Article
Computer Science, Artificial Intelligence
Khabat Soltanian, Ali Ebnenasir, Mohsen Afsharchi
Summary: This article presents a novel method called Modular Grammatical Evolution (MGE) that aims to validate the hypothesis that restricting the solution space of NeuroEvolution to modular and simple neural networks allows for the efficient generation of smaller, more structured networks with acceptable (and sometimes superior) accuracy on large datasets. MGE improves upon state-of-the-art Grammatical Evolution (GE) methods by introducing a modular representation and mitigating scalability and locality issues. Experimental results demonstrate that modularity helps in finding better neural networks faster and MGE outperforms other GE methods in terms of locality and scalability.
EVOLUTIONARY COMPUTATION
(2022)
Article
Energy & Fuels
Songda Wang, Tomislav Dragicevic, Yuan Gao, Remus Teodorescu
Summary: Modular multilevel converter (MMC) has been popular for its advantages in harmonics reduction and efficiency improvement, and model predictive control (MPC) based controllers are widely used. However, the computational burden of MPC limits the control implementation of MMC. To address this, machine learning (ML) based controllers, specifically neural network (NN) regression, have been designed and shown to have better control performance and lower computation burden compared to NN pattern recognition.
IEEE TRANSACTIONS ON ENERGY CONVERSION
(2021)
Article
Computer Science, Artificial Intelligence
Tomasz Praczyk
Summary: This paper presents a modular neuro-evolutionary controller for guiding underwater vehicles to move along a desired trajectory and maintain a certain distance from the sea bottom in an underwater environment. The controller outperforms traditional controllers in simulation tests.
APPLIED SOFT COMPUTING
(2022)
Article
Multidisciplinary Sciences
N. Deshpandea Jhelam, Emanuel A. Fronhofer
Summary: This study investigates the impact of genetic architecture on the dynamics and predictability of invasion into an environmental gradient. The results show that in a gene regulatory network model, range expansions are accelerating and less predictable, primarily driven by an increase in the rate of local adaptation to novel habitats.
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
(2022)
Article
Computer Science, Artificial Intelligence
Min Shi, Yufei Tang, Xingquan Zhu, Yu Huang, David Wilson, Yuan Zhuang, Jianxun Liu
Summary: Neural architecture search (NAS) has gained significant attention in computational intelligence research, but there is limited research on Graph Neural Network (GNN) models for unstructured network data. This paper proposes a novel framework that evolves individual models in a large GNN architecture search space to dynamically approach the optimal fit. Experimental results show that evolutionary NAS matches state-of-the-art reinforcement learning methods for graph representation learning and node classification.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Automation & Control Systems
Junting Fei, Qingxuan Jia, Gang Chen, Tong Li, Ruiquan Wang, Xiaodong Zhang
Summary: This paper proposes a genetic algorithm-based optimal design strategy for modular robot topology. It uses four tuples to represent the topology and introduces a distributed parallel kinematic modeling and analysis method. An optimization model of the topology design is established, and the genetic algorithm is used to solve it. Simulations with a modular robot as an example show that the designed topology allows the robot to successfully execute tasks with fewer modular units and significantly reduces the computing time for kinematics modeling.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2023)
Article
Physics, Fluids & Plasmas
Vaibhav Mohanty, Ard A. Louis
Summary: Investigated the robustness of spin glasses in the glassy phase, finding high robustness and topological properties that scale similarly to other systems.
Article
Computer Science, Artificial Intelligence
Tomasz Praczyk
Summary: This paper introduces a novel generative Neuro-Evolutionary method called Hill Climb Modular Assembler Encoding (HCMAE) for evolving modular Artificial Neural Networks (ANNs). By testing different variants on two ANN benchmarks, the effectiveness of the HCMAE method is demonstrated.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Automation & Control Systems
Deepan Muthirayan, Pramod P. Khargonekar
Summary: In this article, a novel control architecture is proposed for adaptive control of continuous-time systems using inspiration from neuroscience. The architecture augments an external working memory to a standard neural network (NN)-based adaptive controller. The controller writes the hidden layer feature vector of the NN to the external working memory and can update this information with the observed error in the output. Memory augmentation significantly improves learning, as shown through extensive simulations and specific metrics.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL
(2023)
Article
Ecology
Dina Navon, Paul Hatini, Lily Zogbaum, R. Craig Albertson
Summary: The research found that plasticity is generally integrated across a range of ecologically relevant traits, with traits that have overlapping functions responding in a coordinated manner. Genetic data suggest a compromise between global genetic regulators and integration across traits for plasticity.
Article
Computer Science, Software Engineering
Yuan Liu, Peng Wang, Cheng Lin, Xiaoxiao Long, Jiepeng Wang, Lingjie Liu, Taku Komura, Wenping Wang
Summary: NeRO is a neural rendering-based method that reconstructs the geometry and BRDF of reflective objects from multiview images captured in an unknown environment. It uses a two-step approach to accurately reconstruct the object's geometry and then recover the environment lights and BRDF.
ACM TRANSACTIONS ON GRAPHICS
(2023)
Article
Medicine, Research & Experimental
Guiyang Zhang, Qiang Tang, Pengmian Feng, Wei Chen
Summary: In this study, an attention-based bidirectional gated recurrent unit network called IPs-GRUAtt was proposed to identify phosphorylation sites in SARS-CoV-2-infected host cells. Comparative results showed that IPs-GRUAtt outperformed state-of-the-art machine-learning methods and existing models in identifying phosphorylation sites. Moreover, the attention mechanism allowed IPs-GRUAtt to extract key features from protein sequences. These results demonstrate that IPs-GRUAtt is a powerful tool for identifying phosphorylation sites.
MOLECULAR THERAPY-NUCLEIC ACIDS
(2023)
Article
Computer Science, Artificial Intelligence
Ali Gholami-Rahimabadi, Hadi Razmi, Hasan Doagou-Mojarrad
Summary: The paper introduces a multiple-deme parallel genetic algorithm for load shedding control to prevent voltage collapse and instability. A modular neural network method is implemented to estimate the voltage stability margin index, and a simultaneous equilibrium tracing technique is employed to consider the detailed model of generator components. Test results on the New England-39 bus test system show the efficiency of the proposed method.
Article
Multidisciplinary Sciences
Oystein H. Opedal, W. Scott Armbruster, Thomas F. Hansen, Agnes Holstad, Christophe Pelabon, Stefan Andersson, Diane R. Campbell, Christina M. Caruso, Lynda F. Delph, Christopher G. Eckert, Asa Lankinen, Greg M. Walter, Jon Agren, Geir H. Bolstad
Summary: Understanding the causes and limits of population divergence in phenotypic traits is important for evolutionary biology and can predict adaptation to environmental change. This study analyzed a large database of plant populations and found that evolutionary divergence scaled positively with genetic variability within populations. Additionally, vegetative traits showed greater divergence compared to reproductive traits. These results suggest that there is predictability and genetic constraints in trait divergence.
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
(2022)
Article
Computer Science, Artificial Intelligence
Sydney M. Katz, Kyle D. Julian, Christopher A. Strong, Mykel J. Kochenderfer
Summary: Neural networks are effective controllers in complex settings, but their difficult-to-verify outputs restrict their use in safety-critical applications. Recent research focuses on using formal methods to verify neural network outputs. This study proposes a method to provide probabilistic safety guarantees for neural network controllers using results from neural network verification tools.