Deep learning for computer vision: a comparison between convolutional neural networks and hierarchical temporal memories on object recognition tasks

In recent years, Deep Learning techniques have shown to perform well on a large variety of problems both in Computer Vision and Natural Language Processing, reaching and often surpassing the state of the art on many tasks. The rise of deep learning is also revolutionizing the entire field of Machine Learning and Pattern Recognition pushing forward the concepts of automatic feature extraction and unsupervised learning in general. However, despite the strong success both in science and business, deep learning has its own limitations. It is often questioned if such techniques are only some kind of brute-force statistical approaches and if they can only work in the context of High Performance Computing with tons of data. Another important question is whether they are really biologically inspired, as claimed in certain cases, and if they can scale well in terms of "intelligence". The dissertation is focused on trying to answer these key questions in the context of Computer Vision and, in particular, Object Recognition, a task that has been heavily revolutionized by recent advances in the field. Practically speaking, these answers are based on an exhaustive comparison between two, very different, deep learning techniques on the aforementioned task: Convolutional Neural Network (CNN) and Hierarchical Temporal memory (HTM). They stand for two different approaches and points of view within the big hat of deep learning and are the best choices to understand and point out strengths and weaknesses of each of them. CNN is considered one of the most classic and powerful supervised methods used today in machine learning and pattern recognition, especially in object recognition. CNNs are well received and accepted by the scientific community and are already deployed in large corporation like Google and Facebook for solving face recognition and image auto-tagging problems. HTM, on the other hand, is known as a new emerging paradigm and a new meanly-unsupervised method, that is more biologically inspired. It tries to gain more insights from the computational neuroscience community in order to incorporate concepts like time, context and attention during the learning process which are typical of the human brain. In the end, the thesis is supposed to prove that in certain cases, with a lower quantity of data, HTM can outperform CNN.

Oxidative stress and inflammation response after nanoparticle exposure: differences between human lung cell monocultures and an advanced three-dimensional model of the human epithelial airways

Combustion-derived and manufactured nanoparticles (NPs) are known to provoke oxidative stress and inflammatory responses in human lung cells; therefore, they play an important role during the development of adverse health effects. As the lungs are composed of more than 40 different cell types, it is of particular interest to perform toxicological studies with co-cultures systems, rather than with monocultures of only one cell type, to gain a better understanding of complex cellular reactions upon exposure to toxic substances. Monocultures of A549 human epithelial lung cells, human monocyte-derived macrophages and monocyte-derived dendritic cells (MDDCs) as well as triple cell co-cultures consisting of all three cell types were exposed to combustion-derived NPs (diesel exhaust particles) and to manufactured NPs (titanium dioxide and single-walled carbon nanotubes). The penetration of particles into cells was analysed by transmission electron microscopy. The amount of intracellular reactive oxygen species (ROS), the total antioxidant capacity (TAC) and the production of tumour necrosis factor (TNF)-alpha and interleukin (IL)-8 were quantified. The results of the monocultures were summed with an adjustment for the number of each single cell type in the triple cell co-culture. All three particle types were found in all cell and culture types. The production of ROS was induced by all particle types in all cell cultures except in monocultures of MDDCs. The TAC and the (pro-)inflammatory reactions were not statistically significantly increased by particle exposure in any of the cell cultures. Interestingly, in the triple cell co-cultures, the TAC and IL-8 concentrations were lower and the TNF-alpha concentrations were higher than the expected values calculated from the monocultures. The interplay of different lung cell types seems to substantially modulate the oxidative stress and the inflammatory responses after NP exposure.

Numerical model of the human cornea based on stromal microstructure: identification of mechanical properties for two age groups

The optical quality of the human eye mainly depends on the refractive performance of the cornea. The shape of the cornea is a mechanical balance between intraocular pressure and tissue intrinsic stiffness. Several surgical procedures in ophthalmology alter the biomechanics of the cornea to provoke local or global curvature changes for vision correction. Legitimated by the large number of surgical interventions performed every day, the demand for a deeper understanding of corneal biomechanics is rising to improve the safety of procedures and medical devices. The aim of our work is to propose a numerical model of corneal biomechanics, based on the stromal microstructure. Our novel anisotropic constitutive material law features a probabilistic weighting approach to model collagen fiber distribution as observed on human cornea by Xray scattering analysis (Aghamohammadzadeh et. al., Structure, February 2004). Furthermore, collagen cross-linking was explicitly included in the strain energy function. Results showed that the proposed model is able to successfully reproduce both inflation and extensiometry experimental data (Elsheikh et. al., Curr Eye Res, 2007; Elsheikh et. al., Exp Eye Res, May 2008). In addition, the mechanical properties calculated for patients of different age groups (Group A: 65-79 years; Group B: 80-95 years) demonstrate an increased collagen cross-linking, and a decrease in collagen fiber elasticity from younger to older specimen. These findings correspond to what is known about maturing fibrous biological tissue. Since the presented model can handle different loading situations and includes the anisotropic distribution of collagen fibers, it has the potential to simulate clinical procedures involving nonsymmetrical tissue interventions. In the future, such mechanical model can be used to improve surgical planning and the design of next generation ophthalmic devices.

Biomechanical effects of environmental and engineered particles on human airway smooth muscle cells

The past decade has seen significant increases in combustion-generated ambient particles, which contain a nanosized fraction (less than 100 nm), and even greater increases have occurred in engineered nanoparticles (NPs) propelled by the booming nanotechnology industry. Although inhalation of these particulates has become a public health concern, human health effects and mechanisms of action for NPs are not well understood. Focusing on the human airway smooth muscle cell, here we show that the cellular mechanical function is altered by particulate exposure in a manner that is dependent upon particle material, size and dose. We used Alamar Blue assay to measure cell viability and optical magnetic twisting cytometry to measure cell stiffness and agonist-induced contractility. The eight particle species fell into four categories, based on their respective effect on cell viability and on mechanical function. Cell viability was impaired and cell contractility was decreased by (i) zinc oxide (40-100 nm and less than 44 microm) and copper(II) oxide (less than 50 nm); cell contractility was decreased by (ii) fluorescent polystyrene spheres (40 nm), increased by (iii) welding fumes and unchanged by (iv) diesel exhaust particles, titanium dioxide (25 nm) and copper(II) oxide (less than 5 microm), although in none of these cases was cell viability impaired. Treatment with hydrogen peroxide up to 500 microM did not alter viability or cell mechanics, suggesting that the particle effects are unlikely to be mediated by particle-generated reactive oxygen species. Our results highlight the susceptibility of cellular mechanical function to particulate exposures and suggest that direct exposure of the airway smooth muscle cells to particulates may initiate or aggravate respiratory diseases.

Properties of low-threshold motor axons in the human median nerve

This study investigated the excitability and accommodative properties of low-threshold human motor axons to test whether these motor axons have greater expression of the persistent Na(+) conductance, I(NaP). Computer-controlled threshold tracking was used to study 22 single motor units and the data were compared with compound motor potentials of various amplitudes recorded in the same experimental session. Detailed comparisons were made between the single units and compound potentials that were 40% or 5% of maximal amplitude, the former because this is the compound potential size used in most threshold tracking studies of axonal excitability, the latter because this is the compound potential most likely to be composed entirely of motor axons with low thresholds to electrical recruitment. Measurements were made of the strength-duration relationship, threshold electrotonus, current-voltage relationship, recovery cycle and latent addition. The findings did not support a difference in I(NaP). Instead they pointed to greater activity of the hyperpolarization-activated inwardly rectifying current (I(h)) as the basis for low threshold to electrical recruitment in human motor axons. Computer modelling confirmed this finding, with a doubling of the hyperpolarization-activated conductance proving the best single parameter adjustment to fit the experimental data. We suggest that the hyperpolarization-activated cyclic nucleotide-gated (HCN) channel(s) expressed on human motor axons may be active at rest and contribute to resting membrane potential.

Mesh-based vs. Image-based Statistical Appearance Model of the Human Femur: a Preliminary Comparison Study for the Creation of Finite Element Meshes

Statistical models have been recently introduced in computational orthopaedics to investigate the bone mechanical properties across several populations. A fundamental aspect for the construction of statistical models concerns the establishment of accurate anatomical correspondences among the objects of the training dataset. Various methods have been proposed to solve this problem such as mesh morphing or image registration algorithms. The objective of this study is to compare a mesh-based and an image-based statistical appearance model approaches for the creation of nite element(FE) meshes. A computer tomography (CT) dataset of 157 human left femurs was used for the comparison. For each approach, 30 finite element meshes were generated with the models. The quality of the obtained FE meshes was evaluated in terms of volume, size and shape of the elements. Results showed that the quality of the meshes obtained with the image-based approach was higher than the quality of the mesh-based approach. Future studies are required to evaluate the impact of this finding on the final mechanical simulations.

Ultrasound-guided percutaneous tracheal puncture: a computer-tomographic controlled study in cadavers

Ultrasound-guided techniques are increasingly used in anaesthetic practice to identify tissues beneath the skin and to increase the accuracy of placement of needles close to targeted structures. To examine ultrasound's usefulness for dilatational tracheostomy, we performed ultrasound-guided tracheal punctures in human cadavers followed by computer-tomographic (CT) control.

The interface between morphology and action planning: a comparison of two species of New World monkeys

Recent research with several species of nonhuman primates suggests sophisticated motor-planning abilities observed in human adults may be ubiquitous among primates. However, there is considerable variability in the extent to which these abilities are expressed across primate species. In the present experiment, we explore whether the variability in the expression of anticipatory motor-planning abilities may be attributed to cognitive differences (such as tool use abilities) or whether they may be due to the consequences of morphological differences (such as being able to deploy a precision grasp). We compared two species of New World monkeys that differ in their tool use abilities and manual dexterity: squirrel monkeys, Saimiri sciureus (less dexterous with little evidence for tool use) and tufted capuchins, Sapajus apella (more dexterous and known tool users). The monkeys were presented with baited cups in an untrained food extraction task. Consistent with the morphological constraint hypothesis, squirrel monkeys frequently showed second-order motor planning by inverting their grasp when picking up an inverted cup, while capuchins frequently deployed canonical upright grasping postures. Findings suggest that the lack of ability for precision grasping may elicit more consistent second-order motor planning, as the squirrel monkeys (and other species that have shown a high rate of second-order planning) have fewer means of compensating for inefficient initial postures. Thus, the interface between morphology and motor planning likely represents an important factor for understanding both the ontogenetic and phylogenetic origins of sophisticated motor-planning abilities.

The Interface Between Morphology and Action Planning: a Comparison of Two Species of New World Monkeys

Recent research with several species of nonhuman primates suggests sophisticated motor-planning abilities observed in human adults may be ubiquitous among primates. However, there is considerable variability in the extent to which these abilities are expressed across primate species. In the present experiment, we explore whether the variability in the expression of anticipatory motor-planning abilities may be attributed to cognitive differences (such as tool use abilities) or whether they may be due to the consequences of morphological differences (such as being able to deploy a precision grasp). We compared two species of New World monkeys that differ in their tool use abilities and manual dexterity: squirrel monkeys, Saimiri sciureus (less dexterous with little evidence for tool use) and tufted capuchins, Sapajus apella (more dexterous and known tool users). The monkeys were presented with baited cups in an untrained food extraction task. Consistent with the morphological constraint hypothesis, squirrel monkeys frequently showed second-order motor planning by inverting their grasp when picking up an inverted cup, while capuchins frequently deployed canonical upright grasping postures. Findings suggest that the lack of ability for precision grasping may elicit more consistent second-order motor planning, as the squirrel monkeys (and other species that have shown a high rate of second-order planning) have fewer means of compensating for inefficient initial postures. Thus, the interface between morphology and motor planning likely represents an important factor for understanding both the ontogenetic and phylogenetic origins of sophisticated motor-planning abilities. (C) 2013 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.

Feature Versus Spatial Strategies by Orangutans (Pongo abelii) and Human Children (Homo sapiens) in a Cross-Dimensional Task

Despite the fact that photographic stimuli are used across experimental contexts with both human and nonhuman subjects, the nature of individuals' perceptions of these stimuli is still not well understood. In the present experiments, we tested whether three orangutans and 36 human children could use photographic information presented on a computer screen to solve a perceptually corresponding problem in the physical domain. Furthermore, we tested the cues that aided in this process by pitting featural information against spatial position in a series of probe trials. We found that many of the children and one orangutan were successfully able to use the information cross-dimensionally; however, the other two orangutans and almost a quarter of the children failed to acquire the task. Species differences emerged with respect to ease of task acquisition. More striking, however, were the differences in cues that participants used to solve the task: Whereas the orangutan used a spatial strategy, the majority of children used a feature one. Possible reasons for these differences are discussed from both evolutionary and developmental perspectives. The novel results found here underscore the need for further testing in this area to design appropriate experimental paradigms in future comparative research settings.

A novel interface for hybrid mock circulations to evaluate ventricular assist devices

This paper presents a novel mock circulation for the evaluation of ventricular assist devices (VADs), which is based on a hardware-in-the-loop concept. A numerical model of the human blood circulation runs in real time and computes instantaneous pressure, volume, and flow rate values. The VAD to be tested is connected to a numerical-hydraulic interface, which allows the interaction between the VAD and the numerical model of the circulation. The numerical-hydraulic interface consists of two pressure-controlled reservoirs, which apply the computed pressure values from the model to the VAD, and a flow probe to feed the resulting VAD flow rate back to the model. Experimental results are provided to show the proper interaction between a numerical model of the circulation and a mixed-flow blood pump.

Surfactant protein D modulates allergen particle uptake and inflammatory response in a human epithelial airway model

Background Allergen-containing subpollen particles (SPP) are released from whole plant pollen upon contact with water or even high humidity. Because of their size SPP can preferentially reach the lower airways where they come into contact with surfactant protein (SP)-D. The aim of the present study was to investigate the influence of SP-D in a complex three-dimensional human epithelial airway model, which simulates the most important barrier functions of the epithelial airway. The uptake of SPP as well as the secretion of pro-inflammatory cytokines was investigated. Methods SPP were isolated from timothy grass and subsequently fluorescently labeled. A human epithelial airway model was built by using human Type II-pneumocyte like cells (A549 cells), human monocyte derived macrophages as well as human monocyte derived dendritic cells. The epithelial cell model was incubated with SPP in the presence and absence of surfactant protein D. Particle uptake was evaluated by confocal microscopy and advanced computer-controlled analysis. Finally, human primary CD4+ T-Cells were added to the epithelial airway model and soluble mediators were measured by enzyme linked immunosorbent assay or bead array. Results SPP were taken up by epithelial cells, macrophages, and dendritic cells. This uptake coincided with secretion of pro-inflammatory cytokines and chemokines. SP-D modulated the uptake of SPP in a cell type specific way (e.g. increased number of macrophages and epithelial cells, which participated in allergen particle uptake) and led to a decreased secretion of pro-inflammatory cytokines. Conclusion These results display a possible mechanism of how SP-D can modulate the inflammatory response to inhaled allergen.

Effects of flame made zinc oxide particles in human lung cells - a comparison of aerosol and suspension exposures

Background Predominantly, studies of nanoparticle (NPs) toxicology in vitro are based upon the exposure of submerged cell cultures to particle suspensions. Such an approach however, does not reflect particle inhalation. As a more realistic simulation of such a scenario, efforts were made towards direct delivery of aerosols to air-liquid-interface cultivated cell cultures by the use of aerosol exposure systems. This study aims to provide a direct comparison of the effects of zinc oxide (ZnO) NPs when delivered as either an aerosol, or in suspension to a triple cell co-culture model of the epithelial airway barrier. To ensure dose–equivalence, ZnO-deposition was determined in each exposure scenario by atomic absorption spectroscopy. Biological endpoints being investigated after 4 or 24h incubation include cytotoxicity, total reduced glutathione, induction of antioxidative genes such as heme-oxygenase 1 (HO–1) as well as the release of the (pro)-inflammatory cytokine TNFα. Results Off-gases released as by-product of flame ZnO synthesis caused a significant decrease of total reduced GSH and induced further the release of the cytokine TNFα, demonstrating the influence of the gas phase on aerosol toxicology. No direct effects could be attributed to ZnO particles. By performing suspension exposure to avoid the factor “flame-gases”, particle specific effects become apparent. Other parameters such as LDH and HO–1 were not influenced by gaseous compounds: Following aerosol exposure, LDH levels appeared elevated at both timepoints and the HO–1 transcript correlated positively with deposited ZnO-dose. Under submerged conditions, the HO–1 induction scheme deviated for 4 and 24h and increased extracellular LDH was found following 24h exposure. Conclusion In the current study, aerosol and suspension-exposure has been compared by exposing cell cultures to equivalent amounts of ZnO. Both exposure strategies differ fundamentally in their dose–response pattern. Additional differences can be found for the factor time: In the aerosol scenario, parameters tend to their maximum already after 4h of exposure, whereas under submerged conditions, effects appear most pronounced mainly after 24h. Aerosol exposure provides information about the synergistic interplay of gaseous and particulate phase of an aerosol in the context of inhalation toxicology. Exposure to suspensions represents a valuable complementary method and allows investigations on particle-associated toxicity by excluding all gas–derived effects.

Investigation of Human-Structure Interaction Through Experimental and Analytical Studies

Vibration serviceability is a widely recognized design criterion for assembly-type structures, such as stadiums, that are likely subjected to rhythmic human-induced excitation. Human-induced excitation of a structure occurs from the movement of the occupants such as walking, running, jumping, or dancing. Vibration serviceability is based on the level of comfort that people have with the vibrations of a structure. Current design guidance uses the natural frequency of the structure to assess vibration serviceability. However, a phenomenon known as human-structure interaction suggests that there is a dynamic interaction between the structure and passive occupants, altering the natural frequency of the system. Human-structure interaction is dependent on many factors, including the dynamic properties of the structure, posture of the occupants, and relative size of the crowd. It is unknown if the shift in natural frequency due to humanstructure interaction is significant enough to warrant consideration in the design process. This study explores the interface of both structural and crowd characteristics through experimental testing to determine if human-structure interaction should be considered because of its potential impact on serviceability assessment. An experimental test structure that represents the dynamic properties of a cantilevered stadium structure was designed and constructed. Experimental modal analysis was implemented to determine the dynamic properties of the empty test structure and when occupied with up to seven people arranged in different locations and postures. Comparisons of the dynamic properties were made between the empty and occupied testing configurations and analytical results from the use of a dynamic crowd model recommended from the Joint Working Group of Europe. Data trends lead to the development of a refined dynamic crowd model. This dynamic model can be used in conjunction with a finite element model of the test structure to estimate the dynamic influence due to human-structure interaction due to occupants standing with straight knees. In the future, the crowd model will be refined and can aid in assessing the dynamic properties of in-service stadium structures.