Pith. Sign in

REVIEW

Likelihood-Based Unsupervised Anomaly Detection in CMS Dijet Events

PaperPDF from arXiv

data files

ReviewThe objections and verdict, as JSON Paper breakdownThe authors' claims, conditions and measurements, as JSON Signed receiptProof this review has not been altered
X Bluesky LinkedIn Reddit Hacker News
Bhavishya Chebrolu (1), Hitesh Rasineni (1), Prajwal Aaryan Immadi (1) ((1) VIT-AP University, Amaravati, India, (2) Mohan Babu University, Tirupati, India)

High Energy Physics - Experiment · Submitted 6 Sep 2026 · 36 references

Pith's verdict: Solid methodology, but the central anomaly selection is likely an artifact of low-multiplicity jets with zero substructure — a load-bearing flaw not ruled out by existing tests.

T0 · AI review

How this review was made: An AI referee (deepseek/deepseek-v4-flash) read the paper on September 9, 2026 and answered a fixed list of questions about it. T0 is the starting level every paper gets. The levels describe how deeply Pith has checked the paper, never the authors.

  1. T0 AI review An AI read the paper and answered a fixed list of questions.
  2. T1 Audit Automatic checks that the references and files the paper relies on exist and match.
  3. T2 Compute The paper's code or calculations re-run with the same results.
  4. T3 Formal statement The claims written out in a formal math language a computer can read.
  5. T4 Proof Those claims proved, and the proof checked by computer.

How Pith checks papers →

· 3 major objections · 6 minor · Ask about this paper · The compiled record · Read the full review →
Challenge this review Comment (0) Read on arXiv ↗
Cite ▾
arXiv ID2609.06686 (version 1)
Pith ID4BZ7O2UI, Pith's permanent record of this paper and its review
Citation(2) Mohan Babu University, Amaravati, Bhavishya Chebrolu (1), Hitesh Rasineni (1), India, India), Prajwal Aaryan Immadi (1) ((1) VIT-AP University, Tirupati. "Likelihood-Based Unsupervised Anomaly Detection in CMS Dijet Events." arXiv:2609.06686 (2026). https://pith.science/paper/4BZ7O2UI
Cite this reviewBibTeX for Pith's review of this paper
Claim paper

Why Pith says this

Highlight any text to comment on it.

Referee report

How Pith checks papers → · Full audit of this paper →

The paper

The paper claims

A neural spline flow trained on 61 dijet observables flags extreme events whose jet substructure deviates coherently from the QCD background.

The reading

This paper aims to establish that neural spline flows can serve as a fully unsupervised, model-independent anomaly detection tool on real LHC data. The authors train a flow on 61 jet, dijet and event-level observables from 125 GB of CMS Run2011B dijet data, learning the QCD background density without any simulated signal. They select the 0.1% of events with the lowest likelihood and find that these events exhibit coherent, systematic deviations in jet-substructure observables (girth, N-subjettiness, energy correlation ratios) rather than isolated kinematic outliers. Extensive validation tests—mass decorrelation, permutation nulls, training stability—show that the deviations are not artifacts of the selection procedure. The paper argues that such flow-based density estimation is sensitive to rare, structured departures from the Standard Model and offers a complementary path to hypothesis-driven searches.

What carries the argument

A neural spline flow (NSF) of the rational-quadratic autoregressive variety (RQ-NSF), composed of ten permutation and masked coupling blocks each with 256 hidden units and eight spline bins. This architecture allows exact computation of the per-event log-likelihood −log p(x) in a 67-dimensional standardized feature space (61 original observables plus sine-cosine encodings of angular variables). The log-likelihood serves directly as the anomaly score.

Extended reading notes

Core claim

The paper claims that a neural spline flow density estimator, trained without labels or signal Monte Carlo, can identify a set of 379 dijet events (the 99.9th percentile tail of the learned density) that show statistically significant and physically organized deviations from the QCD background, concentrated in jet substructure variables (Kolmogorov-Smirnov statistic D=0.575 for girth, D=0.549 for tau1). The anomalies are distributed across the mass spectrum and show no narrow resonance, but their multivariate coherence—correlated deviations across girth, N-subjectivity, and energy correlation functions—suggests a genuine difference in internal jet topology rather than a kinematic selection artifact. The paper's chief claim is not that these events represent new physics, but that the methodology provides a validated, model-independent way to surface such structured anomalies from real collider data.

Load-bearing premise

The trained neural spline flow density accurately represents the true Standard Model background distribution in the 61-dimensional feature space, including in the low-density tail regions where the anomaly selection is made, and that the limited data subset (125 GB of a >10 TB dataset) is representative of the full CMS Run2011B dijet sample.

Editorial extensions

If this is right

  • If the method is applied to larger datasets (full Run-1, Run-2,Run-3), the statistical significance of the observed deviations can be evaluated more precisely.
  • The same flow-based anomaly scoring can be combined with explicit signal hypotheses (likeliood ratio) to turn the unsupervised screen into a calibrated search.
  • The validation protocol—mass decorrelation tests, permutation nulls, multi-seed stability—can be adopted as a standard for future unsupervised analyses at the LHC.
  • The observed substructure pattern suggests that the anomalous events may be enriched in boosted W/Z/top jets or other multi-prong signatures, which can be followed up with dedicated taggers.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's finding that the most deviating observables are jet substructure variables (girth, N-subjectivity) rather than simple kinematic ones suggests that the method is particularly sensitive to changes in radiation pattern inside jets, which could arise from new resonances decaying to boosted jets or from modified parton shower, but also from unaccounted miscalibrations of the jet substructur
  • The absence of narrow mass peaks and the broad distribution of anomalous events across mjj implies that this method complements resonant searches: it can detect non-resonant, shape-based new physics that would be missed by bump-hunting.
  • A natural extension would be to apply the same trained flow to score events from a simulated QCD-only sample and verify that no equally strong deviations appear, which would strengthen the case that the real-data anomalies are not a generic property of the tail of the learned density.
Original abstract

We present an unsupervised search for anomalous dijet events in proton--proton collision data using neural spline flow density estimation. A normalizing flow model is trained on a high-dimensional feature space comprising jet, dijet, and event-level observables to learn the dominant Standard Model background directly from data, without assuming a specific signal hypothesis. Events assigned low likelihood under the learned density are identified as potential anomalous events. Using this approach on a CMS Open Data dijet sample, we investigate extreme events in the tail of the anomaly-score distribution and perform an extensive validation and robustness study. This study includes feature-level statistical comparisons, mass decorrelation tests, permutation-based null tests, and evaluations of training stability. The selected anomalous events exhibit notable departures from the background-only expectation, primarily in jet-substructure observables, while remaining stable under several known sources of bias in unsupervised learning. The identified anomalies are distributed across the kinematic phase space and do not exhibit a narrow structure in the dijet invariant-mass spectrum. Instead, they show correlated deviations across multiple observables, consistent with a multivariate difference in jet substructure and event topology rather than a localized resonance. While no claim of new physics is made, this study demonstrates that neural spline flow-based density estimation can be sensitive to rare, structured deviations in collider data and may provide a model-independent exploratory tool for searches for physics beyond the Standard Model at the LHC.

The compiled record

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-09-23: 1 request, 7,028 input tokens.

What the abstract itself states, as probabilities: claims first 0.03 · states limitation 0.93 · human participants 0.01 · controlled comparison 0.40 · randomized assignment 0.01 · states quantitative result 0.06 · compares with prior results 0.11 · mentions code or data availability 0.06.

The abstract, sentence by sentence

arXiv listing abstract v1. Results and the main claim are underlined; .

We present an unsupervised search for anomalous dijet events in proton--proton collision data using neural spline flow density estimation. A normalizing flow model is trained on a high-dimensional feature space comprising jet, dijet, and event-level observables to learn the dominant Standard Model background directly from data, without assuming a specific signal hypothesis. Events assigned low likelihood under the learned density are identified as potential anomalous events. Using this approach on a CMS Open Data dijet sample, we investigate extreme events in the tail of the anomaly-score distribution and perform an extensive validation and robustness study. This study includes feature-level statistical comparisons, mass decorrelation tests, permutation-based null tests, and evaluations of training stability. The selected anomalous events exhibit notable departures from the background-only expectation, primarily in jet-substructure observables, while remaining stable under several known sources of bias in unsupervised learning. The identified anomalies are distributed across the kinematic phase space and do not exhibit a narrow structure in the dijet invariant-mass spectrum. Instead, they show correlated deviations across multiple observables, consistent with a multivariate difference in jet substructure and event topology rather than a localized resonance. While no claim of new physics is made, this study demonstrates that neural spline flow-based density estimation can be sensitive to rare, structured deviations in collider data and may provide a model-independent exploratory tool for searches for physics beyond the Standard Model at the LHC.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#sentencerolethe authors' own workstates the main conclusionstrength
s0 We present an unsupervised search for anomalous dijet events in proton--proton collision data using neural spline flow density estimation. contribution 0.99 0.96 0.05 asserted 0.98
s1 A normalizing flow model is trained on a high-dimensional feature space comprising jet, dijet, and event-level observables to learn the dominant Standard Model background directly from data, without assuming a specific signal hypothesis. method 1.00 0.93 0.03 asserted 1.00
s2 Events assigned low likelihood under the learned density are identified as potential anomalous events. method 0.99 0.81 0.05 asserted 0.98
s3 Using this approach on a CMS Open Data dijet sample, we investigate extreme events in the tail of the anomaly-score distribution and perform an extensive validation and robustness study. contribution 0.56 0.96 0.04 asserted 0.98
s4 This study includes feature-level statistical comparisons, mass decorrelation tests, permutation-based null tests, and evaluations of training stability. method 0.69 0.94 0.03 asserted 0.98
s5 The selected anomalous events exhibit notable departures from the background-only expectation, primarily in jet-substructure observables, while remaining stable under several known sources of bias in unsupervised learning. result 0.99 0.96 0.20 asserted 0.99
s6 The identified anomalies are distributed across the kinematic phase space and do not exhibit a narrow structure in the dijet invariant-mass spectrum. result 1.00 0.95 0.09 asserted 1.00
s7 Instead, they show correlated deviations across multiple observables, consistent with a multivariate difference in jet substructure and event topology rather than a localized resonance. interpretation 0.91 0.94 0.16 hedged possibility 0.81
s8 While no claim of new physics is made, this study demonstrates that neural spline flow-based density estimation can be sensitive to rare, structured deviations in collider data and may provide a model-independent exploratory tool for searches for physics beyond the Standard Model at the LHC. interpretation 0.96 0.72 0.89 asserted 0.51

What kind of paper the classifier reports this is: method or tool 0.67 · evidence from analysis of existing data 1.00.

What was examined, and what was not

sources read
  • arXiv listing abstract v1
not examined
  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers
levels run
  • P abstract
not run
  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

Figures
Figure 1. Schematic overview of the anomaly detection pipeline based on the neural spline
Figure 2. Distribution of neural spline flow anomaly scores, as given by
Figure 3. Two-dimensional subspace projec￾tion of the anomalous events (orange points) in the girth versus τ21 (standardized) subspace. The background density is represented by a two-dimensional color-coded density map (blue = low density, yellow = high density). The anomalous events preferentially occupy the low￾background-density regions (blue), evidencing their systematic displacement relative to the learned background distribution. The concen￾tration of events at τ21 = 0 corresponds to jets with fewer than 3 constituents, a known feature of the CMS particle-flow reconstruction. these…
Figure 4. Feature level distributions for anomalous (99.9-th percentile, red) and background
Figure 5. Dijet invariant mass (mjj ) distribution comparing the inclusive background event sample and anomalous events retained at the 99.9th percentile of anomaly scores. The background shows a smooth distribution typical of a QCD multijet process, while anomalous events are dispersed across the invariant mass spectrum. No evident resonance structure is seen; consistent with a multivariate deviation across multiple correlated kinematic and event-shape variables. clustering in any particular angular or mass window. No correlations are found between anomaly selection and known detector…
Figure 6. Dijet invariant mass (mjj ) distribution for anomalous events identified with the mass-decorrelated neural spline flow model. The model was trained on input variables explicitly excluding mjj and j1 mass to investigate mass sculpting effects. There is no significant excess in the distribution over background predictions, indicating the anomaly selection is not largely influenced by mass variables. consistency of the selection as by its nominal performance; the percentile-based selection and multi-seed, multi-threshold validation adopted here apply the same principle to the…
Figure 7. Dijet invariant mass (mjj ) distribution comparing real anomalies (selected at the 99.9th percentile) with anomalies selected using HT-conditionally permuted scores. The permuted selection shows no systematic deviation from the background distribution, confirming that the observed feature-level deviations are not statistical artifacts. is therefore a proof of concept for the sensitivity of modern unsupervised learning methods to rare, structured features in collider data, and for the role of systematic validation in fully unsupervised analyses. From a physics perspective, the approac
Figure 8. Mean Kolmogorov-Smirnov distance between anomalous and background events as a

Figures reproduced from arXiv:2609.06686 by the authors.

References

Discussion

Disagree with the review, or know something it missed? Leave a comment, highlight any passage above to comment on it, or challenge this review formally.