A method for the local and global interpretation of a black-box model on the basis of the well-known generalized additive models is proposed. It can be viewed as an extension or a modification of the ...algorithm using the neural additive model. The method is based on using an ensemble of gradient boosting machines (GBMs) such that each GBM is learned on a single feature and produces a shape function of the feature. The ensemble is composed as a weighted sum of separate GBMs resulting a weighted sum of shape functions which form the generalized additive model. GBMs are built in parallel using randomized decision trees of depth 1, which provide a very simple architecture. Weights of GBMs as well as features are computed in each iteration of boosting by using the Lasso method and then updated by means of a specific smoothing procedure. In contrast to the neural additive model, the method provides weights of features in the explicit form, and it is simply trained. A lot of numerical experiments with an algorithm implementing the proposed method on synthetic and real datasets demonstrate its efficiency and properties for local and global interpretation.
Search for heavy neutrino in K+→μ+νH decay Sadovsky, A. S.; Kurshetsov, V. F.; Filin, A. P. ...
The European physical journal. C, Particles and fields,
02/2018, Letnik:
78, Številka:
2
Journal Article
Recenzirano
Odprti dostop
A high statistics data sample of the
K
+
→
μ
+
ν
μ
decay was accumulated by the OKA experiment in 2012. The missing mass analysis was performed to search for the decay channel
K
+
→
μ
+
ν
H
with a ...hypothetic stable heavy neutrino in the final state. The obtained missing mass spectrum does not show peaks that could be attributed to existence of stable heavy neutrinos in the mass range
(
270
<
m
ν
H
<
375
)
MeV
/
c
2
. As a result, upper limits on the branching ratio and on the value of the mixing element
|
U
μ
H
|
2
are obtained.
Abstract A high statistics data sample of the $$K^{+}\rightarrow \mu ^{+}\nu _{\mu }$$ K+→μ+νμ decay was accumulated by the OKA experiment in 2012. The missing mass analysis was performed to search ...for the decay channel $$K^{+}\rightarrow \mu ^{+}\nu _{H}$$ K+→μ+νH with a hypothetic stable heavy neutrino in the final state. The obtained missing mass spectrum does not show peaks that could be attributed to existence of stable heavy neutrinos in the mass range $$(270< m_{\nu _{H}} < 375)$$ (270<mνH<375) MeV$$/c^{2}$$ /c2 . As a result, upper limits on the branching ratio and on the value of the mixing element $$|U_{\mu H}|^{2}$$ |UμH|2 are obtained.
Abstract A high statistics data sample of the decays of $$K^+$$ K+ mesons to three charged particles was accumulated by the OKA experiment in 2012 and 2013. This allowed to select a clean sample of ...about 450 events with $$K^{+}\rightarrow \pi ^{+}\pi ^{-}\pi ^{+}\gamma $$ K+→π+π-π+γ decays with the energy of the photon in the kaon rest frame greater than 30 MeV. The measured branching fraction of the $$K^{+}\rightarrow \pi ^{+}\pi ^{-}\pi ^{+}\gamma $$ K+→π+π-π+γ , with $$E_{\gamma }^{*} > 30\ \hbox {MeV}$$ Eγ∗>30MeV is equal to $$(0.71 \pm 0.05) \times 10^{-5}$$ (0.71±0.05)×10-5 . The measured differential branching fraction over photon energy is compared with the prediction of the chiral perturbation theory to $${\mathcal {O}}(p^{4})$$ O(p4) . A search for an up-down asymmetry of the photon with respect to the hadronic system decay plane is also performed.
Tunka-133: Results of 3 year operation Prosin, V.V.; Berezhnev, S.F.; Budnev, N.M. ...
Nuclear instruments & methods in physics research. Section A, Accelerators, spectrometers, detectors and associated equipment,
08/2014, Letnik:
756
Journal Article
Recenzirano
Odprti dostop
The EAS Cherenkov light array Tunka-133, with ~3km2 geometric area, is taking data since 2009. The array permits a detailed study of cosmic ray energy spectrum and mass composition in the PeV energy ...range. After a short description of the methods of EAS parameter reconstruction, we present the all-particle energy spectrum and results of studying CR composition, based on 3 seasons of array operation. In the last part of the paper, we discuss possible interpretations of the obtained results.
The main stages in the creation of the Russian segment of the MASTER network of robotic telescopes is described. This network is designed for studies of the prompt optical emission of gammaray bursts ...(GRBs; optical emission synchronous with the gamma-ray radiation) and surveys of the sky aimed at discovering uncataloged objects and photometric studies for various programs. The first results obtained by the network, during its construction and immediately after its completion in December 2010, are presented. Eighty-nine alert pointings at GRBs (in most cases, being the first ground telescopes to point at the GRBs) were made from September 2006 through July 2011. The MASTER network holds first place in the world in terms of the total number of first pointings, and currently more than half of first pointings at GRBs by ground telescopes are made by the MASTER network. Photometric light curves of GRB 091020, GRB 091127, GRB 100901A, GRB 100906A, GRB 10925A, GRB 110106A, GRB 110422A, and GRB 110530A are presented. It is especially important that prompt emission was observed for GRB 100901A and GRB 100906A, and thar GRB 091127, GRB 110422A, and GRB 110106A were observed from the first seconds in two polarizations. Very-wide-field cameras carried out synchronous observations of the prompt emission of GRB 081102, GRB 081130B, GRB 090305B, GRB 090320B, GRB 090328, and GRB 090424. Discoveries of Type Ia supernovae are ongoing (among them the brightest supernova in 2009): 2008gy, 2009nr, 2010V, and others. In all, photometry of 387 supernovae has been carried out, 43 of which were either discovered or first observed with MASTER telescopes; more than half of these are Type Ia supernovae. Photometric studies of the open clusters NGC 7129 and NGC 7142 have been conducted, leading to the discovery of 38 variable stars. Sixty-nine optical transients have been discovered.
We present the results of the prompt, early and afterglow optical observations of five γ-ray bursts (GRBs): GRB 100901A, GRB 100902A, GRB 100905A, GRB 100906A and GRB 101020A. These observations were ...made with the Mobile Astronomical System of TElescope-Robots in Russia (MASTER-II Net), the 1.5-m telescope of the Sierra Nevada Observatory and the 2.56-m Nordic Optical Telescope. For two sources, GRB 100901A and GRB 100906A, we detected optical counterparts and obtained light curves starting before the cessation of γ-ray emission, at 113 and 48 s after the trigger, respectively. Observations of GRB 100906A were conducted in two polarizing filters. Observations of the other three bursts gave the upper limits on the optical flux; their properties are briefly discussed. A more detailed analysis of GRB 100901A and GRB 100906A, supplemented by Swift data, provides the following results and indicates different origins for the prompt optical radiation in the two bursts. The light-curve patterns and spectral distributions suggest that there is a common production site for the prompt optical and high-energy emission in GRB 100901A. The results of the spectral fits for GRB 100901A in the range from optical to X-ray favour power-law energy distributions and a consistent value of the optical extinction in the host galaxy. GRB 100906A produced a smoothly peaking optical light curve, suggesting that the prompt optical radiation in this GRB originated in a front shock. This is supported by a spectral analysis. We have found that the Amati and Ghirlanda relations are satisfied for GRB 100906A. We obtain an upper limit on the value of the optical extinction on the host of GRB 100906A.
Plant mitochondrial genomes (mitogenomes) can be structurally complex while their size can vary from ~ 222 Kbp in Brassica napus to 11.3 Mbp in Silene conica. To date, in comparison with the number ...of plant species, only a few plant mitogenomes have been sequenced and released, particularly for conifers (the Pinaceae family). Conifers cover an ancient group of land plants that includes about 600 species, and which are of great ecological and economical value. Among them, Siberian larch (Larix sibirica Ledeb.) represents one of the keystone species in Siberian boreal forests. Yet, despite its importance for evolutionary and population studies, the mitogenome of Siberian larch has not yet been assembled and studied.
Two sources of DNA sequences were used to search for mitochondrial DNA (mtDNA) sequences: mtDNA enriched samples and nucleotide reads generated in the de novo whole genome sequencing project, respectively. The assembly of the Siberian larch mitogenome contained nine contigs, with the shortest and the largest contigs being 24,767 bp and 4,008,762 bp, respectively. The total size of the genome was estimated at 11.7 Mbp. In total, 40 protein-coding, 34 tRNA, and 3 rRNA genes and numerous repetitive elements (REs) were annotated in this mitogenome. In total, 864 C-to-U RNA editing sites were found for 38 out of 40 protein-coding genes. The immense size of this genome, currently the largest reported, can be partly explained by variable numbers of mobile genetic elements, and introns, but unlikely by plasmid-related sequences. We found few plasmid-like insertions representing only 0.11% of the entire Siberian larch mitogenome.
Our study showed that the size of the Siberian larch mitogenome is much larger than in other so far studied Gymnosperms, and in the same range as for the annual flowering plant Silene conica (11.3 Mbp). Similar to other species, the Siberian larch mitogenome contains relatively few genes, and despite its huge size, the repeated and low complexity regions cover only 14.46% of the mitogenome sequence.
A new approach called ABRF (the attention-based random forest) and its modifications for applying the attention mechanism to the random forest (RF) for regression and classification are proposed. The ...main idea behind the proposed ABRF models is to assign attention weights with trainable parameters to decision trees in a specific way. The attention weights depend on the distance between an instance, which falls into a corresponding leaf of a tree, and training instances, which fall in the same leaf. This idea stems from representation of the Nadaraya–Watson kernel regression in the form of a RF. Three modifications of the general approach are proposed. The first one is based on applying the Huber’s contamination model and on computing the attention weights by solving quadratic or linear optimization problems. The second and the third modifications use the gradient-based algorithms for computing an extended set of the attention trainable parameters. Numerical experiments with various regression and classification datasets illustrate the proposed method. The code implementing the approach is publicly available.
PDF
