MFAN: Multi-Level Features Attention Network for Fake Certificate Image Detection

Up to now, most of the forensics methods have attached more attention to natural content images. To expand the application of image forensics technology, forgery detection for certificate images that can directly represent people’s rights and interests is investigated in this paper. Variable tampered region scales and diverse manipulation types are two typical characteristics in fake certificate images. To tackle this task, a novel method called Multi-level Feature Attention Network (MFAN) is proposed. MFAN is built following the encoder–decoder network structure. In order to extract features with rich scale information in the encoder, on the one hand, we employ Atrous Spatial Pyramid Pooling (ASPP) on the final layer of a pre-trained residual network to capture the contextual information at different scales; on the other hand, low-level features are concatenated to ensure the sensibility to small targets. Furthermore, the resulting multi-level features are recalibrated on channels for irrelevant information suppression and enhancing the tampered regions, guiding the MFAN to adapt to diverse manipulation traces. In the decoder module, the attentive feature maps are convoluted and unsampled to effectively generate the prediction mask. Experimental results indicate that the proposed method outperforms some state-of-the-art forensics methods.

A DEEP DIVE INTO THE BASICS OF DEEP LEARNING

Deep learning is a type of machine learning (ML) that is growing in importance in the medical field. It can often perform better than traditional ML models on different metrics, and it can handle non-linear problems due to activation functions. Activation functions are different non-linear functions that are used to restrict the values propagated to an interval. In deep learning, information propagates forward, passing through different layers of weights and activation functions, before reaching the final layer. Then a cost function is evaluated and propagated back through the network to adjust weights. A convolutional neural network (CNN) is a form of deep learning that is used primarily in imaging. CNNs perform significantly well with grid-like inputs because they learn shapes well. CNNs compute dot products between layers and kernels in a convolutional layer, prior to pooling, which outputs summary statistics. CNNs are better than trivial neural networks for imaging due to a number of reasons, like sparse interaction and equivariance of translation

Crosslinked layered surfaces of heparin and poly(L-lysine) enhance mesenchymal stromal cells behavior in the presence of soluble interferon gamma

Human mesenchymal stromal cells (hMSCs) are multipotent cells that have been proposed for the treatment of immune-mediated diseases. Culturing hMSCs on tissue culture plastic reduces their therapeutic potential in part due to the lack of extracellular matrix components. The aim of this study is to evaluate multilayers of heparin and poly(L-lysine) (HEP/PLL) as a bioactive surface for hMSCs stimulated with soluble interferon gamma (IFN‐γ). Multilayers were formed, via layer-by-layer assembly, with HEP as the final layer and supplemented with IFN-γ in the culture medium. Multilayer construction and chemistry were confirmed using Azure A staining, quartz crystal microbalance (QCM), and X-ray photoelectron spectroscopy. hMSCs adhesion, viability, and differentiation, were assessed. Results showed that (HEP/PLL) multilayer coatings were poorly adhesive for hMSCs. However, performing chemical crosslinking using 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide and N-hydroxysuccinimide (EDC/NHS) significantly enhanced hMSCs adhesion and viability. The immunosuppressive properties of hMSCs cultured on crosslinked (HEP/PLL) multilayers were confirmed by measuring the level of indoleamine 2,3-dioxygenase (IDO) secretion. Lastly, hMSCs cultured on crosslinked (HEP/PLL) multilayers in the presence of soluble IFN- γ successfully differentiated towards the osteogenic and adipogenic lineages as confirmed by Alizarin red, and oil-red O staining, as well as alkaline phosphatase activity. This study suggests that crosslinked (HEP/PLL) films can modulate hMSCs response to soluble factors, which may improve hMSCs-based therapies aimed at treating several immune diseases.

A SIMULATION STUDY ON THE EFFECT OF PARTICLE SIZE DISTRIBUTION ON THE PRINTED GEOMETRY IN SELECTIVE LASER MELTING

Selective laser melting (SLM) process is a powder bed fusion additive manufacturing process that finds applications in aerospace and medical industries for its ability to produce complex geometry parts. As the raw material used is in powder form, particle size distribution (PSD) is a significant characteristic that influences the build quality in turn affecting the functionality and aesthetics aspects of the product. This paper investigates the effect of PSD on the printed geometry for 316L stainless steel powder, where three coupled in-house simulation tools based on Discrete Element Method (DEM), Computational Fluid Dynamics (CFD), and Structural Mechanics are employed. DEM is used for simulating the powder bed distribution based on the different powder PSD. The CFD is used as a virtual testbed to determine thermal parameters such as heat capacity and thermal conductivity of the powder bed viewed as a continuum. The values found as a stochastic function of the powder distribution is used to analyse the effect on the melted zone and deformation using Structural Mechanics. Results showed that mean particle size and PSD had a significant effect on the packing density, melt pool layer thickness, and the final layer thickness after deformation. Specifically, a narrow particle size distribution with smaller mean particle size and standard deviation produced solidified final layer thickness closest to nominal layer thickness. The proposed simulation approach and the results will catalyze in development of geometry assurance strategies to minimize the effect of particle size distribution on the geometric quality of the printed part.

Using Neural Networks in Order to Analyze Telework Adaptability across the European Union Countries: A Case Study of the Most Relevant Scenarios to Occur in Romania

Our study addresses the issue of telework adoption by countries in the European Union and draws up a few feasible scenarios aimed at improving telework’s degree of adaptability in Romania. We employed the dataset from the 2020 Eurofound survey on Living, Working and COVID-19 (Round 2) in order to extract ten relevant determinants of teleworking on the basis of 24,123 valid answers provided by respondents aged 18 and over: the availability of work equipment; the degree of satisfaction with the experience of working from home; the risks related to potential contamination with SARS-CoV-2 virus; the employees’ openness to adhering to working-from-home patterns; the possibility of maintaining work–life balance objectives while teleworking; the level of satisfaction on the amount and the quality of work submitted, etc. Our methodology entailed the employment of SAS Enterprise Guide software to perform a cluster analysis resulting in a preliminary classification of the EU countries with respect to the degree that they have been able to adapt to telework. Further on, in order to refine this taxonomy, a multilayer perceptron neural network with ten input variables in the initial layer, six neurons in the intermediate layer, and three neurons in the final layer was successfully trained. The results of our research demonstrate the existence of significant disparities in terms of telework adaptability, such as: low to moderate levels of adaptability (detected in countries such as Greece, Croatia, Portugal, Spain, Lithuania, Latvia, Poland, Italy); fair levels of adaptability (encountered in France, Slovakia, the Czech Republic, Hungary, Slovenia, or Romania); and high levels of adaptability (exhibited by intensely digitalized economies such Denmark, Sweden, Finland, Germany, Ireland, the Netherlands, Belgium, etc.).

Refined Continuous Control of DDPG Actors via Parametrised Activation

Continuous action spaces impose a serious challenge for reinforcement learning agents. While several off-policy reinforcement learning algorithms provide a universal solution to continuous control problems, the real challenge lies in the fact that different actuators feature different response functions due to wear and tear (in mechanical systems) and fatigue (in biomechanical systems). In this paper, we propose enhancing the actor-critic reinforcement learning agents by parameterising the final layer in the actor network. This layer produces the actions to accommodate the behaviour discrepancy of different actuators under different load conditions during interaction with the environment. To achieve this, the actor is trained to learn the tuning parameter controlling the activation layer (e.g., Tanh and Sigmoid). The learned parameters are then used to create tailored activation functions for each actuator. We ran experiments on three OpenAI Gym environments, i.e., Pendulum-v0, LunarLanderContinuous-v2, and BipedalWalker-v2. Results showed an average of 23.15% and 33.80% increase in total episode reward of the LunarLanderContinuous-v2 and BipedalWalker-v2 environments, respectively. There was no apparent improvement in Pendulum-v0 environment but the proposed method produces a more stable actuation signal compared to the state-of-the-art method. The proposed method allows the reinforcement learning actor to produce more robust actions that accommodate the discrepancy in the actuators’ response functions. This is particularly useful for real life scenarios where actuators exhibit different response functions depending on the load and the interaction with the environment. This also simplifies the transfer learning problem by fine-tuning the parameterised activation layers instead of retraining the entire policy every time an actuator is replaced. Finally, the proposed method would allow better accommodation to biological actuators (e.g., muscles) in biomechanical systems.

A New Framework for Matching Forensic Composite Sketches With Digital Images

Face sketch recognition is considered as a sub-problem of face recognition. Matching composite sketches with its corresponding digital image is one of the challenging tasks. A new convolution neural network (CNN) framework for matching composite sketches with digital images is proposed in this work. The framework consists of a base CNN model that uses swish activation function in the hidden layers. Both composite sketches and digital images are trained separately in the network by providing matching pairs and mismatching pairs. The final output resulted from the network's final layer is compared with the threshold value, and then the pair is assigned to the same or different class. The proposed framework is evaluated on two datasets, and it exhibits an accuracy of 78.26% with extended-PRIP (E-PRIP) and 69.57% with composite sketches with age variations (CSA) respectively. Experimental analysis shows the improved results compared to state-of-the-art composite sketch matching systems.

Convolutional neural networks reveal topic-level representations of sentences in medial prefrontal cortex, lateral anterior temporal lobe, precuneus, and angular gyrus

When reading a sentence, individual words can be combined to create more complex meaning. In this study, we sought to uncover brain regions that reflect the representation of meaning at the sentence level, as opposed to only the meaning of their individual constituent words. Using fMRI, we recorded the neural activity of participants while reading sentences. We constructed sentence topic-level representations using the final layer of a convolutional neural network (CNN) trained to classify Wikipedia sentences into broad semantic categories. This model was contrasted with word-level sentence representations constructed using the average of the word embeddings constituting the sentence. Using representational similarity analysis, we found that the medial prefrontal cortex, lateral anterior temporal lobe, precuneus, and angular gyrus more strongly represent sentence topic-level, compared to word-level, meaning, uncovering the important role of these semantic system regions in the representation of integrated meaning. Conversely, these results validate the capacity of CNNs to capture human sentence-level representations.

Topological Uncertainty: Monitoring Trained Neural Networks through Persistence of Activation Graphs

Although neural networks are capable of reaching astonishing performance on a wide variety of contexts, properly training networks on complicated tasks requires expertise and can be expensive from a computational perspective. In industrial applications, data coming from an open-world setting might widely differ from the benchmark datasets on which a network was trained. Being able to monitor the presence of such variations without retraining the network is of crucial importance. In this paper, we develop a method to monitor trained neural networks based on the topological properties of their activation graphs. To each new observation, we assign a Topological Uncertainty, a score that aims to assess the reliability of the predictions by investigating the whole network instead of its final layer only as typically done by practitioners. Our approach entirely works at a post-training level and does not require any assumption on the network architecture, optimization scheme, nor the use of data augmentation or auxiliary datasets; and can be faithfully applied on a large range of network architectures and data types. We showcase experimentally the potential of Topological Uncertainty in the context of trained network selection, Out-Of-Distribution detection, and shift-detection, both on synthetic and real datasets of images and graphs.

Predictive and discriminative localization of pathology using high resolution class activation maps with CNNs

Purpose Existing class activation mapping (CAM) techniques extract the feature maps only from a single layer of the convolutional neural net (CNN), generally from the final layer and then interpolate to upsample to the original image resolution to locate the discriminative regions. Consequently these provide a coarse localization that may not be able to capture subtle abnormalities in medical images. To alleviate this, our work proposes a technique called high resolution class activation mapping (HR-CAMs) that can provide enhanced visual explainability to the CNN models. Methods HR-CAMs fuse feature maps by training a network using the input from multiple layers of a trained CNN, thus gaining information from every layer that can localize abnormalities with greater details in original image resolution. The technique is validated qualitatively and quantitatively on a simulated dataset of 8,000 images followed by applications on multiple image analysis tasks that include (1) skin lesion classification (ISIC open dataset—25,331 cases) and (2) predicting bone fractures (MURA open dataset—40,561 images) (3) predicting Parkinson’s disease (PD) from neuromelanin sensitive MRI (small cohort-80 subjects). Results We demonstrate that our model creates clinically interpretable subject specific high resolution discriminative localizations when compared to widely used CAMs and Gradient-CAMs. Conclusion HR-CAMs provide finer delineation of abnormalities thus facilitating superior explainability to CNNs as has been demonstrated from its rigorous validation.