Specific Expert Learning: Enriching Ensemble Diversity via Knowledge Distillation.

In recent years, ensemble methods have shown sterling performance and gained popularity in visual tasks. However, the performance of an ensemble is limited by the paucity of diversity among the models. Thus, to enrich the diversity of the ensemble, we present the distillation approach-learning from experts (LFEs). Such method involves a novel knowledge distillation (KD) method that we present, specific expert learning (SEL), which can reduce class selectivity and improve the performance on specific weaker classes and overall accuracy. Through SEL, models can acquire different knowledge from distinct networks with various areas of expertise, and a highly diverse ensemble can be obtained afterward. Our experimental results demonstrate that, on CIFAR-10, the accuracy of the ResNet-32 increases 0.91% with SEL, and that the ensemble trained by SEL increases accuracy by 1.13%. Compared to state-of-the-art approaches, for example, DML only improves accuracy by 0.3% and 1.02% on single ResNet-32 and the ensemble, respectively. Furthermore, our proposed architecture also can be applied to ensemble distillation (ED), which applies KD on the ensemble model. In conclusion, our experimental results show that our proposed SEL not only improves the accuracy of a single classifier but also boosts the diversity of the ensemble model.

New steroidal glycosides from Hosta plantaginea (Lam.) Aschers.

Four new furostanol glycosides were isolated from the flowers of Hosta plantaginea (Lam.) Aschers. On the basis of spectroscopic methods including 1D and 2D NMR experiments, their structures were elucidated as 26-O-ß-d-glucopyranosyl-(25R)-22-O-methyl-5α-furostan-2α,3ß,22ξ,26-tetrol 3-O-α-l-rhamnopyranosyl-(1 â†’ 4)-O-ß-d-xylopyranosyl-(1 â†’ 3)-[O-ß-d-glucopyranosyl-(1 â†’ 2)]-O-ß-d-glucopyranosyl-(1 â†’ 4)-ß-d-galactopyranoside (hostaplantagineoside A, 1), 26-O-ß-d-glucopyranosyl-(25R)-5α-furostan-20(22)-ene-2α,3ß,26-triol-3-O-ß-d-glucopyranosyl-(1 â†’ 2)-[O-ß-d-xylopyranosyl-(1 â†’ 3)]-O-ß-d-glucopyranosyl-(1 â†’ 4)-ß-d-galactopyranoside (hostaplantagineoside B, 2), 26-O-ß-d-glucopyranosyl-(25R)-5α-furostan-22(23)-ene-2α,3ß,20α,26-tetraol-3-O-ß-d-glucopyranosyl-(1 â†’ 2)-[O-ß-d-xylopyranosyl-(1 â†’ 3)]-O-ß-d-glucopyranosyl-(1 â†’ 4)-O-ß-d-galactopyranoside (hostaplantagineoside C, 3), 26-O-ß-d-glucopyranosyl-(25R)-5α-furostan-20(22)-ene-2α,3ß,26-triol-3-O-α-l-rhamnopyranosyl-(1 â†’ 4)-O-ß-d-xylopyranosyl-(1 â†’ 3)-[O-ß-d-glucopyranosyl-(1 â†’ 2)]-O-ß-d-glucopyranosyl-(1 â†’ 4)-ß-d-galactopyranoside (hostaplantagineoside D, 4).

[Spatial and temporal distributions of rainfall erosivity in the Yanhe River Basin].

Rainfall erosivity (R) is a measure of effects of the rainfall factor on the potential capacity for soil water erosion. The characteristic of spatial and temporal of R value is the basis for soil erosion prediction. The spatial and temporal distributions of rainfall erosivity were analyzed using daily rainfall data collected between 1980 and 2003 from 22 rainfall stations located across the Yanhe River Basin. The results showed that the seasonal distribution of R values formed a single peak, concentrated between May and September and accounted for 91.61% of the total annual R value. The changes in annual average rainfall erosivity followed a similar trend to those for the annual average rainfall and erosive rainfall. The average annual R value was 1580.58 MJ x mm x (hm2 x h x a)(-1), the maximum annual rainfall erosivity (1981) was 2417.70 MJ x mm x (hm2 x h x a)(-1) and the minimum (1999) was 585.29 MJ x mm x (hm2 x h x a)(-1). The variation coefficient of annual rainfall erosivity was the medium with a value of 0.32. The annual rainfall erosivity was the greatest at the Shaofangbian station with a value of 2190.33 MJ x mm x (hm2 x h x a)(-1), while the least erosivity was found at the Liandaowan and Yangshan station with values of 1151.37 MJ x mm (hm2 x h x a)(-1) and 1146.87 MJ x mm x (hm2 x h x a)(-1), respectively. The annual average R values had a similar trend of spatial distribution with annual average erosive rainfall. The annual changes in R values increased at the rainfall stations of north, while those were found to decrease at other stations of Yanhe River Basin. Overall, the annual rainfall erosivity tended to decrease in the Yanhe River Basin with the trend coefficient value of -0.004.

[Clinical analysis of 255 dental crown fracture].

OBJECTIVE: To analyse dental crown fracture occurrence situation in the long-term mastication process, and discuss the correlation factor of crown fracture occurrence. METHODS: From December, 2005 to December, 2007, 255 crown fracture teeth were chosen as the object of study. The sex, age, teeth position, fracture position and occlusal state of patients were analyzed. RESULTS: In 255 teeth, 1) 76 teeth (29.8%) were maxillary first molars, 45 teeth (17.6%) were mandibular first molars, 41 teeth (16.1%) were maxillary second molars, 37 teeth (14.5%) were mandibular second molars, 32 teeth (12.5%) were maxillary second premolars, 15 teeth (5.9%) were maxillary first premolars, 9 teeth (3.5%) were mandibular second premolars. 2) The fracture position of 158 teeth (62.0%) overlapped with pit and fissure. The fracture position of 97 teeth (38.0%) occurred in dental inclined surface. 3) The occlusal state of 85 patients (33.3%) was normal. The occlusal surface shape of 55 patients (21.6%) was unnormal. The occlusal shape of 115 patients (45.1%) was unnormal. CONCLUSION: The teeth position and occlusal state has certain relations with occurrence of dental crown fracture. Occlusal force was the basic factor for dental crown fracture, but it was not the only one.