Neural Approach Research Articles

LuckyMera: a modular AI framework for building hybrid NetHack agents

In the last few decades we have witnessed a significant development in Artificial Intelligence (AI) thanks to the availability of a variety of testbeds, mostly based on simulated environments and video games. Among those, roguelike games offer a very good trade-off in terms of complexity of the environment and computational costs, which makes them perfectly suited to test AI agents generalization capabilities. In this work, we present LuckyMera, a flexible, modular, extensible and configurable AI framework built around NetHack, a popular terminal-based, single-player roguelike video game. This library is aimed at simplifying and speeding up the development of AI agents capable of successfully playing the game and offering a high-level interface for designing game strategies. LuckyMera comes with a set of off-the-shelf symbolic and neural modules (called "skills"): these modules can be either hard-coded behaviors, or neural Reinforcement Learning approaches, with the possibility of creating compositional hybrid solutions. Additionally, LuckyMera comes with a set of utility features to save its experiences in the form of trajectories for further analysis and to use them as datasets to train neural modules, with a direct interface to the NetHack Learning Environment and MiniHack. Through an empirical evaluation we validate our skills implementation and propose a strong baseline agent that can reach state-of-the-art performances in the complete NetHack game. LuckyMera is open-source and available at https://github.com/Pervasive-AI-Lab/LuckyMera.

Variability of bio-optical properties of Sundarbans mangrove estuarine ecosystem using elemental analysis, Sentinel 3 OLCI imageries and neural network models

Coastal water bodies face significant contamination risks stemming from various factors, such as growing populations, rapid urban development, dam construction, heightened industrial operations (notably in ports), and increasing tourism activities. The release of untreated sewage from municipalities, industrial discharges, and various recreational and commercial activities along the coast significantly deteriorates water quality and presents a serious risk to marine ecosystems, food chains, and human health. Hence, continual monitoring of coastal water quality is essential for the ecological sustainability of marine and aquatic ecosystems. This study examines the spatiotemporal variability of optical characteristics within the Sundarbans estuary system, a transboundary ecosystem shared by India and Bangladesh. The study employs Sentinel-3 OLCI satellite data alongside in-situ measurements of critical water quality parameters, including chlorophyll-a (Chl-a), total suspended matter (TSM), and chromophoric dissolved organic matter (CDOM), collected from 100 sampling sites across the Sundarbans estuarine system during the pre-monsoon, monsoon, and post-monsoon periods of 2021 and 2022.This study employs the sophisticated C2RCC model, a machine learning-based neural network-driven approach recognized for its robustness and precision. It examines the various bio-optical properties in tropical estuarine environments. The C2RCC model addresses this challenge directly. This advanced processor, powered by neural network technology, effectively interprets the complex optical signals found in Case-2 waters, offering essential insights into their quality. The findings indicate the highest TSM levels during the pre-monsoon seasons of both years. Additionally, the estimated Chl-a concentrations indicated a rise during the monsoon season (0.03–4 mg m3), whereas the intensified pre-monsoon CDOM levels showed consistency across both years. The strong positive correlation observed between in-situ and satellite data, with an average R2 value ranging from 0.90 to 0.99, underscores the reliability of the high-resolution remote sensing approach. The study illustrates the efficacy of OLCI data in observing near-shore and coastal ecosystems, clearly exceeding conventional methods for evaluating Chl-a, TSM, and CDOM. This study underscores the potential of high spatial and spectral resolution of Sentinel-3 Ocean color data for effective monitoring and management of the complex estuarine ecosystem of the Sundarbans. The findings of this study provide important insights to inform the development of sustainable coastal water resource management strategies. Furthermore, the findings will be crucial in achieving Sustainable Development Goal 14, which focuses on life below water, particularly in the conservation and management of marine and coastal biodiversity.

Bridging Hydrological Ensemble Simulation and Learning Using Deep Neural Operators

AbstractEnsemble‐based simulation and learning (ESnL) has long been used in hydrology for parameter inference, but computational demands of process‐based ESnL can be quite high. To address this issue, we propose a deep neural operator learning approach. Neural operators are generic machine learning algorithms that can learn functional mappings between infinite‐dimensional spaces, providing a highly flexible tool for scientific machine learning. Our approach is built upon DeepONet, a specific deep neural operator, and is designed to address several common problems in hydrology, namely, model parameter estimation, prediction at ungaged locations, and uncertainty quantification. Here we demonstrate the effectiveness of our DeepONet‐based workflow using an existing large model ensemble created for an eastern U.S. watershed that is instrumented with 10 streamflow gages. Results suggest DeepONet achieves high efficiency in learning an ML surrogate model from the model ensemble, with the modified Kling‐Gupta Efficiency exceeding 0.9 on holdout test sets. Parameter inference, carried out using the trained DeepONet surrogate model and genetic algorithm, also yields robust results. Additionally, we formulate and train a separate DeepONet model for physics‐informed, seq‐to‐seq streamflow forecasting, which further reduces biases in the pre‐trained DeepONet surrogate model. While this study focuses primarily on a single watershed, our approach is general and may be extended to enable learning from model ensembles across multiple basins or models. Thus, this research represents a significant contribution to the application of hybrid machine learning in hydrology.

Label distribution-driven multi-view representation learning

In multi-view representation learning (MVRL), the challenge of category uncertainty is significant. Existing methods excel at deriving shared representations across multiple views, but often neglect the uncertainty associated with cluster assignments from each view, thereby leading to increased ambiguity in the category determination. Additionally, methods like kernel-based or neural network-based approaches, while revealing nonlinear relationships, lack attention to category uncertainty. To address these limitations, this paper proposes a method leveraging the uncertainty of label distributions to enhance MVRL. Specifically, our approach combines uncertainty reduction based on label distribution with view representation learning to improve clustering accuracy and robustness. It initially computes the within-view representation of the sample and semantic labels. Then, we introduce a novel constraint based on either variance or information entropy to mitigate class uncertainty, thereby improving the discriminative power of the learned representations. Extensive experiments conducted on diverse multi-view datasets demonstrate that our method consistently outperforms existing approaches, producing more accurate and reliable class assignments. The experimental results highlight the effectiveness of our method in enhancing MVRL by reducing category uncertainty and improving overall classification performance. This method is not only very interpretable but also enhances the model’s ability to learn multi-view consistent information.

Investigating the impact of system parameters on flow-induced vibration hard galloping based on deep neural network

Abstract In this paper, a classification model is established for the flow-induced vibration response based on the numerical and experimental data, using a deep neural network-based machine learning approach. The model effectively distinguishes between hard galloping and soft galloping in flow-induced vibrations by identifying the corresponding range of system parameters. Moreover, a regression model is established to determine the relationship between the critical reduced velocity of hard galloping and system parameters, then an exploratory function strategy is utilized to establish the functional relationship between the critical reduced velocity of the hard galloping and the system parameters. The results reveal that the system parameter range with the occurrence of hard galloping is fn<0.85Uζ>-0.1fn+0.19. Additionally, the functional relationship between the critical reduced velocity and system parameters facilitates the adjustment of vibration states in flow-induced vibrations and enables deeper investigation into the phenomenon of hard galloping.

An Improved Predefined-Time Adaptive Neural Control Approach for Nonlinear Multiagent Systems

An Improved Predefined-Time Adaptive Neural Control Approach for Nonlinear Multiagent Systems

Enhancing Worker Recruitment in Collaborative Mobile Crowdsourcing: A Graph Neural Network Trust Evaluation Approach

Enhancing Worker Recruitment in Collaborative Mobile Crowdsourcing: A Graph Neural Network Trust Evaluation Approach

An Accurate Deep Neural Network-Based Inversion Approach for Multichannel Microwave Interferometer Characterization of Plasma Wake Electron Density Distribution

An Accurate Deep Neural Network-Based Inversion Approach for Multichannel Microwave Interferometer Characterization of Plasma Wake Electron Density Distribution

Comparative Analysis of the Methods of Planning and Coordinating of Manipulator Robot Movement

This paper presents a comparative analysis of two methods for planning and coordinating the movement of robot manipulators in dynamic environments: a neural network-based approach for solving dynamic production scenarios and the rapidly exploring random trees algorithm. The study aims to enhance the trajectory planning of robot manipulators by leveraging the strengths of intelligent systems. The neural network method is designed to perceive the environment, generate accurate control commands, and adapt to changing conditions in real-time. The paper the processes involved in environmental analysis, collision avoidance, and control signal generation for actuators, with an emphasis on the neural network architecture tailored for these tasks. The results demonstrate that the neural network approach offers significant improvements in adaptability and efficiency, providing a robust solution for optimizing automated processes in dynamic production environments.

Neutral Benchmarking of Survival Models in Health Sciences: Comparative Study of Classical and Machine Learning Techniques

Survival analysis plays a central role in diverse research fields, especially in health sciences. As an analytical tool, it can be used to help improve patients’ survival time, or at least, reduce the prospects of recurrence in cancer studies. However, approaches to the predictive performance of the current survival models mainly center on clinical data along with the classical survival methods. For censored “omics” data, the performance of survival models has not been thoroughly studied, either often due to their high dimensionality issues or reliance on binarizing the survival time for classification analysis. We aim to present a neural benchmark approach that analyzes and compares a broad range of classical and state-of-the-art machine learning survival models for “omics” and clinical datasets. All the methods considered in our study are evaluated using predictability as a performance measure. The study is systematically designed to make 36 comparisons (9 methods over 4 datasets, i.e., 2 clinical and 2 omics), and shows that, in practice, predictability of survival models does vary across real-world datasets, model choice, as well as the evaluation metric. From our results, we emphasize that performance criteria can play a key role in a balanced assessment of diverse survival models. Moreover, the Multitask Logistic Regression (MTLR) showed remarkable predictability for almost all the datasets. We believe this outstanding performance presents a unique opportunity for a wider use of MTLR for survival risk factors. For translational clinicians and scientists, we hope our findings provide practical guidance for benchmark studies of survival models, as well as highlight potential areas of research interest.

Chemical characterization of hibiscus rosa-sinensis plant fibers facilitated through design of experiments and artificial neural network hybrid approach

The integration of natural fibers into Fiber Reinforced Polymers (FRPs) has emerged as a promising avenue for sustainable and high-performance composite materials. Natural fibers, derived from plants, offer notable advantages such as renewability, low cost, and environmental friendliness. Among these natural fibers, Hibiscus Rosa-Sinensis (HRS) plant fibers have gained significant attention owing to their widespread availability and unique mechanical properties. In this study, HRS fibers were chemically treated using Sodium Hydroxide (NaOH), Potassium Permanganate (KMnO4), and Acetic Acid (CH3COOH) at different weight percentages (3, 4, 5 Wt.%) and solutionizing times (1, 2, 3 h) based on Taguchi’s L27 orthogonal array. The fibers, extracted from epidermis of the stems, underwent cleaning and chemical treatment after water retting. The crystallinity index, determined via X-ray Diffraction (XRD), indicated a maximum value of 65.77%. Thermo-gravimetric analysis (TGA) exhibited a degradation temperature of 365.24 °C and a material loss of 63.11%. Potassium Permanganate treatment at 4 Wt.% and 3 h of solutionizing time has yielded the best results. Multi-Layer Perceptron Artificial Neural Network (MLP-ANN) has been successfully applied to accurately predict the output physical characteristics of chemically treated HRS fibers using experimental data. The results are in close alignment with the literature. Scanning Electron Microscopy (SEM) and Energy Dispersive X-ray Spectroscopy (EDS) analyses have provided valuable insights into the microstructure and constituents of the chemically treated HRS fibers. This research emphasises on the effectiveness of the chemical treatment process in enhancing the properties of HRS plant fibers for potential composite applications.

K-plex-based community detection with graph neural networks

Community detection is an effective way to determine the structure and characteristics of a complex network. With the expansion of the network scale, traditional community detection approaches such as modularity-based optimization models face new challenges related to representing and learning the topological structure and node attributes from a large scale complex network. Moreover, in many practical applications, there is little knowledge about label information or the number of communities, which greatly limits the performance of existing supervised or semi-supervised community detection approaches. To solve these problems, in this paper, we propose a graph neural network-based unsupervised community detection approach, which first applies the k-plex to generate the community seeds, then uses a node sampling algorithm to reduce the network complexity, and finally constructs a graph neural network model to learn the relationships of the network nodes and assign the nodes to different communities. Extensive empirical studies on various scale networks demonstrate both the effectiveness and efficiency of the proposed approach. Our codes are available at https://github.com/lol12854/KPGN.

A combined neural ODE-Bayesian optimization approach to resolve dynamics and estimate parameters for a modified SIR model with immune memory

We propose a novel hybrid approach that integrates Neural Ordinary Differential Equations (NODEs) with Bayesian optimization to address the dynamics and parameter estimation of a modified time-delay-type Susceptible-Infected-Removed (SIR) model incorporating immune memory. This approach leverages a neural network to produce continuous multi-wave infection profiles by learning from both data and the model. The time-delay component of the SIR model, expressed through a convolutional integral, results in an integro-differential equation. To resolve these dynamics, we extend the NODE framework, employing a Runge-Kutta solver, to handle the challenging convolution integral, enabling us to fit the data and learn the parameters and dynamics of the model. Additionally, through Bayesian optimization, we enhance prediction accuracy while focusing on long-term dynamics. Our model, applied to COVID-19 data from Mexico, South Africa, and South Korea, effectively learns critical time-dependent parameters and provides accurate short- and long-term predictions. This combined methodology allows for early prediction of infection peaks, offering significant lead time for public health responses.

Hybrid artificial neural networks-based approaches predicting the heat transfer and flow characteristics of manifold microchannels

Aiming at meeting the requirement of high heat flux electronics cooling, the geometric complexity of manifold microchannels as an energy-saving technique has been increasing rapidly. It is well worth exploring optimal prediction models for the heat transfer and flow characteristics of manifold microchannels to reduce computing resources and time costs in the optimization process. In this paper, hybrid prediction approaches based on artificial neural networks are proposed for manifold microchannels. Four optimization algorithms, namely, genetic algorithm, particle swarm optimization algorithm, artificial hummingbird algorithm, and zebra optimization algorithm were adopted to enhance the prediction performance. Initially, all prediction models were trained and tested by a numerical database including six input parameters, and the thermal resistance and pumping power were taken as the target outputs. Compared to the original method, hybrid prediction models show higher accuracy and reliability on the numerical data with regression coefficients beyond 0.987. The prediction model optimized by the artificial hummingbird algorithm shows the best performance with the mean average error for thermal resistance and pumping power decreased by 88.7% and 81.4% respectively. Additionally, the runtime of all prediction models is compared and the model optimized by the zebra optimization algorithm takes the longest time, but is still two orders of magnitude shorter than that of the traditional simulation method. Finally, the generalizability of all prediction models is further validated by a database amassed from six experimental studies on manifold microchannels. The results showed that three hybrid models can provide accurate outputs with regression coefficients beyond 0.84 which differs from the poor prediction of the original method. The model optimized by the zebra optimization algorithm shows superior prediction performance for six individual studies, two fluids, three fin shapes, and wide flow rate conditions. The trained hybrid prediction models in this study perform higher accuracy and can be cost-effective and reliable prediction tools for the rapid optimization process of various manifold microchannel applications.

Research Progresses and Applications of Knowledge Graph Embedding Technique in Chemistry.

A knowledge graph (KG) is a technique for modeling entities and their interrelations. Knowledge graph embedding (KGE) translates these entities and relationships into a continuous vector space to facilitate dense and efficient representations. In the domain of chemistry, applying KG and KGE techniques integrates heterogeneous chemical information into a coherent and user-friendly framework, enhances the representation of chemical data features, and is beneficial for downstream tasks, such as chemical property prediction. This paper begins with a comprehensive review of classical and contemporary KGE methodologies, including distance-based models, semantic matching models, and neural network-based approaches. We then catalogue the primary databases employed in chemistry and biochemistry that furnish the KGs with essential chemical data. Subsequently, we explore the latest applications of KG and KGE in chemistry, focusing on risk assessment, property prediction, and drug discovery. Finally, we discuss the current challenges to KG and KGE techniques and provide a perspective on their potential future developments.

Quantum computations on a new neural network potential for the proton-bound noble-gas Ar[formula omitted]H[formula omitted] complex: Isotopic effects

High-quality data-driven potentials were developed aiming to predict rovibrational traits and analyze the influence of the isotopic substitution on the molecular spectroscopic properties of Ar2H+. Neural networks machine-learning approaches trained on CCSD(T)/CBS datasets were implemented. Our full-dimensional quantum MCTDH results were discussed in comparison with experimental data in gas phase and solid matrix environments, as well as against theoretical estimates available. The new data indicate that both fundamental and progression bands are dominantly driven by the strength and shape of the underlying interactions. Our simulations could enable the spectroscopic characterization of these species, assisting investigations for their astrophysical observation.

A stochastic approach for elliptic problems in perforated domains

A wide range of applications in science and engineering involve a PDE model in a domain with perforations, such as perforated metals or air filters. Solving such perforated domain problems suffers from computational challenges related to resolving the scale imposed by the geometries of perforations. We propose a neural network-based mesh-free approach for perforated domain problems. The method is robust and efficient in capturing various configuration scales, including the averaged macroscopic behavior of the solution that involves a multiscale nature induced by small perforations. The new approach incorporates the derivative-free loss method that uses a stochastic representation or the Feynman-Kac formulation. In particular, we implement the Neumann boundary condition for the derivative-free loss method to handle the interface between the domain and perforations. A suite of stringent numerical tests is provided to support the proposed method's efficacy in handling various perforation scales.

Learning to quantify uncertainty in off-target activity for CRISPR guide RNAs.

CRISPR-based genome editing technologies have revolutionised the field of molecular biology, offering unprecedented opportunities for precise genetic manipulation. However, off-target effects remain a significant challenge, potentially leading to unintended consequences and limiting the applicability of CRISPR-based genome editing technologies in clinical settings. Current literature predominantly focuses on point predictions for off-target activity, which may not fully capture the range of possible outcomes and associated risks. Here, we present crispAI, a neural network architecture-based approach for predicting uncertainty estimates for off-target cleavage activity, providing a more comprehensive risk assessment and facilitating improved decision-making in single guide RNA (sgRNA) design. Our approach makes use of the count noise model Zero Inflated Negative Binomial (ZINB) to model the uncertainty in the off-target cleavage activity data. In addition, we present the first-of-its-kind genome-wide sgRNA efficiency score, crispAI-aggregate, enabling prioritization among sgRNAs with similar point aggregate predictions by providing richer information compared to existing aggregate scores. We show that uncertainty estimates of our approach are calibrated and its predictive performance is superior to the state-of-the-art in silico off-target cleavage activity prediction methods. The tool and the trained models are available at https://github.com/furkanozdenn/crispr-offtarget-uncertainty.

Physics-informed probabilistic deep network with interpretable mechanism for trustworthy mechanical fault diagnosis

The application of data-driven models based on the neural network is pivotal to developing an intelligent fault diagnostic flowchart. However, the reliability and interpretability of these models for fault prediction have presented challenges in neural network-based diagnostic approach. Another challenge is data availability, which has also been a limiting factor in using artificial Intelligence for condition monitoring and fault assessment in industrial mechanical systems. Consequently, this study proposes a Physics Informed Probabilistic Deep Network (PIPDN) framework to overcome these challenges. The PIPDN comprises two main components: the physical labelling module, designed to enhance physical labels with mechanical failure information, and the main body of PIPDN, responsible for learning fault representative features and generating smart data guided by conditional and physical labels. Furthermore, a multi-scale PIPDN model is developed to integrate the proposed uncertainty quantification (UQ) with decision-fusion module for accurate interpretation and enhanced fault diagnosis. The applicability, effectiveness, and superiority of the proposed framework and approach are validated using an experimental bearing dataset. The results indicate that integrating physical labels significantly assists the PIPDN model in capturing more accurate fault characteristics. This increases the importance of latent space features for subsequent fault diagnosis and also enhances the diagnostic interpretability. Furthermore, the addition of UQ-based decision-making module improves the reliability of the MS-PIPDN model by reducing epistemic uncertainty in the predictions.

Modelling of Joule-Thomson cooling effect using a modified shift-DeepONet method for predicting hydrate onset during CO2 sequestration

Carbon Storage in underground depleted oil and gas reservoirs play a key role in reducing anthropogenic Carbon Dioxide (CO2) emissions around the world. Injection of CO2 at high pressure into a low pressure depleted gas formations containing methane (CH4) leads to Joule-Thompson (J-T) cooling, possibly resulting in injection blocking precipitation of CO2 or CH4 hydrates. Here, we present a deep neural operator-based approach to model the J-T effects during CO2 injection in depleted reservoirs to demonstrate the effectiveness of Deep Operator Networks (DeepONets) in modelling piece-wise functions with sharp changes in gradient for cases of varying permeabilities. We propose a modified version of the shift-DeepONet model using non-linear constructors in the neural network architecture to accurately model temperature and pressure profiles in space and time. The modified shift-DeepONet model achieves a higher accuracy in modelling temperature for permeability in the range of 1.01mD to 101mD with three to up to ten times improvement in reducing Relative L2 (RL2) norm and with a testing RL2 norm of 3.8 × 10−4 compared to 2.97 × 10−3 for the Vanilla DeepONet. Using the temperature and pressure profiles, estimates are made for hydrate onset in space and time. The proposed modified shift-DeepONet method achieves high accuracy in hydrate classification with precision values > 0.99, recall values > 0.86 and F1 scores >0.92 in classification of CO2 or CH4 hydrate onset during CO2 injection into depleted gas reservoirs.