Pith. sign in

REVIEW 1 major objections 7 minor 85 references

Reviewed by Pith at T0; open to challenge.

T0 means a machine referee read the full paper against a public rubric. The mark states how deep the mechanical check went, never who wrote it. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · glm-5.2

When AI confidently mislabels a new star, two kinds of doubt catch it

2026-07-10 01:03 UTC pith:FH6PAQ6Z

load-bearing objection The cross-family the 1 major comments →

arxiv 2607.08044 v1 pith:FH6PAQ6Z submitted 2026-07-09 astro-ph.IM astro-ph.HE

Catching Disguised Transients with ASTRANet: Anomaly-Aware Spectroscopic Classification and Conformal Calibration

Argyro Sasli , Maojie Xu , Alexandra Junell , Hailey Markoff , Avyukt Raghuvanshi , Felipe F. Nunes , Theophile Jegou Du Laz , Jesper Sollerman
show 16 more authors
Christoffer Fremling Drew Oldag Antoine Le Calloch Sushant Sharma Chaudhary Sneha Maharjan Maxine West Benny Border Nabeel Rehemtulla Richard Dekany Joahan Castaneda Jaimes Russ R. Laher Reed Riddle Mansi M. Kasliwal Matthew J. Graham Ashish A. Mahabal Michael W. Coughlin
This is my paper
classification astro-ph.IM astro-ph.HE
keywords spectroscopictexttttransientsanomalyconformalastranetastranet-cpclass
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Deep classifiers for astronomical spectra assign every input to a known class with high confidence, even when the input belongs to no known class. This paper identifies five physically distinct ways these classifiers fail on rare transients, from blazars disguised as dwarf-star outbursts to genuinely novel explosions the classifier has never seen. The central claim is that no single anomaly score can detect all five failure modes: roughly 43 percent of rare anomalies produce weak signals in every individual detection method, and only a non-linear combination of sixteen complementary scores spanning four families recovers them. The paper further argues that classifier-internal uncertainty (the model's own self-doubt) and embedding-based anomaly detection (how far the spectrum sits from known classes in a learned representation space) are structurally complementary axes that catch largely disjoint failure populations, and that both must be deployed simultaneously rather than chosen between.

Core claim

The paper's central object is the failure-mode taxonomy: five physically distinct regimes in which closed-set spectroscopic classifiers confidently misclassify rare transients. Two of these regimes, comprising about 43 percent of the rare-anomaly population, are recoverable only through non-linear cross-family combination of multiple anomaly scores, because no single score family fires above threshold on those spectra. This finding directly motivates the architecture: a gradient-boosted combiner integrating sixteen scores from four families (uncertainty, distance, density, hybrid) into a single anomaly probability, supplemented by a conformal calibration layer that provides distribution-free

What carries the argument

A hierarchical spectral classifier operating on observer-frame spectra without requiring host-galaxy redshift or spectral phase; sixteen embedding-space anomaly scores spanning four physically motivated families (uncertainty, distance, density, hybrid); a gradient-boosted non-linear combiner producing a single cross-family anomaly score; a conformal prediction layer with Mondrian calibration providing per-spectrum p-values and prediction sets; and an AD-stratified Mondrian variant that partitions the calibration set by both predicted class and anomaly-score quintile to achieve uniform conditional coverage across anomaly strata.

Load-bearing premise

The 289-spectrum rare-anomaly evaluation set, with some classes having as few as one or ten members, is assumed to represent the full physical diversity of rare transients. If the five failure-mode taxonomy is incomplete or the sample is biased toward particular astrophysical configurations, the necessity of cross-family combination may not generalize to rare classes not represented in this set.

What would settle it

A rare-anomaly class not represented in the evaluation set whose spectra are confidently misclassified by the classifier, produce weak signals in all sixteen anomaly scores including the non-linear combiner, and are structurally undetectable by the framework. Such a class would demonstrate that the five-regime taxonomy is incomplete and that cross-family combination is not sufficient for all failure modes.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Survey pipelines deploying only softmax confidence or only Mahalanobis distance silently lose roughly half of rare-anomaly populations by construction, since cross-family combination is necessary for the disguised-class regimes.
  • The AD-stratified conformal calibration approach is generalizable: any anomaly score used as a stratification variable for Mondrian conformal prediction can recover uniform conditional coverage in the high-anomaly operational regime where marginal methods under-cover.
  • The finding that classifier-internal UQ and embedding-based AD are structurally orthogonal suggests a three-way operational policy for survey triage: commit when both are low, abstain when UQ is high but AD is low, and alert when AD is high regardless of UQ.
  • At LSST scale, every one percent improvement in confidence-aware triage corresponds to roughly one hundred peculiar transients per year that would otherwise be lost to routine classification.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 7 minor

Summary. This manuscript presents ASTRANet, a confidence-aware framework for spectroscopic transient classification comprising three modules: a hierarchical classifier operating on observer-frame spectra without requiring redshift or phase, an anomaly detection layer (ASTRANet-Sentinel) combining 16 embedding-space scores via a gradient-boosted model, and a conformal uncertainty quantification layer (ASTRANet-CP) with a novel AD-stratified Mondrian conformal prediction scheme. The framework is validated on 289 rare/out-of-taxonomy spectra spanning 11 excluded classes. The central claims are: (1) closed-set classifier failure modes separate into five physically distinct regimes, (2) cross-family non-linear combination of anomaly scores is necessary to recover all five regimes, and (3) classifier-internal uncertainty and embedding-based anomaly detection are structurally complementary axes. The conformal coverage properties are empirically validated, and AD-MCP achieves uniform conditional coverage where vanilla Mondrian under-covers.

Significance. The work addresses a genuine and timely problem: spectroscopic triage at the scale of ZTF and LSST requires automated, calibrated confidence assessment, not just point-prediction classification. The redshift-free, phase-free observer-frame architecture is a practical advance for real-time deployment. The failure-mode taxonomy (Sec. II) is a useful conceptual contribution that connects physical regimes to specific anomaly-score signatures. The AD-MCP extension of atypicality-stratified conformal prediction to a learned cross-family ensemble is methodologically novel for astronomical spectroscopy. The leakage-free k-fold protocol (Sec. VD) and the empirical validation of conformal coverage within finite-sample error (Table IV, IX) are commendable. The structural complementarity argument between classifier-internal UQ and cross-family AD (Sec. VIF, Table VII) is well-supported by the population-scale analysis. The framework is being prepared for production integration with the ZTF BOOM broker, which lends practical credibility.

major comments (1)
  1. Sec. VC, Table I, and the claim that cross-family non-linear combination is 'necessary' (Contributions item 2, Sec. II.f): The key ablation comparing family-only mini-ensembles (86/289 at 1% FAR, no outlier exposure) against the full pipeline (238/289, with outlier exposure) confounds two variables: cross-family combination and outlier-exposure (OE) fine-tuning. OE reshapes the encoder embeddings that all 16 scores depend on (Sec. IVD), so the 137/238 spectra attributed to cross-family interaction may be partly or wholly recoverable by OE-enhanced single-family scores. The non-OE comparison (86 vs. 99, Table I) shows cross-family combination adds only 13 spectra over the best single family without OE — a modest gain that does not support the strong necessity claim. To establish that cross-family combination (rather than OE) drives the improvement, the authors should run OE-enhancedfamily
minor comments (7)
  1. Sec. IIIb: The rare-anomaly evaluation set includes classes with N<=10 (Ca-rich N=10, BL Lac N=1, Classical Nova N=8). Per-class recovery statistics for these classes (Table III) have large uncertainties. The authors acknowledge this in Sec. VIII (Limitations) but the headline recovery rates (e.g., '40.0%' for Ca-rich) are presented without error bars or confidence intervals. Adding bootstrap intervals for the smallest classes would set appropriate expectations.
  2. Table X caption: The regimes are labeled A/B/C in the table but I/V in the text (Sec. II, Sec. VII). This mismatch requires cross-referencing and could be unified.
  3. Sec. VIA: The statement 'the LightGBM-OOF score satisfies the exchangeability assumption conformal prediction requires' is an empirical observation, not a proof. The p-value ECDF being consistent with uniform (Fig. 4) is reassuring but does not formally guarantee exchangeability under arbitrary distribution shift. The authors should soften this language.
  4. Appendix C, feature importances: The top features (cosine, energy, mahal_global, mahal_within, mahal_class, gmm) span three of four families, which supports the cross-family design. However, the two lowest-ranked features (pca_recon, mrs) are from the density and hybrid families. If these contribute little, a 14-score combiner might perform comparably. An ablation over score count would strengthen the design justification.
  5. Sec. IVA: The claim that ASTRANet is 'the first fine-grained, multi-class classifier' to operate without redshift on observer-frame spectra should note SNIascore (Ref. 14) as a binary precedent, which it does, but the novelty claim would benefit from explicit comparison to any other redshift-free multi-class attempts if they exist.
  6. Table XIV: The 'Category' column uses 'SN CC (H)' and 'SN CC (SE)' but the coarse head (Sec. IVA) uses 'SN-CC' as a single superclass. The mapping between the three-way coarse partition and the finer category labels in Table XIV should be clarified.
  7. Sec. VIH: The AD-MCP fallback for cells with n_min < 10 uses the marginal class threshold. With 1,669 calibration samples and Q=5, this activates for sparse cells. The authors note this affects only rare-class extreme-quintile cells, but the fraction of test samples falling into fallback cells is not reported. This number would help assess the practical impact.

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained against external benchmarks and theory.

full rationale

The paper's derivation chain contains no circular steps. The conformal prediction guarantees (Eqs. 1-2, 6-7) rest on established distribution-free theory from Vovk et al. [67], Angelopoulos & Bates [68], and Romano et al. [69] — all external citations with independently verified theorems. The AD-MCP extension (Sec. VIH) applies this theory to a new stratification scheme; its conditional coverage guarantee (Eq. 7) follows from within-cell exchangeability, a standard conformal assumption, not from any result defined in this paper. The cross-family combination claim (Sec. VC) is empirically validated on held-out data under a leakage-free k-fold protocol — the LightGBM combiner is trained on validation data and evaluated on held-out folds, so no detection reduces to its own training input. The structural complementarity claim (Sec. VIF, Table VII) is an empirical observation on held-out test data, not a derived result. The five failure-mode taxonomy (Sec. II) is descriptive. Self-citations to companion works [16, 27, 28, 37] provide context, baselines, and infrastructure but are not load-bearing for any mathematical or methodological claim. The skeptic's concern that OE is confounded with cross-family combination in the ablation (Table I) is a validity concern, not circularity: the paper does not define a quantity in terms of what it claims to predict. Score 1 reflects the presence of non-load-bearing self-citations to companion works.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physical entities or postulated particles. The 'invented' components are architectural choices (FiLM conditioning, hierarchical heads, 16-score combiner) rather than new physical postulates. Free parameters are standard ML hyperparameters and loss weights. The axioms are domain assumptions about data exchangeability and sample representativeness, standard for conformal prediction and ML evaluation.

free parameters (5)
  • Loss weights (w_f, w_c, w_sc,f, w_sc,c, w_conf, w_H) = 1.5, 0.3, 0.03, 0.05, 0.5, 0.01
    Hand-set loss term weights (Table XI) chosen to balance classification and contrastive objectives.
  • LightGBM hyperparameters = 300 estimators, eta=0.05, num_leaves=31
    Chosen by 5-fold CV on validation set for the anomaly score combiner.
  • AD-MCP quintile bin edges = 5 equal-frequency bins on calibration set
    Choice of Q=5 quintiles and n_min=10 fallback threshold are design decisions affecting conditional coverage granularity.
  • Outlier exposure margins (lambda_OE, lambda_emb) = 1.0, 1.0
    Margins for the outlier-exposure fine-tuning objective, stated as fixed values.
  • Confusion penalty pairs and weights = (TDE->Ia, 0.5), (SLSN->Ibc, 0.3), (II->Ibc, 0.8)
    Astrophysically motivated targeted misclassification penalties, hand-selected.
axioms (4)
  • domain assumption Exchangeability of calibration and test spectra
    Conformal prediction validity (Sec. VI) requires that calibration data and new test data are exchangeable. Empirically checked on ZTF held-out set but not guaranteed under LSST distribution shift.
  • domain assumption The 289-spectrum rare set spans the full physical diversity of rare anomalies
    Sec. IIIb states the set is 'constructed to span the five failure-mode regimes.' The taxonomy completeness depends on this.
  • standard math Supervised contrastive loss produces a metrically faithful embedding space
    The AD layer operates on the 192-d embedding assuming distance correlates with typicality. Appendix H provides empirical support (kNN accuracy = classifier accuracy).
  • domain assumption Within-cell exchangeability holds for AD-MCP
    AD-MCP conditional coverage guarantee (Eq. 7) requires exchangeability within each (class, AD-quintile) cell. Not independently verified beyond empirical coverage tracking.

pith-pipeline@v1.1.0-glm · 47122 in / 2704 out tokens · 364053 ms · 2026-07-10T01:03:25.147737+00:00 · methodology

0 comments
read the original abstract

Time-domain surveys discover thousands of transients per year, but the spectroscopic identification of rare and physically peculiar objects remains rate-limited by closed-set classifiers that confidently assign every input to a known class -- including spectra that genuinely belong to no known class. We present the \texttt{ASTRANet} framework, a confidence-aware infrastructure for spectroscopic transient classification built around three coupled modules: a hierarchical spectral classifier that operates directly on observer-frame spectra without requiring host-galaxy redshift or spectral phase as inputs; an anomaly detection layer (\texttt{ASTRANet-Sentinel}) that non-linearly combines $16$ embedding-space anomaly scores spanning four physically motivated families; and a conformal uncertainty quantification layer (\texttt{ASTRANet-CP}). We validate the framework on a held-out evaluation set of $289$ rare and out-of-taxonomy transients spanning $11$ classes deliberately excluded from training, chosen to span the full physical diversity of the rare-anomaly population: AGN-related outliers, GRB-related events, gap transients, novae, and peculiar supernovae. Through five astrophysically distinct failure modes of closed-set classifiers, we show that classifier-internal uncertainty and embedding-based anomaly detection are structurally complementary axes of confidence rather than alternative implementations of the same estimator. We further introduce AD-stratified Mondrian conformal prediction (AD-MCP) within \texttt{ASTRANet-CP}, achieving uniform conditional coverage across anomaly-score strata where vanilla Mondrian under-covers in the operational regime. This establishes the methodological infrastructure for confidence-aware spectroscopic discovery in the Vera C.\ Rubin Observatory era.

Figures

Figures reproduced from arXiv: 2607.08044 by Alexandra Junell, Antoine Le Calloch, Argyro Sasli, Ashish A. Mahabal, Avyukt Raghuvanshi, Benny Border, Christoffer Fremling, Drew Oldag, Felipe F. Nunes, Hailey Markoff, Jesper Sollerman, Joahan Castaneda Jaimes, Mansi M. Kasliwal, Maojie Xu, Matthew J. Graham, Maxine West, Michael W. Coughlin, Nabeel Rehemtulla, Reed Riddle, Richard Dekany, Russ R. Laher, Sneha Maharjan, Sushant Sharma Chaudhary, Theophile Jegou Du Laz.

Figure 1
Figure 1. Figure 1: FIG. 1. Architecture of [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Confusion matrix on the held-out test set ( [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Per-anomaly detection map at [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Mondrian conformal p-value distribution on the [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Embedding-space nearest-neighbor retrieval, both [PITH_FULL_IMAGE:figures/full_fig_p027_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

85 extracted references · 85 canonical work pages · 37 internal anchors

  1. [1]

    Stochastic depth [79] with linearly increasing drop probability0→0.1provides implicit ensemble reg- ularisation

    A residual skip connection adds the block input to its output. Stochastic depth [79] with linearly increasing drop probability0→0.1provides implicit ensemble reg- ularisation. Cumulative receptive field:∼250samples. c. Metadata conditioning via FiLM.The metadata vectorc∈R 64 encodes three quantities: spectral phase ϕ(rest-frame days from maximum light, re...

    work page

  2. [2]

    Included Subtypes

    with learning rateη0 = 3×10 −4, weight decay10−4, gradient clip1.0. The schedule uses a5-epoch linear warmup followed by cosine annealing with warm restarts [80],T 0 = 30,T mult = 2,η min = 0.05η 0. EMA [81] of model weights is maintained with decayα= 0.996and used for all validation/test evaluations. Early stopping usesvalidationmacrorecallwithpatience45...

    work page

  3. [3]

    find more like this

    over the test-set embeddings reaches86.8%five-fold accuracy – statistically indistinguishable from the full hi- erarchical classifier head (86.6%, Sec. VA). The class structure visible in the UMAP projection is therefore metrically faithful, not a visualization artifact. The same retrieval operates in two complementary regimes (Fig. 6). (ii) In-distributi...

    work page

  4. [4]

    E. C. Bellm, S. R. Kulkarni, M. J. Graham, R. Dekany, R. M. Smith, R. Riddle, F. J. Masci, G. Helou, T. A. Prince, S. M. Adams, C. Barbarino, T. Barlow, J. Bauer, R. Beck, J. Belicki, R. Biswas, N. Blagorodnova, D. Bodewits, B. Bolin, V. Brinnel, T. Brooke, B. Bue, M. Bulla, R. Burruss, S. B. Cenko, C.-K. Chang, A. Con- nolly, M. Coughlin, J. Cromer, V. C...

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  5. [5]

    F. J. Masci, R. R. Laher, B. Rusholme, D. L. Shupe, S. Groom, J. Surace, E. Jackson, S. Monkewitz, R. Beck, D. Flynn, S. Terek, W. Landry, E. Hacopians, V. Desai, J. Howell, T. Brooke, D. Imel, S. Wachter, Q.-Z. Ye, H.- W. Lin, S. B. Cenko, V. Cunningham, U. Rebbapragada, B.Bue, A.A.Miller, A.Mahabal, E.C.Bellm, M.T.Pat- terson, M. Jurić, V. Z. Golkhou, E...

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  6. [6]

    M. J. Graham, S. R. Kulkarni, E. C. Bellm, S. M. Adams, C. Barbarino, N. Blagorodnova, D. Bodewits, B. Bolin, P. R. Brady, S. B. Cenko, C.-K. Chang, M. W. Coughlin, K. De, G. Eadie, T. L. Farnham, U. Feindt, A. Franck- owiak, C. Fremling, S. Gezari, S. Ghosh, D. A. Gold- stein, V. Z. Golkhou, A. Goobar, A. Y. Q. Ho, D. Hup- penkothen, Ž. Ivezić, R. L. Jon...

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  7. [7]

    M. T. Pattersonet al., PASP131, 018001 (2019)

    work page 2019

  8. [8]

    Dekany, R

    R. Dekany, R. M. Smith, R. Riddle, M. Feeney, M. Porter, D. Hale, J. Zolkower, J. Belicki, S. Kaye, J. Henning, R. Walters, J. Cromer, A. Delacroix, H. Ro- driguez, D. J. Reiley, P. Mao, D. Hover, P. Murphy, R. Burruss, J. Baker, M. Kowalski, K. Reif, P. Mueller, E. Bellm, M. Graham, and S. R. Kulkarni, PASP132, 038001 (2020), arXiv:2008.04923 [astro-ph.IM]

    work page arXiv 2020

  9. [9]

    LSST: from Science Drivers to Reference Design and Anticipated Data Products

    Ž. Ivezić, S. M. Kahn,et al., The Astrophysical Journal 873, 111 (2019), arXiv:0805.2366

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  10. [10]

    Blagorodnova, J

    N. Blagorodnova, J. D. Neill, R. Walters, S. R. Kulkarni, C. Fremling, S. Ben-Ami, R. G. Dekany, J. R. Fucik, N. Konidaris, R. Nash, C.-C. Ngeow, E. O. Ofek, D. O. Sullivan, R. Quimby, A. Ritter, and K. E. Vyhmeister, PASP130, 035003 (2018)

    work page 2018

  11. [11]

    A Fully Automated Integral Field Spectrograph Pipeline for the SEDMachine: pysedm

    M. Rigault, J. D. Neill, N. Blagorodnova, A. Dugas, M. Feeney, R. Walters, V. Brinnel, Y. Copin, C. Frem- ling, J. Nordin, and J. Sollerman, A&A627, A115 (2019), arXiv:1902.08526 [astro-ph.IM]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  12. [12]

    R. J. Foley, P. J. Challis, R. Chornock,et al., The Astro- physical Journal767, 57 (2013), arXiv:1212.2209

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  13. [13]

    H. B. Perets, A. Gal-Yam, P. A. Mazzali, D. Arnett, D. Kagan, A. V. Filippenko, W. Li, I. Arcavi, S. B. Cenko, D. B. Fox,et al., Nature465, 322 (2010), arXiv:0906.2003 [astro-ph.HE]

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  14. [14]

    M. R. Drout, R. Chornock, A. M. Soderberg, N. E. Sanders, R. McKinnon, A. Rest, R. J. Foley, D. Milisavl- jevic, R. Margutti, E. Berger,et al., The Astrophysical Journal794, 23 (2014), arXiv:1405.3668 [astro-ph.HE]

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  15. [15]

    A. Y. Q. Ho, D. A. Perley, A. Gal-Yam, R. Lunnan, J. Sollerman, S. Schulze, K. K. Das, D. Dobie, Y. Yao, C. Fremling, S. Adams, S. Anand, I. Andreoni, E. C. Bellm, R. J. Bruch, K. B. Burdge, A. J. Castro-Tirado, A. Dahiwale, K. De, R. Dekany, A. J. Drake, D. A. Duev, M. J. Graham, G. Helou, D. L. Kaplan, V. Karambelkar, M. M. Kasliwal, E. C. Kool, S. R. K...

    work page 2023

  16. [16]

    T. A. Thompson, J. L. Prieto, K. Z. Stanek, M. D. Kistler, J. F. Beacom, and C. S. Kochanek, The As- trophysical Journal705, 1364 (2009), arXiv:0809.0510 [astro-ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2009

  17. [17]

    SNIascore: Deep Learning Classification of Low-Resolution Supernova Spectra

    C. Fremling, X. J. Hall, M. W. Coughlin, A. S. Dahiwale, D. A. Duev, M. J. Graham, M. M. Kasliwal, E. C. Kool, A. A. Mahabal, A. A. Miller, J. D. Neill, D. A. Perley, M. Rigault, P. Rosnet, B. Rusholme, Y. Sharma, K. M. Shin, D. L. Shupe, J. Sollerman, and R. S. Walters, ApJL 917, L2 (2021), arXiv:2104.12980 [astro-ph.IM]

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  18. [18]

    Sharma, A

    Y. Sharma, A. A. Mahabal, J. Sollerman, C. Fremling, S. R. Kulkarni, N. Rehemtulla, A. A. Miller, M. Aubert, 29 T. X. Chen, M. W. Coughlin, M. J. Graham, D. Hale, M. M. Kasliwal, Y.-L. Kim, J. D. Neill, J. N. Pur- dum, B. Rusholme, A. Singh, and N. Sravan, PASP137, 034507 (2025)

    work page 2025

  19. [19]

    M. Xu, A. Sasli, A. Junell, F. F. Nunes, Y.-J. Qin, C. Fremling, S. Rose, T. J. D. Laz, B. Border, A. L. Calloch, S. S. Chaudhary, H. Markoff, A. Raghuvan- shi, N. Rehemtulla, J. Sollerman, Y. Sharma, N. Sravan, J. Adler, T. X. Chen, R. Dekany, R. Riddle, M. M. Kasli- wal, M. J. Graham, and M. W. Coughlin, Applecider ii: Spectranet – a deep learning netwo...

    work page arXiv 2025

  20. [20]

    W. F. Fortino, F. B. Bianco, P. Protopapas, D. Muthukr- ishna, and A. Brockmeier, Abc-sn: Attention based classifier for supernova spectra (2025), arXiv:2507.22106 [astro-ph.IM]

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  21. [21]

    Hendrycks and K

    D. Hendrycks and K. Gimpel, ICLR (2017)

    work page 2017

  22. [22]

    C. Guo, G. Pleiss, Y. Sun, and K. Q. Weinberger, in Proceedings of the 34th International Conference on Ma- chine Learning (ICML), Proceedings of Machine Learn- ing Research, Vol. 70 (PMLR, 2017) pp. 1321–1330, arXiv:1706.04599 [cs.LG]

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  23. [23]

    M. Hein, M. Andriushchenko, and J. Bitterwolf, inPro- ceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR)(2019) pp. 41–50, arXiv:1812.05720 [cs.LG]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  24. [24]

    V. A. Villar, M. Cranmer, E. Berger,et al., The As- trophysical Journal Supplement Series255, 24 (2021), arXiv:2103.12102

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  25. [25]

    M. V. Pruzhinskaya, K. L. Malanchev, M. V. Kornilov, et al., Monthly Notices of the Royal Astronomical Society 489, 3591 (2019), arXiv:1905.11516

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  26. [26]

    Chaini, F

    S. Chaini, F. B. Bianco, and A. Mahabal, In search of the unknown unknowns: A multi-metric distance ensemble for out of distribution anomaly detection in astronomical surveys (2025), arXiv:2510.23702 [astro-ph.IM]

    work page arXiv 2025

  27. [27]

    Guptaet al., arXiv e-prints (2025),Update with ac- tual MCIF reference., 2503.00000

    R. Guptaet al., arXiv e-prints (2025),Update with ac- tual MCIF reference., 2503.00000

    work page arXiv 2025

  28. [28]

    Perez, F

    E. Perez, F. Strub, H. de Vries, V. Dumoulin, and A. Courville, inProceedings of the AAAI Conference on Artificial Intelligence, Vol. 32 (2018) p. N/A

    work page 2018

  29. [29]

    Muthukrishna, D

    D. Muthukrishna, D. Parkinson, and B. E. Tucker, ApJ 885, 85 (2019)

    work page 2019

  30. [30]

    T. J. du Laz, M. W. Coughlin, P. Bachant, J. E. Simones, T. Culino, A. L. Calloch, S. S. Chaudhary, X. J. Hall, T. Barna, D. Warshofsky, M. Graham, M. M. Kasliwal, A. Mahabal, J. S. Bloom, A. Palmese, F. J. Masci, S. L. Groom, R. Dekany, R. L. Riddle, and G. Helou, Boom and babamul: a real-time, multi-survey, optical alert bro- ker system operating at sca...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  31. [31]

    Applying multimodal learning to Classify transient Detections Early (AppleCiDEr) I: Data set, methods, and infrastructure

    A. Junell, A. Sasli, F. F. Nunes, M. Xu, B. Border, N. Re- hemtulla, M. Rizhko, Y.-J. Qin, T. J. D. Laz, A. L. Calloch, S. S. Chaudhary, S. Wu, J. Sollerman, N. Sra- van, S. L. Groom, D. Hale, M. M. Kasliwal, J. Purdum, A. Wold, M. J. Graham, and M. W. Coughlin, Apply- ing multimodal learning to classify transient detections early (applecider) i: Data set...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  32. [32]

    Beyond Confidence: Reliable Models Should Also Consider Atypicality

    M. Yuksekgonul, L. Zhang, J. Zou, and C. Guestrin, Be- yond confidence: Reliable models should also consider atypicality (2023), arXiv:2305.18262 [cs.LG]

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  33. [33]

    An embedded X-ray source shines through the aspherical AT2018cow: revealing the inner workings of the most luminous fast-evolving optical transients

    R. Margutti, B. D. Metzger, R. Chornock, I. Vurm, N. Roth, B. W. Grefenstette, V. Savchenko, R. Cartier, J. F. Steiner, G. Terreran,et al., The Astrophysical Jour- nal872, 18 (2019), arXiv:1810.10720 [astro-ph.HE]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  34. [34]

    D. A. Perley, P. A. Mazzali, L. Yan, S. B. Cenko, S. Gezari, K. Taggart, N. Blagorodnova, C. Frem- ling, B. Mockler, A. Singh,et al., Monthly Notices of the Royal Astronomical Society484, 1031 (2019), arXiv:1808.00969 [astro-ph.HE]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  35. [35]

    C. S. Kochanek, D. M. Szczygieł, and K. Z. Stanek, The Astrophysical Journal758, 142 (2012), arXiv:1202.0281 [astro-ph.SR]

    work page internal anchor Pith review Pith/arXiv arXiv 2012

  36. [36]

    Barna, C

    T. Barna, C. Fremling, T. Ahumada, I. Andreoni, S. Banerjee, J. S. Bloom, M. Bulla, T. X. Chen, M. W. Coughlin, T.Dietrich, X.J.Hall, A.Junell, B.Rusholme, J. Sollerman, and N. Sravan, Publications of the Astro- nomical Society of the Pacific137, 084105 (2025)

    work page 2025

  37. [37]

    Fremling, D

    C. Fremling, D. Perley, A. Y. Q. Ho, and ZTF Collabo- ration, GRB Coordinates Network31629, 1 (2022)

    work page 2022

  38. [38]

    S. J. van der Walt, A. Crellin-Quick, and J. S. Bloom, Journal of Open Source Software4, 1247 (2019)

    work page 2019

  39. [39]

    M. W. Coughlin, J. S. Bloom, G. Nir, S. Antier, T. J. du Laz, S. van der Walt, A. Crellin-Quick, T. Culino, D. A. Duev, D. A. Goldstein, B. F. Healy, V. Karam- belkar, J. Lilleboe, K. M. Shin, L. P. Singer, T. Ahu- mada, S. Anand, E. C. Bellm, R. Dekany, M. J. Graham, M. M. Kasliwal, I. Kostadinova, R. W. Kiendrebeogo, S. R. Kulkarni, S. Jenkins, N. LeBar...

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  40. [40]

    Sasli, IRIS: Identification and Reduction of Interest- ing Spectra,https://github.com/applecider-ml/IRIS (2026), spectral preprocessing pipeline for transient clas- sification

    A. Sasli, IRIS: Identification and Reduction of Interest- ing Spectra,https://github.com/applecider-ml/IRIS (2026), spectral preprocessing pipeline for transient clas- sification

    work page 2026

  41. [41]

    The Zwicky Transient Facility Bright Transient Survey I: Spectroscopic Classification and the Redshift Completeness of Local Galaxy Catalogs

    C. Fremling, A. A. Miller, Y. Sharma, A. Dugas, D. A. Perley, K. Taggart, J. Sollerman, A. Goobar, M. L. Gra- ham, J. D. Neill, J. Nordin, M. Rigault, R. Walters, I. Andreoni, A. Bagdasaryan, J. Belicki, C. Cannella, E. C. Bellm, S. B. Cenko, K. De, R. Dekany, S. Fred- erick, V. Z. Golkhou, M. J. Graham, G. Helou, A. Y. Q. Ho, M. M. Kasliwal, T. Kupfer, R...

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  42. [42]

    D. A. Perley, C. Fremling, J. Sollerman, A. A. Miller, A. S. Dahiwale, Y. Sharma, E. C. Bellm, R. Biswas, T. G. Brink, R. J. Bruch, K. De, R. Dekany, A. J. Drake, D. A. Duev, A. V. Filippenko, A. Gal-Yam, A. Goobar, M. J. Graham, M. L. Graham, A. Y. Q. Ho, I. Irani, M. M. Kasliwal, Y.-L. Kim, S. R. Kulkarni, A. Mahabal, F. J. Masci, S. Modak, J. D. Neill,...

    work page 2020

  43. [43]

    The Zwicky Transient Facility Bright Transient Survey. III. $\texttt{BTSbot}$: Automated Identification and Follow-up of Bright Transients with Deep Learning

    N. Rehemtulla, A. A. Miller, T. Jegou Du Laz, M. W. Coughlin, C. Fremling, D. A. Perley, Y.-J. Qin, J. Sollerman, A. A. Mahabal, R. R. Laher, R. Riddle, B. Rusholme, and S. R. Kulkarni, ApJ972, 7 (2024), arXiv:2401.15167 [astro-ph.IM]

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  44. [44]

    The SED Machine: a robotic spectrograph for fast transient classification

    N. Blagorodnova, J. D. Neill, R. Walters, S. R. Kulka- 30 rni, C. Fremling, S. Ben-Ami, R. G. Dekany, J. R. Fu- cik, N. Konidaris, R. Nash, C.-C. Ngeow, E. O. Ofek, D.O’Sullivan, R.Quimby, A.Ritter,andK.E.Vyhmeis- ter, PASP130, 035003 (2018), arXiv:1710.02917 [astro- ph.IM]

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  45. [45]

    Y. L. Kim, M. Rigault, J. D. Neill, M. Briday, Y. Copin, J. Lezmy, N. Nicolas, R. Riddle, Y. Sharma, M. Smith, J. Sollerman, and R. Walters, PASP134, 024505 (2022), arXiv:2203.01346 [astro-ph.IM]

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  46. [46]

    J. B. Oke, J. G. Cohen, M. Carr, J. Cromer, A. Din- gizian, F. H. Harris, S. Labrecque, R. Lucinio, W. Schaal, H. Epps, and J. Miller, Publications of the Astronomical Society of the Pacific107, 375 (1995)

    work page 1995

  47. [47]

    S. M. Faber, A. C. Phillips, R. I. Kibrick, B. Alcott, S. L. Allen, J. Burrous, T. Cantrall, D. Clarke, A. L. Coil, D. J. Cowley, M. Davis, W. T. S. Deich, K. Di- etsch, D. K. Gilmore, C. A. Harper, D. F. Hilyard, J. P. Lewis, M.McVeigh, J.Newman, J.Osborne, R.Schiavon, R. J. Stover, D. Tucker, V. Wallace, M. Wei, G. Wirth, andC.A.Wright,inInstrument Desi...

    work page 2003

  48. [48]

    R. L. Davies, J. R. Allington-Smith, P. Bettess, E. Chad- wick, R. Content, G. N. Dodsworth, R. Haynes, D. Lee, I. J. Lewis, J. Webster, E. Atad, S. M. Beard, M. El- lis, P. R. Hastings, P. R. Williams, T. Bond, D. Cramp- ton, T. J. Davidge, M. Fletcher, B. Leckie, C. L. Mor- bey, R. G. Murowinski, S. Roberts, L. K. Saddlemyer, J. Sebesta, J. R. Stilburn,...

    work page 1997

  49. [49]

    R. W. e. a. Pogge, inSPIE, Vol. 7735 (2010)

    work page 2010

  50. [50]

    M. M. Kasliwal, C. Fremling, L. Yan, K. Das, F. Verdi, J. Zmudzinas, C. Martin, E. Kirby, S. Xue, L. Ho, G. Herczeg, X. Wu, Z. Hu, H. Ji, M. Matuszewski, R. Bertz, D. Hale, H. Rodriguez, A. Boden, R. Dekany, R. Smith, D. Reiley, R. Nash, J. Milburn, D. Neill, J. Brugger, R. Zarzaca, B. Weber, and C. Shapiro, Tran- sient Name Server AstroNote340, 1 (2024)

    work page 2024

  51. [51]

    Accetta, C

    Abdurro’uf, K. Accetta, C. Aerts, V. Silva Aguirre, R. Ahumada, N. Ajgaonkar, N. Filiz Ak, S. Alam, C. Allende Prieto, A. Almeida, F. Anders, S. F. An- derson, B. H. Andrews, B. Anguiano, E. Aquino-Ortíz, A. Aragón-Salamanca, M. Argudo-Fernández, M. Ata, M. Aubert, V. Avila-Reese, C. Badenes, R. H. Barbá, K. Barger, J. K. Barrera-Ballesteros, R. L. Beaton...

    work page arXiv 2022

  52. [52]

    DESI Collaboration, A. G. Adame, J. Aguilar, S. Ahlen, S. Alam, G. Aldering, D. M. Alexander, R. Alfarsy, C. Allende Prieto, M. Alvarez, O. Alves, A. Anand, F. Andrade-Oliveira, E. Armengaud, J. Asorey, S. Avila, A.Aviles, S.Bailey, A.Balaguera-Antolínez, O.Ballester, C. Baltay, A. Bault, J. Bautista, J. Behera, S. F. Beltran, S. BenZvi, L. Beraldo e Silv...

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  53. [53]

    M. M. Kasliwal, C. Cannella, A. Bagdasaryan, T. Hung, U. Feindt, L. P. Singer, M. Coughlin, C. Fremling, R. Walters, D. Duev, R. Itoh, and R. M. Quimby, PASP 131, 038003 (2019)

    work page 2019

  54. [54]

    D. A. Coulter, D. O. Jones, P. McGill, R. J. Foley, P. D. Aleo, M. J. Bustamante-Rosell, D. Chatterjee, K. W. Davis, C. Dickinson, A. Engel, A. Gagliano, W. V. Jacobson-Galán, C. D. Kilpatrick, J. Kutcka, X. K. Le Saux, K. Malanchev, Y. C. Pan, P. J. Quiñonez, C. Rojas-Bravo, M. R. Siebert, K. Taggart, S. Tinyanont, and Q. Wang, PASP135, 064501 (2023), ar...

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  55. [55]

    S. Bai, J. Z. Kolter, and V. Koltun, arXiv preprint arXiv:1803.01271 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  56. [56]

    M. Ilse, J. M. Tomczak, and M. Welling, inInternational Conference on Machine Learning (ICML)(PMLR, 2018) pp. 2127–2136

    work page 2018

  57. [57]

    Khosla, P

    P. Khosla, P. Teterwak, C. Wang, A. Sarna, Y. Tian, P. Isola, A. Maschinot, C. Liu, and D. Krishnan, inAdvances in Neural Information Processing Systems (NeurIPS), Vol. 33 (2020) pp. 18661–18673

    work page 2020

  58. [58]

    T.-Y. Lin, P. Goyal, R. Girshick, K. He, and P. Dollár, inProceedings of the IEEE International Conference on Computer Vision (ICCV)(2017) pp. 2980–2988

    work page 2017

  59. [59]

    Loshchilov and F

    I. Loshchilov and F. Hutter, inInternational Conference on Learning Representations (ICLR)(2019)

    work page 2019

  60. [60]

    Zhang, M

    H. Zhang, M. Cisse, Y. N. Dauphin, and D. Lopez-Paz, inInternational Conference on Learning Representations (ICLR)(2018)

    work page 2018

  61. [61]

    Y. Cui, M. Jia, T.-Y. Lin, Y. Song, and S. Belongie, inProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR)(2019)pp. 9268–9277

    work page 2019

  62. [62]

    Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles

    B. Lakshminarayanan, A. Pritzel, and C. Blundell, in Advances in Neural Information Processing Systems 30 (NIPS 2017)(2017) pp. 6402–6413, arXiv:1612.01474 [stat.ML]

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  63. [63]

    Enhancing The Reliability of Out-of-distribution Image Detection in Neural Networks

    S. Liang, Y. Li, and R. Srikant, Enhancing the reliability of out-of-distribution image detection in neural networks (2020), arXiv:1706.02690 [cs.LG]

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  64. [64]

    W. Liu, X. Wang, J. Owens, and Y. Li, NeurIPS (2020)

    work page 2020

  65. [65]

    Leeet al., NeurIPS (2018)

    K. Leeet al., NeurIPS (2018)

    work page 2018

  66. [66]

    F. T. Liu, K. M. Ting, and Z.-H. Zhou, inProceedings of the 8th IEEE International Conference on Data Mining (ICDM)(IEEE Computer Society, 2008) pp. 413–422

    work page 2008

  67. [67]

    M. M. Breunig, H.-P. Kriegel, R. T. Ng, and J. Sander, inProceedings of the 2000 ACM SIGMOD International Conference on Management of Data(ACM, 2000) pp. 93–104

    work page 2000

  68. [68]

    G. Ke, Q. Meng, T. Finley, T. Wang, W. Chen, W. Ma, Q. Ye, and T.-Y. Liu, inAdvances in Neural Informa- tion Processing Systems 30 (NIPS 2017)(2017)pp.3146– 3154

    work page 2017

  69. [69]

    Deep Anomaly Detection with Outlier Exposure

    D. Hendrycks, M. Mazeika, and T. Dietterich, inInterna- tional Conference on Learning Representations (ICLR) (2019) arXiv:1812.04606

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  70. [70]

    V. Vovk, A. Gammerman, and G. Shafer,Algorith- mic Learning in a Random World(Springer, New York, 2005)

    work page 2005

  71. [71]

    A. N. Angelopoulos and S. Bates, Foundations and Trends in Machine Learning16, 494 (2023)

    work page 2023

  72. [72]

    Classification with Valid and Adaptive Coverage

    Y. Romano, M. Sesia, and E. J. Candès, inAdvances in Neural Information Processing Systems (NeurIPS) (2020) arXiv:2006.02544

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  73. [73]

    M.L.Li, A.Y.Q.Ho, G.Ryan, D.A.Perley, G.P.Lamb, A. J. Nayana, I. Andreoni, G. C. Anupama, E. C. Bellm, E. Berger, J. S. Bloom, E. Burns, I. Caiazzo, P. Chandra, M. W. Coughlin, K. El-Badry, M. J. Graham, M. Kasli- wal, G. K. Keating, S. R. Kulkarni, H. Kumar, F. J. Masci, R.A.Perley, J.Purdum, R.Rao, A.C.Rodriguez, B. Rusholme, N. Sarin, J. Sollerman, G. ...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  74. [74]

    Chaini, A

    S. Chaini, A. Mahabal, A. Kembhavi, and F. Bianco, Astronomy and Computing48, 100850 (2024)

    work page 2024

  75. [75]

    4MOST Consortium Survey 10: The Time-Domain Extragalactic Survey (TiDES)

    E. Swann, M. Sullivan, D. Carollo,et al., The Messenger 175, 58 (2019), arXiv:1903.02476

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  76. [76]

    Frohmaier, M

    C. Frohmaier, M. Sullivan, K. Maguire,et al., inProceed- ings of SPIE, Vol. 13096 (2024) p. 130963X

    work page 2024

  77. [77]

    Schipani, S

    P. Schipani, S. Campana, R. Claudi,et al., inProceedings of SPIE, Vol. 11447 (2020) p. 1144709

    work page 2020

  78. [78]

    P. J. Groot, S. Scaringi, and N. Elias-Rosa, arXiv e-prints , arXiv:2512.17404 (2025), arXiv:2512.17404 [astro-ph.IM]

    work page arXiv 2025

  79. [79]

    Savitzky and M

    A. Savitzky and M. J. E. Golay, Analytical Chemistry 36, 1627 (1964)

    work page 1964

  80. [80]

    Wu and K

    Y. Wu and K. He, inEuropean Conference on Computer Vision (ECCV)(Springer, 2018) pp. 3–19

    work page 2018

Showing first 80 references.