“Reading Pictures Instead of Looking”: RGB-D Image-Based Action Recognition via Capsule Network and Kalman Filter

This paper proposes an action recognition algorithm based on the capsule network and Kalman filter called “Reading Pictures Instead of Looking” (RPIL). This method resolves the convolutional neural network’s over sensitivity to rotation and scaling and increases the interpretability of the model as per the spatial coordinates in graphics. The capsule network is first used to obtain the components of the target human body. The detected parts and their attribute parameters (e.g., spatial coordinates, color) are then analyzed by Bert. A Kalman filter analyzes the predicted capsules and filters out any misinformation to prevent the action recognition results from being affected by incorrectly predicted capsules. The parameters between neuron layers are evaluated, then the structure is pruned into a dendritic network to enhance the computational efficiency of the algorithm. This minimizes the dependence of in-depth learning on the random features extracted by the CNN without sacrificing the model’s accuracy. The association between hidden layers of the neural network is also explained. With a 90% observation rate, the OAD dataset test precision is 83.3%, the ChaLearn Gesture dataset test precision is 72.2%, and the G3D dataset test precision is 86.5%. The RPILNet also satisfies real-time operation requirements (>30 fps).

Similarity Measures for Learning in Lattice Based Biomimetic Neural Networks

This paper presents a novel lattice based biomimetic neural network trained by means of a similarity measure derived from a lattice positive valuation. For a wide class of pattern recognition problems, the proposed artificial neural network, implemented as a dendritic hetero-associative memory delivers high percentages of successful classification. The memory is a feedforward dendritic network whose arithmetical operations are based on lattice algebra and can be applied to real multivalued inputs. In this approach, the realization of recognition tasks, shows the inherent capability of prototype-class pattern associations in a fast and straightforward manner without need of any iterative scheme subject to issues about convergence. Using an artificially designed data set we show how the proposed trained neural net classifies a test input pattern. Application to a few typical real-world data sets illustrate the overall network classification performance using different training and testing sample subsets generated randomly.

Conditional deletion of E11/Podoplanin in bone protects against ovariectomy-induced increases in osteoclast formation and activity

E11/Podoplanin (Pdpn) is implicated in early osteocytogenesis and the formation of osteocyte dendrites. This dendritic network is critical for bone modelling/remodelling, through the production of receptor activator of nuclear factor κ B (RANK)-ligand (RANKL). Despite this, the role of Pdpn in the control of bone remodelling is yet to be established in vivo. Here we utilised bone-specific Pdpn conditional knockout mice (cKO) to examine the role of Pdpn in the bone loss associated with ovariectomy (OVX). MicroCT revealed that Pdpn deletion had no significant effect on OVX-induced changes in trabecular microarchitecture. Significant differences between genotypes were observed in the trabecular pattern factor (P<0.01) and structure model index (P<0.01). Phalloidin staining of F-actin revealed OVX to induce alterations in osteocyte morphology in both wild-type (WT) and cKO mice. Histological analysis revealed an expected significant increase in osteoclast number in WT mice (P<0.01, compared with sham). However, cKO mice were protected against such increases in osteoclast number. Consistent with this, serum levels of the bone resorption marker Ctx were significantly increased in WT mice following OVX (P<0.05), but were unmodified by OVX in cKO mice. Gene expression of the bone remodelling markers Rank, Rankl, Opg and Sost were unaffected by Pdpn deletion. Together, our data suggest that an intact osteocyte dendritic network is required for sustaining osteoclast formation and activity in the oestrogen-depleted state, through mechanisms potentially independent of RANKL expression. This work will enable a greater understanding of the role of osteocytes in bone loss induced by oestrogen deprivation.

Endoplasmic reticulum mediates mitochondrial transfer within the osteocyte dendritic network

Mitochondrial transfer plays a crucial role in the regulation of tissue homeostasis and resistance to cancer chemotherapy. Osteocytes have interconnecting dendritic networks and are a model to investigate its mechanism. We have demonstrated, in primary murine osteocytes with photoactivatable mitochondria (PhAM)floxed and in MLO-Y4 cells, mitochondrial transfer in the dendritic networks visualized by high-resolution confocal imaging. Normal osteocytes transferred mitochondria to adjacent metabolically stressed osteocytes and restored their metabolic function. The coordinated movement and transfer of mitochondria within the dendritic network rely on contact between the endoplasmic reticulum (ER) and mitochondria. Mitofusin 2 (Mfn2), a GTPase that tethers ER to mitochondria, predominantly mediates the transfer. A decline in Mfn2 expression with age occurs concomitantly with both impaired mitochondrial distribution and transfer in the osteocyte dendritic network. These data show a previously unknown function of ER-mitochondrial contact in mediating mitochondrial transfer and provide a mechanism to explain the homeostasis of osteocytes.

Flow-wise or path-wise: diffusion in a fragmented dendritic network and implications for eels

The European eel (Anguilla anguilla) is a catadromous species that reproduces at sea and inhabits continental waters during its growth phase. River fragmentation due to obstacles is considered one cause of the decline of this species. However, the colonization process of river catchments by eels is still poorly understood. In this article, we compare two scenarios for the diffusion of eels within river catchments: a path-wise scenario, in which movements are totally random, and a flow-wise scenario, in which movements are partially oriented. Based on these two scenarios, we attempted to predict the distribution of eels within dendritic river catchments, explicitly accounting for the presence of obstacles to movement. The model was fitted to a long-time series of electro-fishing data. The results suggest that the path-wise scenario is more predominant than the flow-wise scenario. Moreover, results show that the distribution of eels in river catchment depends on (i) the types of movements carried out by eels, (ii) the configuration of river networks and (iii) the positions of obstacles within catchments.

Opposing Roles of Dendritic Cell Subsets in Experimental GN

Dendritic cells (DCs) are thought to form a dendritic network across barrier surfaces and throughout organs, including the kidney, to perform an important sentinel function. However, previous studies of DC function used markers, such as CD11c or CX3CR1, that are not unique to DCs. Here, we evaluated the role of DCs in renal inflammation using a CD11c reporter mouse line and two mouse lines with DC-specific reporters, Zbtb46-GFP and Snx22-GFP. Multiphoton microscopy of kidney sections confirmed that most of the dendritically shaped CD11c+ cells forming a network throughout the renal interstitium expressed macrophage-specific markers. In contrast, DCs marked by Zbtb46-GFP or Snx22-GFP were less abundant, concentrated around blood vessels, and round in shape. We confirmed this pattern of localization using imaging mass cytometry. Motility measurements showed that resident macrophages were sessile, whereas DCs were motile before and after inflammation. Although uninflamed glomeruli rarely contained DCs, injury with nephrotoxic antibodies resulted in accumulation of ZBTB46+ cells in the periglomerular region. ZBTB46 identifies all classic DCs, which can be categorized into two functional subsets that express either CD103 or CD11b. Depletion of ZBTB46+ cells attenuated the antibody-induced kidney injury, whereas deficiency of the CD103+ subset accelerated injury through a mechanism that involved increased neutrophil infiltration. RNA sequencing 7 days after nephrotoxic antibody injection showed that CD11b+ DCs expressed the neutrophil-attracting cytokine CXCL2, whereas CD103+ DCs expressed high levels of several anti-inflammatory genes. These results provide new insights into the distinct functions of the two major DC subsets in glomerular inflammation.

APC/CCdh1-Rock2 pathway controls dendritic integrity and memory

Disruption of neuronal morphology contributes to the pathology of neurodegenerative disorders such as Alzheimer’s disease (AD). However, the underlying molecular mechanisms are unknown. Here, we show that postnatal deletion of Cdh1, a cofactor of the anaphase-promoting complex/cyclosome (APC/C) ubiquitin ligase in neurons [Cdh1 conditional knockout (cKO)], disrupts dendrite arborization and causes dendritic spine and synapse loss in the cortex and hippocampus, concomitant with memory impairment and neurodegeneration, in adult mice. We found that the dendrite destabilizer Rho protein kinase 2 (Rock2), which accumulates in the brain of AD patients, is an APC/CCdh1 substrate in vivo and that Rock2 protein and activity increased in the cortex and hippocampus of Cdh1 cKO mice. In these animals, inhibition of Rock activity, using the clinically approved drug fasudil, prevented dendritic network disorganization, memory loss, and neurodegeneration. Thus, APC/CCdh1-mediated degradation of Rock2 maintains the dendritic network, memory formation, and neuronal survival, suggesting that pharmacological inhibition of aberrantly accumulated Rock2 may be a suitable therapeutic strategy against neurodegeneration.

The importance of metacommunity processes for long-term turnover of riffle-dwelling fish assemblages depends on spatial position within a dendritic network

Spatial position within a dendritic network may determine environmental filters and connectivity with source of immigrants, influencing species composition and variation in metacommunities. We investigated how long-term turnover of riffle-dwelling fish assemblages is affected by niche- and dispersal-related processes. Multiple linear model selection resulted in four variables important to explain assemblage turnover: (i) habitat change, (ii) channel slope, (iii) dlink (a proxy for connectivity), and (iv) catchment area. These four variables were related to turnover in species abundance, while turnover in species occurrence was only related to habitat change. Models generated for each species indicated they were differently affected by niche- and dispersal-related processes. Our findings provide support for the hypothesis that metacommunity processes affecting temporal turnover of stream fish communities are dependent on the spatial position within a dendritic network, but also vary among species. Our study has implications for sampling design in monitoring programs, as the degree of dissimilarities in fish communities depends not only on local habitat change, but on spatial position within the drainage network and on species particularities.