A Novel Evolutionary Algorithm-Based Approach for Optimizing the Mix Design of High-Performance Fiber-Reinforced Concrete (HPFRCC) DOI Creative Commons

Peyman farhadi yeganeh,

Ata Hojatkashani, Maryam Firoozi Nezamabadi

и другие.

Research Square (Research Square), Год журнала: 2024, Номер unknown

Опубликована: Июль 30, 2024

Abstract With the increasing demand for high-performance concrete, optimizing design of alloy-armed compounds (HPFRCC) has become a major challenge. Optimizing mix enables achievement desired stability and ductility. This study offers an innovative way to overcome this challenge using artificial intelligence (AI). Our method uses neural networks (ANN) model relationship between mixed component final HPFRCC properties. Then, we explored wide space advanced distribution algorithms determine best combination. These are inspired by nature, such as evolutionary algorithm (EO), which replicates scientific process. significantly reduces need laboratory testing employing methodologies. significant optimizes time cost. In addition, comprehensively comparing EO performance with other similar algorithms, excellence innovation were demonstrated in complex mixture, new multitarget optimization called FC-MOEO/AE was used predict can be guide decision-making pre-construction stage.

Язык: Английский

Analysis of argument structure constructions in the large language model BERT DOI Creative Commons

Pegah Ramezani,

Achim Schilling, Patrick Krauß

и другие.

Frontiers in Artificial Intelligence, Год журнала: 2025, Номер 8

Опубликована: Янв. 31, 2025

Understanding how language and linguistic constructions are processed in the brain is a fundamental question cognitive computational neuroscience. In this study, we investigate processing representation of Argument Structure Constructions (ASCs) BERT model, extending previous analyses conducted with Long Short-Term Memory (LSTM) networks. We utilized custom GPT-4 generated dataset comprising 2000 sentences, evenly distributed among four ASC types: transitive, ditransitive, caused-motion, resultative constructions. was assessed using various token embeddings across its 12 layers. Our involved visualizing Multidimensional Scaling (MDS) t-Distributed Stochastic Neighbor Embedding (t-SNE), calculating Generalized Discrimination Value (GDV) to quantify degree clustering. also trained feedforward classifiers (probes) predict construction categories from these embeddings. Results reveal that CLS cluster best according types layers 2, 3, 4, diminished clustering intermediate slight increase final Token for DET SUBJ showed consistent intermediate-level layers, while VERB demonstrated systematic layer 1 12. OBJ exhibited minimal initially, which increased substantially, peaking 10. Probe accuracies indicated initial contained no specific information, as seen low chance-level 1. From 2 onward, probe surpassed 90 percent, highlighting latent category information not evident GDV alone. Additionally, Fisher Discriminant Ratio (FDR) analysis attention weights revealed tokens had highest FDR scores, indicating they play crucial role differentiating ASCs, followed by tokens. SUBJ, CLS, SEP did show significant scores. study underscores complex, layered BERT, revealing both similarities differences compared recurrent models like LSTMs. Future research will compare findings neuroimaging data during continuous speech perception better understand neural correlates processing. This demonstrates potential transformer-based mirror human brain, offering valuable insights into mechanisms underlying understanding.

Язык: Английский

Процитировано

2
AI‐Guided Design of Antimicrobial Peptide Hydrogels for Precise Treatment of Drug‐resistant Bacterial Infections DOI
Zhihui Jiang,

Jianwen Feng,

Fan Wang

и другие.

Advanced Materials, Год журнала: 2025, Номер unknown

Опубликована: Март 30, 2025

Abstract Traditional biomaterial development lacks systematicity and predictability, posing significant challenges in addressing the intricate engineering issues related to infections with drug‐resistant bacteria. The unprecedented ability of artificial intelligence (AI) manage complex systems offers a novel paradigm for materials development. However, no AI model currently guides antibacterial biomaterials based on an in‐depth understanding interplay between In this study, AI‐guided design platform (AMP‐hydrogel‐Designer) is developed generate biomaterials. This utilizes generative multi‐objective constrained optimization thiol‐containing high‐efficiency antimicrobial peptide (AMP), that functionally coupled hydrogel form network structure. Additionally, Cu‐modified barium titanate (Cu‐BTO) incorporated facilitate further cross–linking via Cu 2+ /SH coordination produce AI‐AMP‐hydrogel. vitro, AI‐AMP‐hydrogel exhibits > 99.99% bactericidal efficacy against Methicillin‐resistant Staphylococcus aureus (MRSA) Escherichia coli ( E. coli) . Furthermore, Cu‐BTO converts mechanical stimulation into electrical signals, thereby promoting expression growth factors angiogenesis. rat dynamic wounds, AI‐AMP significantly reduces MRSA load markedly accelerates wound healing. Therefore, strategy innovative solution precisely treat bacterial infections.

Язык: Английский

Процитировано

2
Mathematical modeling of local calcium signaling in neurons using artificial neural networks DOI Open Access
Darshana Upadhyay, Hardik Joshi

Discrete and Continuous Dynamical Systems - S, Год журнала: 2025, Номер 0(0), С. 0 - 0

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

1
Neural dynamics of semantic control underlying generative storytelling DOI Creative Commons
Clara Rastelli, Antonino Greco, Chiara Finocchiaro

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Март 18, 2024

Storytelling has been pivotal for the transmission of knowledge and cultural norms across human history. A crucial process underlying generation narratives is exertion cognitive control on semantic representations stored in memory, a phenomenon referred as control. Despite extensive literature investigating neural mechanisms generative language tasks, little effort done towards storytelling under naturalistic conditions. Here, we probed participants to generate stories response set instructions which triggered narrative that was either appropriate (ordinary), novel (random), or balanced (creative), while recording functional magnetic resonance imaging (fMRI) signal. By leveraging deep models, demonstrated how ideally level during story generation. At level, creative were differentiated by multivariate pattern activity frontal cortices compared ordinary ones fronto- temporo-parietal with respect randomly generated stories. Crucially, similar brain regions also encoding features distinguished behaviourally. Moreover, decomposed dynamics into connectome harmonic modes found specific spatial frequency patterns modulation Finally, different coupling within between default mode, salience networks when contrasting their controls. Together, our findings highlight regulation exploration ideation contribute deeper understanding underpinning role storytelling.

Язык: Английский

Процитировано

5
An Alternative to Cognitivism: Computational Phenomenology for Deep Learning DOI Creative Commons
Pierre Beckmann,

Guillaume Köstner,

Inês Hipólito

и другие.

Minds and Machines, Год журнала: 2023, Номер 33(3), С. 397 - 427

Опубликована: Июнь 29, 2023

Abstract We propose a non-representationalist framework for deep learning relying on novel method computational phenomenology, dialogue between the first-person perspective (relying phenomenology) and mechanisms of models. thereby an alternative to modern cognitivist interpretation learning, according which artificial neural networks encode representations external entities. This mainly relies neuro-representationalism, position that combines strong ontological commitment towards scientific theoretical entities idea brain operates symbolic these proceed as follows: after offering review cognitivism neuro-representationalism in field we first elaborate phenomenological critique positions; then sketch out phenomenology distinguish it from existing alternatives; finally apply this new models trained specific tasks, order formulate conceptual deep-learning, allows one think networks’ terms lived experience.

Язык: Английский

Процитировано

9
Constructing Biologically Constrained RNNs via Dale's Backprop and Topologically-Informed Pruning DOI Creative Commons
Aishwarya H. Balwani,

A. Wang,

Farzaneh Najafi

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2025, Номер unknown

Опубликована: Янв. 13, 2025

A bstract Recurrent neural networks (RNNs) have emerged as a prominent tool for modeling cortical function, and yet their conventional architecture is lacking in physiological anatomical fidelity. In particular, these models often fail to incorporate two crucial biological constraints: i) Dale’s law, i.e., sign constraints that preserve the “type” of projections from individual neurons, ii) Structured connectivity motifs, highly sparse defined connections amongst various neuronal populations. Both are known impair learning performance artificial networks, especially when trained perform complicated tasks; but modern experimental methodologies allow us record diverse populations spanning multiple brain regions, using RNN study interactions without incorporating fundamental properties raises questions regarding validity insights gleaned them. To address concerns, our work develops methods let train RNNs which respect law whilst simultaneously maintaining specific pattern across entire network. We provide mathematical grounding guarantees approaches both types constraints, show empirically match any constraints. Finally, we demonstrate utility inferring multi-regional by training network reconstruct 2-photon calcium imaging data during visual behaviour mice, enforcing data-driven, cell-type between spread layers areas. doing so, find inferred model corroborate findings agreement with theory predictive coding, thus validating applicability methods.

Язык: Английский

Процитировано

0
A Historical and Global Overview of Neuropsychology DOI
Caetano Schmidt Gundlach Knop-Máximo, Vanessa de Almeida Signori, Luís Felipe da Silva Rodrigues

и другие.

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0
Neural dynamics of semantic control underlying generative storytelling DOI Creative Commons
Clara Rastelli, Antonino Greco, Chiara Finocchiaro

и другие.

Communications Biology, Год журнала: 2025, Номер 8(1)

Опубликована: Март 28, 2025

Язык: Английский

Процитировано

0
Synchronization of Kuramoto oscillators via HEOL, and a discussion on AI DOI Open Access
Emmanuel Delaleau, Cédric Join, Michel Fliesś

и другие.

IFAC-PapersOnLine, Год журнала: 2025, Номер 59(1), С. 229 - 234

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0
Artificial Intelligence and Its Impact on the Mental Health Field DOI
Marena Hernández Lugo, Diego D. Díaz‐Guerra, Guillermo Alfredo Jiménez Pérez

и другие.

Lecture notes in networks and systems, Год журнала: 2025, Номер unknown, С. 506 - 516

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0