A New Knowledge Distillation Method for Object Detection Based on EMD

As a common method of model compression, the knowledge distillation method can distill the knowledge from the complex large model with strong learning ability to student small model with weak learning ability in the training process, to improve the accuracy and performance of the small model. At present, there has been much knowledge distillation methods specially designed for object detection and achieved good results. However, almost all methods failed to solve the problem of performance degradation caused by the high noise in the current detection framework. In this study, we proposed a feature automatic weight learning method based on EMD to solve these two problems. That is, the EMD method is used to process the space vector to reduce the impact of negative transfer and noise as much as possible, and at the same time, the weights are allocated adaptive to reduce student model’s learning from the teacher model with poor performance and make students more inclined to learn from good teachers. The loss (EMD Loss) was redesigned, and the HEAD was improved to fit our approach. We have carried out different comprehensive performance tests on multiple datasets, including PASCAL, KITTI, ILSVRC, and MS-COCO, and obtained encouraging results, which can not only be applied to the one-stage and two-stage detectors but also can be used radiatively with some other methods.

Distillation-Adsorption of Bioethanol Using Natural Kaolin

Bioethanol has been known as one of the alternative energy sources to replace petroleum. The use of bioethanol as a fuel has requirements that must meet fuel standards or refer to as Fuel Grade Ethanol (FGE). Further refining is required to achieve this qualification. The distillation method is the most commonly used method in bioethanol purification, but the results are less than optimal and sometimes the ethanol produced will be more volatile. Distillation-adsorption method using clay-based materials was developed to solve this problem. The purpose of this study is to purify bioethanol and determine the physical characteristics of the bioethanol produced and to assess the % purity of bioethanol from adsorption distillation. Characterization of activated natural kaolin as adsorbent, determination of the optimum contact time in the distillation-adsorption process, determination of the optimum mass of kaolin used and determination of the physical properties of bioethanol have been carried out. The results showed that 1 hour adsorption distillation could increase the ethanol content from 7.9% to 29.33%. In the study of the effect of the mass of kaolin, the optimum result was 25.25% ethanol using a mass of 5 grams. physical properties test of bioethanol obtained a density of 0.8211 gr/cm3 and a viscosity of 2.096 cps.

Biological Activity of Essential Oils of Four Juniper Species and Their Potential as Biopesticides

The objective of this study was to assess the biological activity of essential oils (EOs) of four Juniperus species obtained via two different distillation methods and their potential as biopesticides. The studied factors were juniper species (Juniperus communis L., J. oxyсedrus L., J. pygmaea C. Koch., and J. sibirica Burgsd), plant sex (male (M) and female (F)), and distillation method (hydrodistillation via a standard Clevenger apparatus (ClevA) and semi-commercial (SCom) steam distillation). The hypothesis was that the EO will have differential antioxidant, antimicrobial, and insecticidal activities as a function of plant species, plant sex, and distillation method. The two distillation methods resulted in similar EO composition within a given species. However, there were differences in the EO content (yield) due to the sex of the plant, and also differences in the proportions of some EO components. The concentration of α-pinene, β-caryophyllene, δ-cadinene and δ-cadinol was dissimilar between the EO of M and F plants within all four species. Additionally, M and F plants of J. pygmaea, and J. sibirica had significantly different concentrations of sabinene within the respective species. The EOs obtained via ClevA extraction showed higher antioxidant capacity within a species compared with those from SCom extraction. All of the tested EOs had significant repellent and insecticidal activity against the two aphid species Rhopalosiphum padi (bird cherry-oat aphid) and Sitobion avenae (English grain aphid) at concentrations of the EO in the solution of 1%, 2.5%, and 5%. The tested EOs demonstrated moderate activity against selected pathogens Fusarium spp., Botrytis cinerea, Colletotrichum spp., Rhizoctonia solani and Cylindrocarpon pauciseptatum. The results demonstrate that the standard ClevA would provide comparable EO content and composition in comparison with SCom steam distillation; however, even slight differences in the EO composition may translate into differential bioactivity.

Chemical composition and anticancer effects of Zingiber officinale volatile oil

Zingiber officinale rhizome is used in ethnomedicine in treating tumor-related ailments. This study was undertaken to evaluate the cytotoxic effects of this plant. The oil was extracted using a Clavenger apparatus by hydro-distillation method. Preliminary screening was carried out with brine shrimp cytotoxicity test at 10-1000 µg/mL. The volatile oil was further tested on breast cancer (AU 565) and cervical cancer (HeLa) at 50 µg /mL using MTT assay and later subjected to GCMS analysis. LC50 of 157.75 µg/mL was obtained in the brine shrimp mortality assay. Z. officinale oil showed high anticancer activities with 50 and 43 % inhibitions against HeLa and AU 565 cells respectively. GCMS analysis revealed the major constituents of Z. officinale oil as a-citral (11.68 %) and a-citral (10.18 %). These results suggest the medicinal potency of this plant oil.

Self-boosting for Feature Distillation

Knowledge distillation is a simple but effective method for model compression, which obtains a better-performing small network (Student) by learning from a well-trained large network (Teacher). However, when the difference in the model sizes of Student and Teacher is large, the gap in capacity leads to poor performance of Student. Existing methods focus on seeking simplified or more effective knowledge from Teacher to narrow the Teacher-Student gap, while we address this problem by Student's self-boosting. Specifically, we propose a novel distillation method named Self-boosting Feature Distillation (SFD), which eases the Teacher-Student gap by feature integration and self-distillation of Student. Three different modules are designed for feature integration to enhance the discriminability of Student's feature, which leads to improving the order of convergence in theory. Moreover, an easy-to-operate self-distillation strategy is put forward to stabilize the training process and promote the performance of Student, without additional forward propagation or memory consumption. Extensive experiments on multiple benchmarks and networks show that our method is significantly superior to existing methods.