optical discovery, are showing broad spectral peaks. We show that this can be explained by introducing a non-homogeneous CSM structure., Comment: 14 pages, 7 figures, 2 tables, submitted to Astronomy & Astrophysics

Automated program repair (APR) tools have unlocked the potential for the rapid rectification of codebase issues. However, to encourage wider adoption of program repair in practice, it is necessary to address the usability concerns related to generating irrelevant or out-of-context patches. When software engineers are presented with patches they deem uninteresting or unhelpful, they are burdened with more "noise" in their workflows and become less likely to engage with APR tools in future. This paper presents a novel approach to optimally time, target, and present auto-generated patches to software engineers. To achieve this, we designed, developed, and deployed a new tool dubbed B-Assist, which leverages GitHub's Suggested Changes interface to seamlessly integrate automated suggestions into active pull requests (PRs), as opposed to creating new, potentially distracting PRs. This strategy ensures that suggestions are not only timely, but also contextually relevant and delivered to engineers most familiar with the affected code. Evaluation among Bloomberg software engineers demonstrated their preference for this approach. From our user study, B-Assist's efficacy is evident, with the acceptance rate of patch suggestions being as high as 74.56%; engineers also found the suggestions valuable, giving usefulness ratings of at least 4 out of 5 in 78.2% of cases. Further, this paper sheds light on persisting usability challenges in APR and lays the groundwork for enhancing the user experience in future APR tools.

If $G$ is a graph, $A$ and $B$ its induced subgraphs, and $f\colon A\to B$ an isomorphism, we say that $f$ is a \emph{partial automorphism} of $G$. In 1992, Hrushovski proved that graphs have the \emph{extension property for partial automorphisms} (\emph{EPPA}, also called the \emph{Hrushovski property}), that is, for every finite graph $G$ there is a finite graph $H$, an \emph{EPPA-witness} for $G$, such that $G$ is an induced subgraph of $H$ and every partial automorphism of $G$ extends to an automorphism of $H$. The EPPA number of a graph $G$, denoted by $\mathop{\mathrm{eppa}}\nolimits(G)$, is the smallest number of vertices of an EPPA-witness for $G$, and we put $\mathop{\mathrm{eppa}}\nolimits(n) = \max\{\mathop{\mathrm{eppa}}\nolimits(G) : \lvert G\rvert = n\}$. In this note we review the state of the area, prove several lower bounds (in particular, we show that $\mathop{\mathrm{eppa}}\nolimits(n)\geq \frac{2^n}{\sqrt{n}}$, thereby identifying the correct base of the exponential) and pose many open questions. We also briefly discuss EPPA numbers of hypergraphs, directed graphs, and $K_k$-free graphs., Comment: Minor revision

We give a combinatorial characterisation of connected graphs whose binomial edge ideals are of K\"{o}nig type, developed independently to the similar characterisation given by LaClair in arXiv:2304.13299, and exhibit some classes of graphs satisfying our criteria. For any connected Hamiltonian graph $G$ on $n$ vertices, we compute an explicit root of $H_{J(G)}^{n-1}(R)$ as an $F$-finite $F$-module., Comment: 19 pages, v3 substantially reorganises v2, several proofs have been simplified or rewritten for clarity, and small errors have been corrected (whilst the numbering has changed, no new results are introduced)

The distribution of Io's volcanic activity likely reflects the position and magnitude of internal tidal heating. We use new observations of Io's polar regions by the Juno spacecraft Jovian Infrared Auroral Mapper (JIRAM) to complete near-infrared global coverage, revealing the global distribution and magnitude of thermal emission from Io's currently erupting volcanoes. We show that the distribution of volcanic heat flow from 266 active hot spots is consistent with the presence of a global magma ocean, and/or shallow asthenospheric heating. We find that Io's polar volcanoes are less energetic but about the same in number per unit area than at lower latitudes. We also find that volcanic heat flow in the north polar cap is greater than that in the south. The low volcanic advection seen at Io's poles is therefore at odds with measurements of background temperature showing Io's poles are anomalously warm. We suggest that the differences in volcanic thermal emission from Io's poles compared to that at lower latitudes is indicative of lithospheric dichotomies that inhibit volcanic advection towards Io's poles, particularly in the south polar region., Comment: 17 pages, two tables, 7 figures

The detection of gravitational waves from a binary neutron star merger by Advanced LIGO and Advanced Virgo (GW170817), along with the discovery of the electromagnetic counterparts of this gravitational wave event, ushered in a new era of multimessenger astronomy, providing the first direct evidence that BNS mergers are progenitors of short gamma-ray bursts (GRBs). Such events may also produce very-high-energy (VHE, > 100GeV) photons which have yet to be detected in coincidence with a gravitational wave signal. The Cherenkov Telescope Array (CTA) is a next-generation VHE observatory which aims to be indispensable in this search, with an unparalleled sensitivity and ability to slew anywhere on the sky within a few tens of seconds. New observing modes and follow-up strategies are being developed for CTA to rapidly cover localization areas of gravitational wave events that are typically larger than the CTA field of view. This work will evaluate and provide estimations on the expected number of of gravitational wave events that will be observable with CTA, considering both on- and off-axis emission. In addition, we will present and discuss the prospects of potential follow-up strategies with CTA., Comment: Presented at the 38th International Cosmic Ray Conference (ICRC 2023), 2023 (arXiv:2309.08219)

This work explores Doppler information from a millimetre-Wave (mm-W) Frequency-Modulated Continuous-Wave (FMCW) scanning radar to make odometry estimation more robust and accurate. Firstly, doppler information is added to the scan masking process to enhance correlative scan matching. Secondly, we train a Neural Network (NN) for regressing forward velocity directly from a single radar scan; we fuse this estimate with the correlative scan matching estimate and show improved robustness to bad estimates caused by challenging environment geometries, e.g. narrow tunnels. We test our method with a novel custom dataset which is released with this work at https://ori.ox.ac.uk/publications/datasets., Comment: Accepted to ITSC 2023

The Panoramic Search for Extraterrestrial Intelligence (PANOSETI) experiment is designed to detect pulsed optical signals on nanosecond timescales. PANOSETI is therefore sensitive to Cherenkov radiation generated by extensive air showers, and can be used for gamma-ray astronomy. Each PANOSETI telescope uses a 0.5 m Fresnel lens to focus light onto a 1024 pixel silicon photomultiplier camera that images a 9.9$^\circ\times$9.9$^\circ$ square field of view. Recent detections of PeV gamma-rays from extended sources in the Galactic Plane motivate constructing an array with effective area and angular resolution surpassing current observatories. The PANOSETI telescopes are much smaller and far more affordable than traditional imaging atmospheric Cherenkov telescopes (IACT), making them ideal instruments to construct such an array. We present the results of coincident observations between two PANOSETI telescopes and the gamma-ray observatory VERITAS, along with simulations characterizing the performance of a PANOSETI IACT array., Comment: 8 pages, 5 figures, Proceedings for the 38th International Cosmic Ray Conference (ICRC2023) in Nagoya Japan, Jul 26 - Aug 3, 2023

In this paper, we present explicit expressions for conforming finite element function spaces, basis functions, and degrees of freedom on the pentatope and tetrahedral prism elements. More generally, our objective is to construct finite element function spaces that maintain conformity with infinite-dimensional spaces of a carefully chosen de Rham complex. This paper is a natural extension of the companion paper entitled "Conforming Finite Element Function Spaces in Four Dimensions, Part I: Foundational Principles and the Tesseract" by Nigam and Williams, (2023). In contrast to Part I, in this paper we focus on two of the most popular elements which do not possess a full tensor-product structure in all four coordinate directions. We note that these elements appear frequently in existing space-time finite element methods. In order to build our finite element spaces, we utilize powerful techniques from the recently developed 'Finite Element Exterior Calculus'. Subsequently, we translate our results into the well-known language of linear algebra (vectors and matrices) in order to facilitate implementation by scientists and engineers., Comment: 44 pages, 2 figures, 1 table

The stability, robustness, accuracy, and efficiency of space-time finite element methods crucially depend on the choice of approximation spaces for test and trial functions. This is especially true for high-order, mixed finite element methods which often must satisfy an inf-sup condition in order to ensure stability. With this in mind, the primary objective of this paper and a companion paper is to provide a wide range of explicitly stated, conforming, finite element spaces in four-dimensions. In this paper, we construct explicit high-order conforming finite elements on 4-cubes (tesseracts); our construction uses tools from the recently developed `Finite Element Exterior Calculus'. With a focus on practical implementation, we provide details including Piola-type transformations, and explicit expressions for the volumetric, facet, face, edge, and vertex degrees of freedom. In addition, we establish important theoretical properties, such as the exactness of the finite element sequences, and the unisolvence of the degrees of freedom., Comment: 35 pages, 1 figure, 1 table

The Breakthrough Listen Initiative is conducting a program using multiple telescopes around the world to search for "technosignatures": artificial transmitters of extraterrestrial origin from beyond our solar system. The VERITAS Collaboration joined this program in 2018, and provides the capability to search for one particular technosignature: optical pulses of a few nanoseconds duration detectable over interstellar distances. We report here on the analysis and results of dedicated VERITAS observations of Breakthrough Listen targets conducted in 2019 and 2020 and of archival VERITAS data collected since 2012. Thirty hours of dedicated observations of 136 targets and 249 archival observations of 140 targets were analyzed and did not reveal any signals consistent with a technosignature. The results are used to place limits on the fraction of stars hosting transmitting civilizations. We also discuss the minimum-pulse sensitivity of our observations and present VERITAS observations of CALIOP: a space-based pulsed laser onboard the CALIPSO satellite. The detection of these pulses with VERITAS, using the analysis techniques developed for our technosignature search, allows a test of our analysis efficiency and serves as an important proof-of-principle., Comment: 15 pages, 7 figures

We report the detection of very high energy gamma-ray emission from the blazar S3 1227+25 (VER J1230+253) with the Very Energetic Radiation Imaging Telescope Array System (VERITAS). VERITAS observations of the source were triggered by the detection of a hard-spectrum GeV flare on May 15, 2015 with the Fermi-Large Area Telescope (LAT). A combined five-hour VERITAS exposure on May 16th and May 18th resulted in a strong 13$\sigma$ detection with a differential photon spectral index, $\Gamma$ = 3.8 $\pm$ 0.4, and a flux level at 9% of the Crab Nebula above 120 GeV. This also triggered target of opportunity observations with Swift, optical photometry, polarimetry and radio measurements, also presented in this work, in addition to the VERITAS and Fermi-LAT data. A temporal analysis of the gamma-ray flux during this period finds evidence of a shortest variability timescale of $\tau_{obs}$ = 6.2 $\pm$ 0.9 hours, indicating emission from compact regions within the jet, and the combined gamma-ray spectrum shows no strong evidence of a spectral cut-off. An investigation into correlations between the multiwavelength observations found evidence of optical and gamma-ray correlations, suggesting a single-zone model of emission. Finally, the multiwavelength spectral energy distribution is well described by a simple one-zone leptonic synchrotron self-Compton radiation model., Comment: 18 pages, 6 figures. Accepted for publication in the Astrophysical Journal (ApJ)

AT 2021loi is an optical-ultraviolet transient located at the center of its host galaxy. Its spectral features identify it as a member of the ``Bowen Fluorescence Flare'' (BFF) class. The first member of this class was considered to be related to a tidal disruption event, but enhanced accretion onto an already active supermassive black hole was suggested as an alternative explanation. AT 2021loi, having occurred in a previously-known unobscured AGN, strengthens the latter interpretation. Its light curve is similar to those of previous BFFs, showing a rebrightening approximately one year after the main peak (which was not explicitly identified, but might be the case, in all previous BFFs). An emission feature around 4680 A, seen in the pre-flare spectrum, strengthens by a factor of $\sim$2 around the optical peak of the flare, and is clearly seen as a double peaked feature then, suggesting a blend of NIII $\lambda 4640$ with HeII $\lambda4686$ as its origin. The appearance of OIII $\lambda$3133 and possible NIII $\lambda\lambda4097,4103$ (blended with H$\delta$) during the flare further support a Bowen Fluorescence classification. Here, we present ZTF, ATLAS, Keck, Las Cumbres Observatory, NEOWISE-R, $Swift$, AMI and VLA observations of AT 2021loi, making it one of the best observed BFFs to date. AT 2021loi thus provides some clarity on the nature of BFFs but also further demonstrates the diversity of nuclear transients., Comment: Submitted to ApJ. This version addresses comments from the referee