Alternative Architectures Research Articles

Enhancing Intracranial Hemorrhage Diagnosis through Deep Learning Models

Timely diagnosis is crucial for the successful treatment of a serious medical condition like brain hemorrhage. Deep learning algorithms have shown great promise in applications for medical image analysis, like the identification of brain hemorrhages. The goal of this study is to assess how well various deep learning algorithms can detect brain hemorrhages. Using a suitable dataset, the study evaluates the computational efficiency, accuracy, sensitivity, and specificity of the selected algorithms. The results demonstrate the potential of deep learning models to assist physicians in identifying this potentially fatal condition and demonstrate how well they can identify brain hemorrhages. The study’s findings improve automated brain hemorrhage detection technology, improving patient outcomes and the efficiency of healthcare delivery. EfficientNetB3 typically achieves higher accuracy due to its increased model complexity. Despite its increased complexity, EfficientNetB3 is still more parameter-efficient and computationally efficient than many alternative architectures. EfficientNetB3’s strong performance and feature extraction capabilities make it a good choice for transfer learning tasks. Out of all the models implemented in this paper, the proposed model with EfficientNetB3 gave the best accuracy for training as well as validation i.e. 99.95% and 93.29% respectively followed by EfficientNetB2, ResNet, SEResNext and ResNext.

The integration of heterogeneous resources in the CMS Submission Infrastructure for the LHC Run 3 and beyond

The integration of heterogeneous resources in the CMS Submission Infrastructure for the LHC Run 3 and beyond

Efficient prediction of anticancer peptides through deep learning.

Cancer remains one of the leading causes of mortality globally, with conventional chemotherapy often resulting in severe side effects and limited effectiveness. Recent advancements in bioinformatics and machine learning, particularly deep learning, offer promising new avenues for cancer treatment through the prediction and identification of anticancer peptides. This study aimed to develop and evaluate a deep learning model utilizing a two-dimensional convolutional neural network (2D CNN) to enhance the prediction accuracy of anticancer peptides, addressing the complexities and limitations of current prediction methods. A diverse dataset of peptide sequences with annotated anticancer activity labels was compiled from various public databases and experimental studies. The sequences were preprocessed and encoded using one-hot encoding and additional physicochemical properties. The 2D CNN model was trained and optimized using this dataset, with performance evaluated through metrics such as accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC-ROC). The proposed 2D CNN model achieved superior performance compared to existing methods, with an accuracy of 0.87, precision of 0.85, recall of 0.89, F1-score of 0.87, and an AUC-ROC value of 0.91. These results indicate the model's effectiveness in accurately predicting anticancer peptides and capturing intricate spatial patterns within peptide sequences. The findings demonstrate the potential of deep learning, specifically 2D CNNs, in advancing the prediction of anticancer peptides. The proposed model significantly improves prediction accuracy, offering a valuable tool for identifying effective peptide candidates for cancer treatment. Further research should focus on expanding the dataset, exploring alternative deep learning architectures, and validating the model's predictions through experimental studies. Efforts should also aim at optimizing computational efficiency and translating these predictions into clinical applications.

CAD-based Autonomous Vision Inspection Systems

Automated industrial Visual Inspection Systems (VIS) are typically customized for specific applications, limiting their flexibility. They are characterized by a demanding setup, high capital investments, and significant knowledge barriers. In this paper, we propose an alternative architecture for the visual inspection of 3D printed parts or complex assemblies using a robotic arm equipped with hand-eye sensors and controllable lighting system. The core of the proposed Flexible Vision Inspection System (FVIS) is the self-extraction of 3D text annotations from STandard for the Exchange of Product model (STEP) AP242 files. The system self-selects and parametrizes the most suitable inspection algorithm, including lighting settings. Additionally, it autonomously performs self-localization, self-referencing of physical products, and self-planning of robot inspection path based on CAD information. This framework, characterized by self-X, cost-effective, non-invasive, and plug-and-play architecture has the potential to disrupt the business model of vision inspection, enabling an as-a-service solution aligned with the next generation of flexible manufacturing.

Ingredient Inspired Recipe Recommender: Your Culinary Guide

Abstract: This research explores the development of an innovative culinary solution – the ingredient-Inspired Recipe Recommender. Focused on addressing the global challenge of food wastage, exacerbated by both consumer habits and quality deterioration, our study employs advanced deep learning architectures. A key aspect involves a detailed comparison of ResNet models (ResNet-50, ResNet-101, and ResNet-152) alongside alternative architectures like DenseNet, VGG, and xResNet. The objective is to identify the most effective neural network for accurately recognizing ingredients and generating insightful recipe recommendations. In response to the contemporary issues of rushed lifestyles, processed food reliance, and insufficient attention to nutrition, our research aims to empower individuals with a practical, technology-driven tool. By seamlessly integrating deep learning into the culinary landscape, the ingredient-Inspired Recipe Recommender suggests personalized recipes based on available ingredients, fostering healthier eating habits and contributing to a reduction in food wastage. This paper presents the methodology employed for the comparative analysis, reports experimental results, and discusses broader implications within the realms of food sustainability and technological innovation. In addressing the evolving needs of individuals, our research strives to align technology and gastronomy for positive environmental and societal impact.

Merging Minds and Machines: The Role of Advancing AI in Robotics

The relentless pursuit of creating intelligent robotic systems has led to a symbiotic relationship between human inventiveness and artificial intelligence (AI). Artificial intelligence is a theory. It is the development of computer systems that are able to perform tasks that would require human intelligence. This abstract explores the pivotal role that AI plays in advancing the capabilities and applications of robotic systems. The integration of AI algorithms and machine learning techniques has launched robotics beyond mere automation, enabling machines to modify, alter, adjust, learn, and interact with the world in ways previously deemed science fiction. Design fictions that vividly imagines future scenarios of AI or robotics in use offer a means both to explain and query the technological possibilities. Examples of these tasks are visual perception, speech recognition, decision-making, and translation between languages. The three key dimensions of AI’s role in robotics are Cognitive Augmentation, Human-Robot Collaboration, and Autonomous Intelligence. The abstract also discusses the societal implications of this AI-driven advancement in robotic systems, including ethical considerations, job market impacts, and the democratization of access to advanced technology. The convergence of human intellect and artificial intelligence in robotics marks a transformative era where machines become not just tools, but companions, collaborators, and cognitive extensions of human capabilities. Researchers are taking inspiration from the brain and considering alternative architectures in which networks of artificial neurons and synapses process information with high speed and adaptive learning capabilities in an energy-efficient, scalable manner. The indispensable role of AI in shaping the future of robotic systems and bridging the gap between human potential and machine capabilities is highlighted. The major impact of this synergy reverberates across industries, promising the world where robots become not just mechanical contraptions / defective apparatus but intelligent partners in our journey of progress.

Tensor-CSPNet: A Novel Geometric Deep Learning Framework for Motor Imagery Classification.

Deep learning (DL) has been widely investigated in a vast majority of applications in electroencephalography (EEG)-based brain-computer interfaces (BCIs), especially for motor imagery (MI) classification in the past five years. The mainstream DL methodology for the MI-EEG classification exploits the temporospatial patterns of EEG signals using convolutional neural networks (CNNs), which have been particularly successful in visual images. However, since the statistical characteristics of visual images depart radically from EEG signals, a natural question arises whether an alternative network architecture exists apart from CNNs. To address this question, we propose a novel geometric DL (GDL) framework called Tensor-CSPNet, which characterizes spatial covariance matrices derived from EEG signals on symmetric positive definite (SPD) manifolds and fully captures the temporospatiofrequency patterns using existing deep neural networks on SPD manifolds, integrating with experiences from many successful MI-EEG classifiers to optimize the framework. In the experiments, Tensor-CSPNet attains or slightly outperforms the current state-of-the-art performance on the cross-validation and holdout scenarios in two commonly used MI-EEG datasets. Moreover, the visualization and interpretability analyses also exhibit the validity of Tensor-CSPNet for the MI-EEG classification. To conclude, in this study, we provide a feasible answer to the question by generalizing the DL methodologies on SPD manifolds, which indicates the start of a specific GDL methodology for the MI-EEG classification.

Quantum logical controlled-NOT gate in a lithium niobate-on-insulator photonic quantum walk

The two-qubit controlled-NOT gate is one of the central entangling operations in quantum information technology. The controlled-NOT gate for single photon qubits is normally realized as a network of five individual beamsplitters on six optical modes. Quantum walks (QWs) are an alternative photonic architecture involving arrays of coupled waveguides, which have been successful for investigating condensed matter physics, however, have not yet been applied to quantum logical operations. Here, we engineer the tight-binding Hamiltonian of an array of lithium niobate-on-insulator waveguides to experimentally demonstrate the two-qubit controlled-NOT gate in a QW. We measure the two-qubit transfer matrix with 0.938 ± 0.003 fidelity, and we use the gate to generate entangled qubits with 0.945 ± 0.002 fidelity by preparing the control photon in a superposition state. Our results highlight a new application for QWs that use a compact multi-mode interaction region to realize large multi-component quantum circuits.

Candidate pathway association and genome-wide association approaches reveal alternative genetic architectures of carotenoid content in cultivated sunflower (Helianthus annuus).

The explosion of available genomic data poses significant opportunities and challenges for genome-wide association studies. Current approaches via linear mixed models (LMM) are straightforward but prevent flexible assumptions of an a priori genomic architecture, while Bayesian sparse LMMs (BSLMMs) allow this flexibility. Complex traits, such as specialized metabolites, are subject to various hierarchical effects, including gene regulation, enzyme efficiency, and the availability of reactants. To identify alternative genetic architectures, we examined the genetic architecture underlying the carotenoid content of an association mapping panel of Helianthus annuus individuals using multiple BSLMM and LMM frameworks. The LMMs of genome-wide single-nucleotide polymorphisms (SNPs) identified a single transcription factor responsible for the observed variations in the carotenoid content; however, a BSLMM of the SNPs with the bottom 1% of effect sizes from the results of the LMM identified multiple biologically relevant quantitative trait loci (QTLs) for carotenoid content external to the known (annotated) carotenoid pathway. A candidate pathway analysis (CPA) suggested a β-carotene isomerase to be the enzyme with the highest impact on the observed carotenoid content within the carotenoid pathway. While traditional LMM approaches suggested a single unknown transcription factor associated with carotenoid content variation in sunflower petals, BSLMM proposed several QTLs with interpretable biological relevance to this trait. In addition, the CPA allowed for the dissection of the regulatory vs. biosynthetic genetic architectures underlying this metabolic trait.

Radiation characteristics of an aerogel-supported fission fragment rocket engine for crewed interplanetary missions

The ionizing radiation properties of a fission fragment rocket engine concept are described in the context of a crewed Mars mission. This propulsion system could achieve very high specific impulses (>106 s) at a high power density (>kW/kg), utilizing micron-sized fissile fuel particles suspended in an aerogel matrix. The fission core is located within the bore of an electromagnet and external neutron moderator material. The low-density aerogel allows for radiative cooling of fuel particles while minimizing collisional losses with the fission fragments, leading to a more efficient use of fissile fuel in producing thrust compared to previous concepts. This paper presents the estimates of the steady-state ionizing radiation equivalent dose to the astronaut crew from both external (e.g., galactic cosmic rays) and internal (reactor) sources. The spacecraft design includes a centrifugation concept where the transit habitation module rotates around the spacecraft’s center of mass, providing artificial gravity to the crew and the separation distance to the nuclear core. We find that the fission fragment propulsion system combined with centrifugation could lead to reduced transit time, reduced equivalent radiation doses, and a reduced risk of long-term exposure to micro-g environments. Such a high-specific impulse propulsion system would enable other crewed fast transit, high delta-V interplanetary missions with payload mass fractions much greater than those of alternative propulsion architecture (chemical and solar electric).

Early symptom detection of basal stem rot disease in oil palm trees using a deep learning approach on UAV images

The oil palm tree (Elaeis guineensis Jacq.) is a vital and economically important plant, as the oil from its fruit is a primary export for many Southeast Asian countries. Hence the health of these trees is directly linked to their yield and the income of the growers. Unfortunately, basal stem rot (BSR) disease has been identified as the main threat to oil palm trees, causing infected trees to reduce their production and ultimately leading to death. While prior research has shown high accuracy in detecting oil palm trees, there has been limited focus on automating the observation of individual tree health status, particularly in identifying early symptoms of BSR disease within densely populated regions. This study introduces an enhanced approach for early symptom detection of BSR disease in densely populated oil palm tree areas. We present a modified U-Net architecture called the Multi-Convolution Residual (M-CR) U-Net, coupled with an image postprocessing method known as the overlapped contour separation (OVCS) algorithm. The M-CR U-Net differs from the original U-Net as it leverages on multiple convolution kernels of various sizes while incorporating skip connections. This design improvement results in superior segmentation performance compared to approaches utilizing fixed kernel sizes. Our proposed M-CR U-Net outperforms alternative architectures across a range of metrics, demonstrating improved training efficiency and delivering enhanced quantitative and qualitative segmentation outcomes. These achievements underscore its proficiency in accurately delineating the intricate features within input images. The M-CR U-Net excels in capturing the fine details while minimizing overlapping regions, as evidenced by its dice coefficient of 0.978, intersection over union of 0.871, precision of 0.989, and recall of 0.982. Additionally, the OVCS algorithm enables effective segmentation of individual oil palm trees within highly crowded regions. By adeptly separating the boundaries of overlapped areas, this algorithm enhances segmentation results. This innovation ensures precise identification of the health and boundaries of oil palm trees, leading to an average performance improvement of 2.716% across various evaluation metrics. The proposed methodology demonstrates significant potential in detecting early symptoms of BSR disease in oil palm trees, even within crowded regions, leading to more effective management of oil palm plantations.

Genome-Wide Expression Analysis Unravels Fluoroalkane Metabolism in Pseudomonas sp. Strain 273.

Pseudomonas sp. strain 273 grows with medium-chain terminally fluorinated alkanes under oxic conditions, releases fluoride, and synthesizes long-chain fluorofatty acids. To shed light on the genes involved in fluoroalkane metabolism, genome, and transcriptome sequencing of strain 273 grown with 1,10-difluorodecane (DFD), decane, and acetate were performed. Strain 273 harbors three genes encoding putative alkane monooxygenases (AlkB), key enzymes for initiating alkane degradation. Transcripts of alkB-2 were significantly more abundant in both decane- and DFD-grown cells compared to acetate-grown cells, suggesting AlkB-2 catalyzes the attack on terminal CH3 and CH2F groups. Coordinately expressed with alkB-2 was an adjacent gene encoding a fused ferredoxin-ferredoxin reductase (Fd-Fdr). Phylogenetic analysis distinguished AlkB that couples with fused Fd-Fdr reductases from AlkB with alternate architectures. A gene cluster containing an (S)-2-haloacid dehalogenase (had) gene was up-regulated in cells grown with DFD, suggesting a possible role in the removal of the ω-fluorine. Genes involved in long-chain fatty acid biosynthesis were not differentially expressed during growth with acetate, decane, or DFD, suggesting the bacterium's biosynthetic machinery does not discriminate against monofluoro-fatty acid intermediates. The analysis sheds first light on genes and catalysts involved in the microbial metabolism of fluoroalkanes.

Dual input neural networks for positional sound source localization

In many signal processing applications, metadata may be advantageously used in conjunction with a high dimensional signal to produce a desired output. In the case of classical Sound Source Localization (SSL) algorithms, information from a high dimensional, multichannel audio signals received by many distributed microphones is combined with information describing acoustic properties of the scene, such as the microphones’ coordinates in space, to estimate the position of a sound source. We introduce Dual Input Neural Networks (DI-NNs) as a simple and effective way to model these two data types in a neural network. We train and evaluate our proposed DI-NN on scenarios of varying difficulty and realism and compare it against an alternative architecture, a classical Least-Squares (LS) method as well as a classical Convolutional Recurrent Neural Network (CRNN). Our results show that the DI-NN significantly outperforms the baselines, achieving a five times lower localization error than the LS method and two times lower than the CRNN in a test dataset of real recordings.

A Multilayer Perceptron Learning Architecture for Black-Scholes Call Options Pricing

This work explores whether an artificial neural network trained with synthetic call option pricing data can perform well on semi-real market data. The approach that is followed is based on a multilayer perceptron using a large synthetic dataset for training and a semi-real dataset from market data simulating top ten S&P 100 stocks. The baseline for errors and price estimations is the Black-Sholes model. Results, demonstrate the overall capability of the network to learn the Black-Scholes function using synthetic data, and estimate call option prices with high level of accuracy, competitively to Black-Sholes model. This is a valuable result for practitioners who might want to use machine learning paradigms for option pricing in various assets and markets. Future work will include feature selection and advanced sampling for training, plus alternative network architectures and further experimentation with hyperparameters. The most important outcome from the work is that machine learning models can be used from practitioners as an alternative method for option pricing.

A Study on Thermal Management Systems for Hybrid–Electric Aircraft

The electrification of an aircraft’s propulsive system is identified as a potential solution towards a lower carbon footprint in the aviation industry. One of the effects of increased electrification is the generation of a large amount of waste heat that needs to be removed. As high-power systems must be cooled to avoid performance deterioration such as battery thermal runaway, a suitable thermal management system is required to regulate the temperature of the powertrain components. With this in mind, the main objective of this research is to identify promising heat transfer technologies to be integrated into a thermal management system (TMS) such that power, mass, and drag can be minimised for a parallel hybrid–electric regional aircraft in the context of the EU-funded FutPrInt50 project. Five different TMS architectures are modelled using the Matlab/Simulink environment based on thermodynamic principles, heat transfer fundamentals, and fluid flow equations. The systems are a combination of a closed-loop liquid cooling integrated with different heat dissipation components, namely ram air heat exchanger, skin heat exchanger, and fuel. Their cooling capacity and overall aircraft performance penalties under different flight conditions are estimated and compared to each other. Then, a parametric study is conducted, followed by a multi-objective optimisation analysis with the aim of minimising the TMS impact. As expected, none of the investigated architectures exhibit an ideal performance across the range of the studied metrics. The research revealed that, while planning the TMS for future hybrid–electric aircraft, alternative architectures will have to be developed and studied in light of the power requirements.

Graph-based learning for automated code checking – Exploring the application of graph neural networks for design review

Although automated code checking (ACC) has been a subject of interest for many years, we have not yet seen significant breakthroughs in the field that may lead to the development of generic, comprehensive tools for ACC. Hard-coded rules are the backbone of all emerging platforms for ACC. These rules require a significant amount of engineering, which often requires manual labor; and the resulting rule sets are strict and difficult to scale to other building models. On the other hand, approaches relying purely on classic machine learning (e.g. SVM) are too coarse and unable to accurately express building information. In our hope to come up with a more scalable solution, we investigate here a novel workflow that relies on graph-based learning algorithms instead of processing rule sets. We illustrate the suggested workflow by checking accessibility requirements in residential houses, which we believe is one of the more promising rule sets that can be checked using graph-based learning methods. The high accuracy of the obtained results is encouraging to continue exploring Graph Neural Networks (GNN) for this type of ACC, yet rule-based and classic ML-based approaches show other advantages as well (rigor and speed, respectively). The main contribution of this work is therefore its identification of meaningful limitations and directions for future research, including alternative graph structures and GNN architectures.

Context-Aware Expert for Software Architecture Recovery (CAESAR): An automated approach for recovering software architectures

Software architecture recovery approaches help to reconstruct the architecture of complex software systems. However, the manual effort and expertise required to use such techniques are high, as the user demands detailed knowledge about software architecture recovery process and preparing additional software artifacts to use as input for such approaches. Such practices lack consideration for the context standards of the recovered software. Additionally, proper module detection rules that are aware of the modeled system’s context should be studied to enhance the cluster quality of the detected modules. We introduce a novel approach to software architecture recovery by implementing an automated knowledge-based rules solution that considers the context of the developed software while identifying and clustering that system’s software modules. Such solution is aided by a novel code distance model that is utilized to enhance the cohesiveness of the recovered software modules, improving the recovered software architecture quality, and enhancing readability through providing visualizations at different abstraction levels for the recovered architecture. The proposed method has been evaluated on two software systems with varying sizes and from different contexts and involved experienced participants. The results demonstrate significant improvements in cohesion and coupling of the produced clusters compared to alternative architecture recovery approaches.

Implementation of fuzzy associative memory toward optimizing a neural network model to predict total iron binding capacity

ObjectivesDetermining the optimal number of hidden layer neurons in a neural network (NN) is an important issue. In this research, a fuzzy associative memory (FAM) model will be developed to determine the hidden neuron of neural network architecture. MethodsThe material for this research is data from chronic kidney disease (CKD) patients on hemodialysis to predict total iron binding capacity (TIBC). The research begins by choosing the most appropriate backpropagation algorithm. The FAM system was developed using the product-correlation coding operator, the product-max composition relation, and a combination of winner-take-all and weighted average. Testing is done by calculating the average error (MSE) of training and testing, as well as the average error for gradients. ResultsWe selected 12 alternative network architectures. The Levenberg-Marquardt backpropagation algorithm was chosen for its best performance. In testing the training data, the average percentage of errors was 7.16% for MSE and 12.93% for gradients. The test results of the FAM system show an average error of 0%. The test results on the test data show an average error for MSE of 7.67% and an average error for gradients of 8.55%. ConclusionBesides being able to predict the MSE and gradient for a particular NN architecture, FAM can also determine the NN architectural model that provides the optimal combination of MSE and gradient. Further research will improve the performance of the FAM system.

Alternative Formulations of Decision Rule Learning from Neural Networks

This paper extends recent work on decision rule learning from neural networks for tabular data classification. We propose alternative formulations to trainable Boolean logic operators as neurons with continuous weights, including trainable NAND neurons. These alternative formulations provide uniform treatments to different trainable logic neurons so that they can be uniformly trained, which enables, for example, the direct application of existing sparsity-promoting neural net training techniques like reweighted L1 regularization to derive sparse networks that translate to simpler rules. In addition, we present an alternative network architecture based on trainable NAND neurons by applying De Morgan’s law to realize a NAND-NAND network instead of an AND-OR network, both of which can be readily mapped to decision rule sets. Our experimental results show that these alternative formulations can also generate accurate decision rule sets that achieve state-of-the-art performance in terms of accuracy in tabular learning applications.

Near-optimal Multiplier-Less Broadband Noise Shaping Filters

Noise shaping (NS) filters reduce the quantization noise power in one or more frequency band(s) while amplifying it in other bands. Narrow-band noise shaping filters are state of the art in audio signal processing, analog-digital and digital-analog conversion, direct digital synthesis, and other applications. However, it is much more difficult to design broadband NS filters. Since NS filters are used in feedback branches, they must therefore be designed direct path free which imposes a constraint on the filter coefficients. This constraint leads to prohibitive large filter coefficients employing state of the art filter design techniques.This paper investigates the theoretical bound for NS filters and shows results about a novel design method for broadband FIR and IIR noise shaping filters and its multiplier-less hardware implementation. The method employed is a purely numerical approximation technique and leads to filter designs close to the discussed theoretical bound. The quantization of the filter coefficients is performed by a Canonical Signed Digit (CSD) representation of the coefficients. Two alternative architectures for the implementation of the filters are discussed. The design technique and the CSD quantization are realized in a MATLAB toolbox. The filters were moreover implemented in VHDL.