Table 3.
Impact of multimodality on different datasets.
Experiences | σ | η | ⟨Δz⟩ | Count |
---|---|---|---|---|
10−3 | % | 10−3 | 103 | |
SDSS | ||||
Baseline | 07.99 | 0.18 | 0.34 | 516.5 |
Multimodal | 07.85 | 0.16 | 0.31 | 516.5 |
G(M) | 1.74% | 10.88% | 6.28% | – |
pvalue | 0.0 | 0.0 | 0.0 | – |
CFHTLS | ||||
Baseline | 16.01 | 0.85 | 0.22 | 108.5 |
Multimodal | 15.35 | 0.79 | 0.29 | 108.5 |
G(M) | 4.13% | 7.22% | −24.05% | – |
pvalue | 0.0 | 0.0002 | 0.15 | – |
HSC-6b | ||||
Baseline | 09.14 | 1.25 | 1.97 | 46.8 |
Multimodal | 08.87 | 1.20 | 1.63 | 46.8 |
G(M) | 2.96% | 3.94% | 17.33% | – |
pvalue | 0.0 | 0.0575 | 0.04 | – |
HSC-9b | ||||
Baseline | 08.41 | 1.24 | 1.58 | 33.1 |
Multimodal | 07.60 | 1.19 | 1.64 | 33.1 |
G(M) | 10.1% | 3.67% | −3.1% | – |
pvalue | 0.0 | 0.11 | 0.40 | – |
HSC-9b with 3DHST redshifts | ||||
Baseline | 14.44 | 2.46 | 13.28 | 2.2 |
Multimodal | 13.88 | 2.37 | 10.6 | 2.2 |
G(M) | 3.93% | 3.71% | 20.19% | – |
pvalue | 0.069 | 0.27 | 0.10 | – |
HSC-9b with PRIMUS redshifts | ||||
Baseline | 12.34 | 2.66 | 11.84 | 15 |
Multimodal | 11.38 | 1.85 | 09.23 | 15 |
G(M) | 7.74% | 30.4% | 22.01% | – |
pvalue | 0.0 | 0.0 | 0.0 | – |
HSC-9b with COSMOS2020 photometric redshifts | ||||
Baseline | 12.01 | 1.01 | 8.74 | 43.7 |
Multimodal | 11.46 | 0.83 | 6.82 | 43.7 |
G(M) | 4.57% | 17.08% | 21.97% | – |
pvalue | 0.0 | 0.0 | 0.0001 | – |
Notes. The MAD, the outlier fraction, and the bias are reported for the baseline and the multimodal models, along with the relative difference and the pvalue as a measure of the significativity of the observed difference. The sizes of the datasets down to i = 24 are reported in the last column. For the nine-band experiments, some objects were missing the j band, so we used redshift estimations of models trained in those conditions.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.