RGP: Neural Network Pruning Through Regular Graph With Edges Swapping.

Deep learning technology has found a promising application in lightweight model design, for which pruning is an effective means of achieving a large reduction in both model parameters and float points operations (FLOPs). The existing neural network pruning methods mostly start from the consideration of the importance of model parameters and design parameter evaluation metrics to perform parameter pruning iteratively. These methods were not studied from the perspective of network model topology, so they might be effective but not efficient, and they require completely different pruning for different datasets. In this article, we study the graph structure of the neural network and propose a regular graph pruning (RGP) method to perform a one-shot neural network pruning. Specifically, we first generate a regular graph and set its node-degree values to meet the preset pruning ratio. Then, we reduce the average shortest path-length (ASPL) of the graph by swapping edges to obtain the optimal edge distribution. Finally, we map the obtained graph to a neural network structure to realize pruning. Our experiments demonstrate that the ASPL of the graph is negatively correlated with the classification accuracy of the neural network and that RGP has a strong precision retention capability with high parameter reduction (more than 90%) and FLOPs reduction (more than 90%) (the code for quick use and reproduction is available at https://github.com/Holidays1999/Neural-Network-Pruning-through-its-RegularGraph-Structure).

[Clinical observation on treatment of PC-BPPV by dynamic and static acupuncture method combined with manual reduction].

OBJECTIVE: To observe the clinical effect and advantages of dynamic and static acupuncture method combined with manual reduction on posterior semicircular canal benign paroxysmal positional vertigo (PC-BPPV). METHODS: Ninety patients with PC-BPPV who met the inclusion criteria were randomly divided into manual reduction control group, acupuncture control group and experimental group, with 30 cases in each group. Epley reduction method was used for manual reduction control treatment of patients in the manual reduction control group, until there was no obvious vertigo of patients. Patients in the acupuncture control group received ordinary acupuncture treatment, while patients in the experiment group received dynamic and static acupuncture treatment, both on the basis of manual reduction control treatment. Baihui(GV20) and Yintang(GV24+), Sanyinjiao(SP6), Zhongzhu(TE3), Houxi(SI3) and Waiguan(TE5) on the healthy side, and the vertigo-auditory area and Fengchi(GB20) on the affected side were selected for acupuncture intervention, which was performed once a day, with needles retained for 30 minutes in two acupuncture groups. Every six times was taken as a session and two sessions were required. Dizziness handicap inventory (DHI) scale and visual analogue score (VAS) were used to evaluate the degree of vertigo before, after 1 and 2 sessions of treatment respectively. RESULTS: Compared with those before treatment, the DHI score and VAS score of each group after 1 and 2 sessions of treatment were both significantly decreased (P<0.05). Compared with the acupuncture control group and the manual reduction control group, the DHI score of the experiment group was significantly decreased (P<0.05) after 1 and 2 sessions of treatment. VAS score of the experiment group was significantly decreased compared with that of the manual reduction control group(P<0.05) after 1 and 2 sessions of treatment. The total effective rate of the experiment group was 86.67%, better than those of the acupuncture control group (83.33%, P<0.05) and the manual reduction control group(66.67%, P<0.05). CONCLUSION: The dynamic and static acupuncture method combined with manual reduction effectively improved vertigo symptoms in PC-BPPV with rapid and lasting effects, which is worthy of further clinical promotion and application.

[Effects of electroacupuncture at "Siguan" points on the expression of colonic 5-hydroxytryptamine and fecal short-chain fatty acids in rats with post-stroke depression].

OBJECTIVE: To observe the effects of electroacupuncture at "Siguan" points on behavior, colonic 5-hydroxytryptamine (5-HT) and fecal short-chain fatty acids (SCFAs) in rats with post-stroke depression (PSD), and explore the effect mechanism of electroacupuncture at Siguan points on PSD. METHODS: Fifty SD rats were randomly divided into a sham-operation group, a stroke group, a PSD group, a drug group and an electroacupuncture group, with 10 rats in each one. The stroke model was established by middle cerebral artery occlusion (MCAO) method in the stroke group; except for the sham-operation group, the rats in the other groups were intervened with MCAO combined with solitary and chronic unpredictable mild stress (CUMS) to establish PSD model. In the electroacupuncture group, electroacupuncture was delivered at "Hegu" (LI 4) and "Taichong" (LR 3), with disperse-dense wave, 2 Hz/10 Hz in frequency, for 30 min in each intervention, once daily, for consecutive 21 days. Simultaneously, distilled water (0.01 L•kg-1•d-1) was administrated intragastrically. Fluoxetine solution (2.33 mg•kg-1•d-1) was given by gavage , once a day and for 21 days in the drug group. The same procedure of fixation and gavage with distilled water were adopted in the sham-operation group, the stroke group and the PSD group. Separately, before stroke modeling, after PSD modeling and after 21-day intervention, the consumption of sugar water and the scores of horizontal movement and vertical movement in open-field test were observed. After 21-day intervention, the content of colonic 5-HT was detected by immunohistochemical method, and that of fecal SCFAs was determined by gas chromatography mass spectrometry. RESULTS: After PSD modeling, compared with the stroke group, the sugar water consumption, the horizontal movement scores and vertical movement scores of the open-field test were all reduced in the PSD group, the drug group and the electroacupuncture group (P<0.05). After 21-day intervention, the sugar water consumption and the scores of horizontal movement and vertical movement of the open-field test were increased in the drug group and the electroacupuncture group (P<0.05) when compared with the PSD group; and the horizontal movement score in the electroacupuncture group was lower than that of the drug group (P<0.05). Compared with the sham-operation group, the contents of total fecal SCFAs and acetic acid were lower in the stroke group (P<0.05), and the contents of colonic 5-HT and total fecal SCFAs, acetic acid, propionic acid and butyric acid were reduced in the PSD group (P<0.05). In comparison with the PSD group, the contents of colonic 5-HT and total fecal SCFAs, acetic acid and propionic acid were increased in the drug group and the electroacupuncture group (P<0.05); and the content of colonic 5-HT in the electroacupuncture group was lower than that of the drug group (P<0.05). The level of colonic 5-HT was positively correlated with the contents of total fecal SCFAs and propionic acid (r=0.424, P=0.005; r=0.427, P=0.004). CONCLUSION: Electroacupuncture at "Siguan" points can relieve the depression-like behavior of PSD rats, and its underlying mechanism may be related to the regulation of fecal SCFAs, which affects the release of colonic 5-HT.

QSAR study of phenolic compounds and their anti-DPPH radical activity by discriminant analysis.

Phenolic compounds (PCs) could be applied to reduce reactive oxygen species (ROS) levels, and are used to prevent and treat diseases related to oxidative stress. QSAR study was applied to elucidate the relationship between the molecular descriptors and physicochemical properties of polyphenol analogues and their DPPH radical scavenging capability, to guide the design and discovery of highly-potent antioxidant substances more efficiently. PubMed database was used to collect 99 PCs with antioxidant activity, whereas, 105 negative PCs were found in ChEMBL database; their molecular descriptors were generated with Python's Rdkit package. While the molecular descriptors significantly related to the antioxidant activity of PCs were filtered by t-test. The prediction QSAR model was then established by discriminant analysis, and the obtained model was verified by the back-substitution and Leave-One-Out cross-validation methods along with heat map. It was revealed that the anti-DPPH radical activity of PCs was correlated with the drug-likeness and molecular fingerprints, physicochemical, topological, constitutional and electronic property. The established QSAR model could explicitly predict the antioxidant activity of polyphenols, thus were applicable to evaluate the potential of candidates as antioxidants.