Search for: All records

Out-of-distribution (OOD) detection plays a crucial role in ensuring the safe deployment of deep neural network (DNN) classifiers. While a myriad of methods have focused on improving the performance of OOD detectors, a critical gap remains in interpreting their decisions. We help bridge this gap by providing explanations for OOD detectors based on learned high-level concepts. We first propose two new metrics for assessing the effectiveness of a particular set of concepts for explaining OOD detectors: 1) detection completeness, which quantifies the sufficiency of concepts for explaining an OOD-detector’s decisions, and 2) concept separability, which captures the distributional separation between in-distribution and OOD data in the concept space. Based on these metrics, we propose an unsupervised framework for learning a set of concepts that satisfy the desired properties of high detection completeness and concept separability, and demonstrate its effectiveness in providing concept-based explanations for diverse off-the-shelf OOD detectors. We also show how to identify prominent concepts contributing to the detection results, and provide further reasoning about their decisions. 

We present photometric and spectroscopic data for SN 2022joj, a nearby peculiar Type Ia supernova (SN Ia) with a fast decline rate (Δm_15,B= 1.4 mag). SN 2022joj shows exceedingly red colors, with a value of approximatelyB−V≈ 1.1 mag during its initial stages, beginning from 11 days before maximum brightness. As it evolves, the flux shifts toward the blue end of the spectrum, approachingB−V≈ 0 mag around maximum light. Furthermore, at maximum light and beyond, the photometry is consistent with that of typical SNe Ia. This unusual behavior extends to its spectral characteristics, which initially displayed a red spectrum and later evolved to exhibit greater consistency with typical SNe Ia. Spectroscopically, we find strong agreement between SN 2022joj and double detonation models with white dwarf masses of around 1M_⊙and a thin He shell between 0.01 and 0.05M_⊙. Moreover, the early red colors are explained by line-blanketing absorption from iron peak elements created by the double detonation scenario in similar mass ranges. The nebular spectra in SN 2022joj deviate from expectations for double detonation, as we observe strong [Feiii] emission instead of [Caii] lines as anticipated, though this is not as robust a prediction as early red colors and spectra. The fact that as He shells get thinner these SNe start to look more like normal SNe Ia raises the possibility that this is the triggering mechanism for the majority of SNe Ia, though evidence would be missed if the SNe are not observed early enough.

 

Human activities are rapidly changing ecosystems around the world. These changes have widespread implications for the preservation of biodiversity, agricultural productivity, prevalence of zoonotic diseases, and sociopolitical conflict. To understand and improve the predictive capacity for these and other biological phenomena, some scientists are now relying on observatory networks, which are often composed of systems of sensors, teams of field researchers, and databases of abiotic and biotic measurements across multiple temporal and spatial scales. One well-known example is NEON, the US-based National Ecological Observatory Network. Although NEON and similar networks have informed studies of population, community, and ecosystem ecology for years, they have been minimally used by organismal biologists. NEON provides organismal biologists, in particular those interested in NEON's focal taxa, with an unprecedented opportunity to study phenomena such as range expansions, disease epidemics, invasive species colonization, macrophysiology, and other biological processes that fundamentally involve organismal variation. Here, we use NEON as an exemplar of the promise of observatory networks for understanding the causes and consequences of morphological, behavioral, molecular, and physiological variation among individual organisms.

 

This paper presents an approach to detect out-of-context (OOC) objects in an image. Given an image with a set of objects, our goal is to determine if an object is inconsistent with the scene context and detect the OOC object with a bounding box. In this work, we consider commonly explored contextual relations such as co-occurrence relations, the relative size of an object with respect to other objects, and the position of the object in the scene. We posit that contextual cues are useful to determine object labels for in-context objects and inconsistent context cues are detrimental to determining object labels for out-of-context objects. To realize this hypothesis, we propose a graph contextual reasoning network (GCRN) to detect OOC objects. GCRN consists of two separate graphs to predict object labels based on the contextual cues in the image: 1) a representation graph to learn object features based on the neighboring objects and 2) a context graph to explicitly capture contextual cues from the neighboring objects. GCRN explicitly captures the contextual cues to improve the detection of in-context objects and identify objects that violate contextual relations. In order to evaluate our approach, we create a large-scale dataset by adding OOC object instances to the COCO images. We also evaluate on recent OCD benchmark. Our results show that GCRN outperforms competitive baselines in detecting OOC objects and correctly detecting in-context objects. 

Normalizing flows map an independent set of latent variables to their samples using a bijective transformation. Despite the exact correspondence between samples and latent variables, their high level relationship is not well understood. In this paper we characterize the geometric structure of flows using principal manifolds and understand the relationship between latent variables and samples using contours. We introduce a novel class of normalizing flows, called principal component flows (PCF), whose contours are its principal manifolds, and a variant for injective flows (iPCF) that is more efficient to train than regular injective flows. PCFs can be constructed using any flow architecture, are trained with a regularized maximum likelihood objective and can perform density estimation on all of their principal manifolds. In our experiments we show that PCFs and iPCFs are able to learn the principal manifolds over a variety of datasets. Additionally, we show that PCFs can perform density estimation on data that lie on a manifold with variable dimensionality, which is not possible with existing normalizing flows. 

We present the Keck Infrared Transient Survey, a NASA Key Strategic Mission Support program to obtain near-infrared (NIR) spectra of astrophysical transients of all types, and its first data release, consisting of 105 NIR spectra of 50 transients. Such a data set is essential as we enter a new era of IR astronomy with the James Webb Space Telescope (JWST) and the upcoming Nancy Grace Roman Space Telescope (Roman). NIR spectral templates will be essential to search JWST images for stellar explosions of the first stars and to plan an effective Roman SN Ia cosmology survey, both key science objectives for mission success. Between 2022 February and 2023 July, we systematically obtained 274 NIR spectra of 146 astronomical transients, representing a significant increase in the number of available NIR spectra in the literature. Here, we describe the first release of data from the 2022A semester. We systematically observed three samples: a flux-limited sample that includes all transients <17 mag in a red optical band (usually ZTFror ATLASobands); a volume-limited sample including all transients within redshiftz< 0.01 (D≈ 50 Mpc); and an SN Ia sample targeting objects at phases and light-curve parameters that had scant existing NIR data in the literature. The flux-limited sample is 39% complete (60% excluding SNe Ia), while the volume-limited sample is 54% complete and is 79% complete toz= 0.005. Transient classes observed include common Type Ia and core-collapse supernovae, tidal disruption events, luminous red novae, and the newly categorized hydrogen-free/helium-poor interacting Type Icn supernovae. We describe our observing procedures and data reduction usingPypeIt, which requires minimal human interaction to ensure reproducibility.