SHAining: Multi-Domain Disambiguation
- SHAining is a polysemous research label defining distinct methods in process mining, exoplanet surveys, accelerator science, and computational imaging.
- In process mining, SHAining employs Shapley value analyses to quantify event-log feature contributions, while in astronomy and accelerator studies it underpins robust detection and beam control techniques.
- All applications share the goal of revealing latent structures through innovative statistical methods, calibration-free imaging, and high-repetition operational designs.
to=arxiv.search 彩神争霸能json {"query":"all:SHAining OR ti:SHAining OR abs:SHAining", "max_results": 10, "sort_by": "submittedDate", "sort_order": "descending"} to=arxiv.search 彩神争霸大发快json {"query":"id:(Yang et al., 13 Jan 2026) OR id:(Chomez et al., 21 Jan 2025) OR id:(Maldonado et al., 10 Sep 2025) OR id:(Squicciarini et al., 26 May 2026) OR id:(Mitjans et al., 22 May 2026) OR id:(Hu et al., 2021) OR id:(Cholopoulou et al., 2024) OR id:(Wang et al., 24 Feb 2026) OR id:(Ma et al., 30 Nov 2025) OR id:(Yang et al., 28 May 2026)", "max_results": 10, "sort_by": "relevance", "sort_order": "descending"} SHAining is not a single standardized technical term. In recent arXiv literature, it appears as a polysemous research label spanning several unrelated domains: a Shapley-based explainability framework for process mining; wordplay around SHINE in the SPHERE infrared survey for exoplanets; SHINE-based accelerator concepts for beyond extreme ultraviolet free-electron-laser radiation, advanced external seeding, and an ultrafast neutron source; and acronymic or near-homophonous usages in computational imaging, shading-aware harmonization, and shadow removal (Maldonado et al., 10 Sep 2025, Chomez et al., 21 Jan 2025, Yang et al., 13 Jan 2026, Wang et al., 24 Feb 2026).
1. Nomenclature and scope
In the cited corpus, the term is best understood as a disambiguation problem rather than as the name of one unified theory. Several expansions and usages coexist.
| Usage | Expansion or designation | Domain |
|---|---|---|
| SHAining | Shapley-based framework for event-log characteristic impact analysis | Process mining |
| SHINE | SpHere INfrared survey for Exoplanets | Direct imaging of exoplanets |
| SHINE | Shanghai high-repetition-rate extreme light facility | XFEL, FEL seeding, neutron-source concepts |
| SHINE | Spatial Harmonic Invariant Nonlinear Encoding | Multimode-fiber imaging |
The exoplanet-survey usage is tied to SPHERE@VLT and large-sample direct imaging (Chomez et al., 21 Jan 2025). The accelerator usage is tied to the Shanghai high-repetition-rate extreme light facility and its FEL and neutron-source extensions (Yang et al., 13 Jan 2026). The process-mining usage is the only one in which “SHAining” is itself the formal name of the method (Maldonado et al., 10 Sep 2025). The multimode-fiber imaging usage defines SHINE as “Spatial Harmonic Invariant Nonlinear Encoding,” a calibration and feedback-free coherent imaging paradigm (Wang et al., 24 Feb 2026).
A common misconception is to treat these labels as different parts of the same research program. They are not. They designate technically unrelated systems, infrastructures, and methods developed for distinct scientific objectives.
2. SHINE in direct imaging of exoplanets
SHINE, in the astronomical literature, denotes the SPHERE infrared survey for exoplanets, a large Guaranteed-Time survey carried out with SPHERE on the ESO VLT. One final-results paper presents the observing strategy, data quality, point-source analysis, and detection performances for the full SHINE statistical sample and snapSHINE, using 650 datasets corresponding to 400 stars and uniform processing with PACO ASDI (Chomez et al., 21 Jan 2025). A later full-sample characterization paper reports that SHINE observed 460 stars between 2015 and 2023 and defines a subsample of 333 stars for the final statistical analysis of the survey (Squicciarini et al., 26 May 2026).
The survey was designed to search for and characterize young giant planets and brown dwarfs on wide orbits, measure occurrence rates and radial distributions as a function of stellar mass and age, and study the link between planets and circumstellar disks. The target population emphasizes young, nearby stars, typically younger than 500 Myr and generally within 150 pc, with strong representation of solar-type stars, A/B stars, and moving-group members such as Sco–Cen (Chomez et al., 21 Jan 2025). The full-sample characterization further formalizes age determination from kinematic indicators, lithium abundance, rotation, activity, and isochrone fitting, and it undertakes a thorough vetting for binarity using astrometric, spectroscopic, and imaging data (Squicciarini et al., 26 May 2026).
On the data-analysis side, the 2025 survey paper emphasizes the transition from earlier SPECAL reductions to PACO ASDI. PACO models speckle noise in small patches using a mixture of scaled multivariate Gaussians, outputs statistically grounded S/N maps, and improves IRDIS contrasts by 1–2 magnitudes, or –, at many separations relative to the earlier F150 analysis (Chomez et al., 21 Jan 2025). The same paper reports more than 3500 physical sources, 1180 physical background sources confirmed via common proper motion tests, 34 background sources identified with Gaia DR3 astrometry alone, 26 confirmed bound companions, and 434 physical sources remaining in PROMISING + AMBIGUOUS (Chomez et al., 21 Jan 2025).
The survey’s demographic significance lies in the rarity of wide-orbit giant planets around FGK stars and the higher occurrence around A/B stars. Using F150, planetary-mass companions with $5$– at $10$–$100$ au were reported at for FGK stars and for A/B stars (Chomez et al., 21 Jan 2025). A later deep-learning reanalysis of the F150 sample applied NA-SODINN to every IRDIS H2/H3 dataset within $1.2''$, recovered all known companions, and identified 13 new substellar candidates; after photometric and multi-epoch vetting, only the candidate around Smethells 20 remained a strong target for follow-up (Mitjans et al., 22 May 2026).
3. SHINE as accelerator platform: BEUV, external seeding, and ultrafast neutrons
In accelerator science, SHINE refers to a superconducting-linac XFEL facility whose design parameters make several additional source concepts feasible. One study analyzes beyond extreme ultraviolet radiation at SHINE and treats wavelengths around $6.x$ nm as an almost natural by-product of the FEL-II soft-X-ray beamline. Using 0 GeV electrons, a 1 mm planar undulator, 2 pC bunch charge, 3 A peak current, 4, 5 slice energy spread, and up to 6 MHz repetition rate, it finds that kilowatt-level BEUV radiation with controllable polarization is achievable through tapering and an APPLE-III EPU afterburner (Yang et al., 13 Jan 2026). The reported operating envelope includes linearly polarized BEUV at about 7–8 kW in tapered operation and circularly polarized BEUV at about 9 kW with an estimated degree of circular polarization of $5$0 (Yang et al., 13 Jan 2026).
A second accelerator paper uses the SHINE bypass line as a platform for advanced externally seeded soft-X-ray FEL operation. It studies three compatible high-repetition-rate configurations—self-modulation cascaded HGHG, self-modulation EEHG, and direct-amplification-driven EEHG—under representative bypass parameters of about $5$1 GeV beam energy, $5$2 A peak current, $5$3 pC bunch charge, $5$4 fs bunch length, $5$5, and $5$6 keV (Yang et al., 28 May 2026). Numerical simulations indicate harmonic generation beyond the 30th order, with SM-EEHG and SM-HGHG cascade targeting $5$7–$5$8 harmonics at $5$9 MHz and DE-EEHG targeting 0–1 harmonics at 2 MHz (Yang et al., 28 May 2026). The same study proposes a common modulator-chicane layout with five 3 m modulators, four chicanes, three seed-laser injection points, and one optical delay platform to preserve compatibility among candidate modes (Yang et al., 28 May 2026).
A third SHINE-centered concept is NeutrSHINE, a high repetition rate ultrafast neutron source driven by the 8 GeV SHINE electron beam. The concept uses 3 mA average current, 4 MHz repetition rate, and 5 fs bunches on a tantalum target to produce a broad neutron spectrum from keV to 6 GeV (Ma et al., 30 Nov 2025). The optimized target is a 7 base with 7 cm tantalum thickness, water-cooled in a layered disk-style structure, and it yields about 8 on average (Ma et al., 30 Nov 2025). For high-energy neutrons, the temporal structure is ultrafast on neutron-source standards: at 500 MeV, 9 ps, corresponding to an FWHM of about $10$0 ps (Ma et al., 30 Nov 2025).
4. Shading, shadow removal, and invariant spectral encoding
A separate cluster of work uses names that are conceptually adjacent to “SHAining” because they center on shading, illumination, and shadow-aware reconstruction. In image harmonization, SIDNet introduces a “Shading-aware Illumination Descriptor” and decomposes harmonization into illumination estimation of the background image and re-rendering of foreground objects under background illumination (Hu et al., 2021). Its descriptor $10$1 combines with learned shading bases $10$2 to generate a full shading field, while albedo is estimated separately and recombined in a neural rendering framework (Hu et al., 2021). On the synthetic test set, the reported all-scene results for SIDNet are fMAE $10$3, fPSNR $10$4, fSSIM $10$5, and LPIPS $10$6, with further gains when the illumination map itself is used (Hu et al., 2021).
In shadow removal, SHAU denotes the Soft-Hard Attention U-Net and is paired with the Multiscale Shadow Removal Dataset. The architecture uses multiscale feature extraction blocks with $10$7, $10$8, and $10$9 convolutions, hard attention guided by a binary shadow mask in the encoder, and soft attention in the decoder (Cholopoulou et al., 2024). MSRD contains 8,582 synthetic image triplets—shadowed image, shadow-free image, and binary shadow mask—at $100$0 resolution, generated in Unreal Engine 5.1 and designed to exhibit complex, overlapping, multiscale shadows (Cholopoulou et al., 2024). The paper reports that SHAU improves the Peak Signal-to-Noise Ratio and Root Mean Square Error for the shadow area by $100$1 and $100$2, respectively, over the relevant state-of-the-art shadow removal methods (Cholopoulou et al., 2024).
In multimode-fiber imaging, SHINE is redefined as “Spatial Harmonic Invariant Nonlinear Encoding.” The method uses angle-dependent phase-matching conditions of second-harmonic generation to encode spatial features into broadband spectral signatures that are intrinsically insensitive to modal scrambling, fiber bending, movement, and even structural variations across distinct multimode fibers (Wang et al., 24 Feb 2026). Experimentally, it achieves an average Pearson correlation coefficient of $100$3 for reconstruction on Fashion-MNIST, a classification accuracy of $100$4 on the HERLEV biomedical dataset, and cross-fiber reconstruction with PCC $100$5 when a model trained on a single MMF is tested on previously unseen fibers (Wang et al., 24 Feb 2026).
These works are independent of the SHINE exoplanet survey and the SHINE accelerator facility. Their shared emphasis is on making latent structure visible through explicit illumination, attention, or invariant encoding.
5. SHAining in process mining
The formal use of “SHAining” appears in process mining, where it names a framework for explaining how event-log characteristics affect algorithmic evaluation metrics. The method addresses a recurring problem: process discovery metrics vary widely with structural event-log characteristics, but prior studies usually benchmark algorithms on fixed sets of real-world logs without quantifying the marginal contribution of individual characteristics, especially when these characteristics co-occur (Maldonado et al., 10 Sep 2025).
The framework is explicitly associational rather than causal. Because event logs are generated from processes in which characteristics overlap, SHAining does not attempt to isolate the causal effect of one characteristic while holding all others fixed. Instead, it uses Shapley values over the space of data-generating configurations to quantify the average marginal contribution of each characteristic to a metric. For a player $100$6 and coalition $100$7, the Shapley value is
$100$8
The downstream task is process discovery, and the evaluation metrics include fitness, precision, F-score, size, control-flow complexity, and execution time (Maldonado et al., 10 Sep 2025).
The experimental pipeline combines controlled log generation with black-box algorithm evaluation. Event logs are generated with GEDI from feature configurations sampled over eight selected meta-features: activities_q1, n_unique_start_activities, start_activities_q1, eventropy_k_block_ratio_3, ratio_top_5_variants, skewness_variant_occurance, trace_len_kurtosis, and trace_len_variance (Maldonado et al., 10 Sep 2025). Using eight features, ten equidistant values per feature, and coalitions up to size three, the theoretical configuration space contains 58,880 combinations; in practice, the study analyzes over 22,000 event logs (Maldonado et al., 10 Sep 2025). The process discovery algorithms are Inductive Miner, ILP Miner, and Split Miner.
The reported findings are highly structured. Across metrics, activities_q1 and skewness_variant_occurance are the most influential features, whereas ratio_top_5_variants is consistently among the least influential (Maldonado et al., 10 Sep 2025). For robustness, Inductive Miner shows high average Shapley values with low variance, ILP Miner is about 40% more sensitive, and Split Miner lies between them (Maldonado et al., 10 Sep 2025). The framework also studies how the value of a characteristic correlates with its Shapley contribution. For example, higher eventropy_k_block_ratio_3 tends to increase fitness while decreasing precision and F-score and increasing control-flow complexity and size for ILP Miner (Maldonado et al., 10 Sep 2025). This makes SHAining not merely a ranking device but an associational map from log structure to algorithmic behavior.
6. Disambiguation, significance, and recurrent motifs
The principal encyclopedic fact about SHAining is terminological: it is not a unified research area. In one domain it is a formal Shapley-based explainability method; in others it is a wordplay or acronymic construction attached to exoplanet direct imaging, FEL and neutron-source extensions of a superconducting linac, multimode-fiber imaging, and shading- or shadow-aware vision systems (Maldonado et al., 10 Sep 2025, Chomez et al., 21 Jan 2025, Yang et al., 13 Jan 2026, Wang et al., 24 Feb 2026).
A second point is methodological. Despite the absence of direct technical overlap, many of these works emphasize robustness under structured variability. SHINE direct imaging centers on statistically reliable detection limits and homogeneous processing with PACO ASDI (Chomez et al., 21 Jan 2025). SHINE accelerator studies emphasize MHz-level repetition, flexible polarization, and compatibility across seeded-FEL modes (Yang et al., 13 Jan 2026, Yang et al., 28 May 2026). SHINE multimode-fiber imaging explicitly targets operation without calibration or feedback under bending, motion, and cross-fiber transfer (Wang et al., 24 Feb 2026). SHAining in process mining quantifies how log characteristics alter fitness, precision, complexity, and runtime across algorithms (Maldonado et al., 10 Sep 2025). This suggests a shared rhetorical function of the label family: making hidden structure measurable, explainable, or operationally controllable.
A plausible implication is that future encounters with “SHAining” in the literature will continue to require local disambiguation. In current arXiv usage, the safest interpretation is always contextual: in astronomy it most often refers to the SHINE survey; in accelerator physics to the SHINE facility and its extensions; in process mining to the Shapley-based framework; and in computational imaging to illumination-, shadow-, or spectral-invariance-centered methods.
