Sequence Prediction and Classification of Echo State Networks DOI Creative Commons

Jingyu Sun,

Lixiang Li, Haipeng Peng

et al.

Mathematics, Journal Year: 2023, Volume and Issue: 11(22), P. 4640 - 4640

Published: Nov. 14, 2023

The echo state network is a unique form of recurrent neural network. Due to its feedback mechanism, it exhibits superior nonlinear behavior compared traditional networks and highly regarded for simplicity efficiency in computation. In recent years, as development has progressed, the security threats faced by have increased. To detect counter these threats, analysis traffic become crucial research focus. demonstrated exceptional performance sequence prediction. this article, we delve into impact on time series. We enhanced model increasing number layers adopting different data input approach. apply predict chaotic systems that appear ostensibly regular but are inherently irregular. Additionally, utilize classification sound data. Upon evaluating using root mean squared error micro-F1, observed our commendable accuracy stability.

Language: Английский

Evolution beats random chance: Performance-dependent network evolution for enhanced computational capacity DOI Creative Commons
Manish Yadav, Sudeshna Sinha, M. Stender

et al.

Physical review. E, Journal Year: 2025, Volume and Issue: 111(1)

Published: Jan. 29, 2025

The quest to understand relationships in networks across scientific disciplines has intensified. However, the optimal network architecture remains elusive, particularly for complex information processing. Therefore, we investigate how and specific structures form efficiently solve distinct tasks using a framework of performance-dependent evolution, leveraging reservoir computing principles. Our study demonstrates that task-specific minimal obtained through this consistently outperform generated by alternative growth strategies Erdős-Rényi random networks. Evolved exhibit unexpected sparsity adhere scaling laws node-density space while showcasing distinctive asymmetry input readout node distribution. Consequently, propose heuristic quantifying task complexity from performance-dependently evolved networks, offering valuable insights into evolutionary dynamics structure-function relationship. findings advance fundamental understanding process-specific evolution shed light on design optimization processing mechanisms, notably machine learning. Published American Physical Society 2025

Language: Английский

Citations

3
Simulating the impact of white matter connectivity on processing time scales using brain network models DOI Creative Commons
Paul Triebkorn, Viktor Jirsa, Peter Ford Dominey

et al.

Communications Biology, Journal Year: 2025, Volume and Issue: 8(1)

Published: Feb. 7, 2025

The capacity of the brain to process input across temporal scales is exemplified in human narrative, which requires integration information ranging from words, over sentences long paragraphs. It has been shown that this processing distributed a hierarchy multiple areas with close sensory cortex, on faster time scale than associative cortex. In study we used reservoir computing derived connectivity investigate effect structural regions during narrative task paradigm. We systematically tested removal selected fibre bundles (IFO, ILF, MLF, SLF I/II/III, UF, AF) regions. show distance pathways such as IFO provide form shortcut whereby driven activation visual cortex can directly impact distant frontal areas. To validate our model demonstrated significant correlation predicted ordering empirical results intact/scrambled fMRI This emphasizes connectivity's role hierarchies, providing framework for future research structure and neural dynamics cognitive tasks.

Language: Английский

Citations

1
Computational memory capacity predicts aging and cognitive decline DOI Creative Commons
Mite Mijalkov, Lance Storm, Blanca Zufiria Gerbolés

et al.

Nature Communications, Journal Year: 2025, Volume and Issue: 16(1)

Published: March 20, 2025

Abstract Memory is a crucial cognitive function that deteriorates with age. However, this ability normally assessed using tests instead of the architecture brain networks. Here, we use reservoir computing, recurrent neural network computing paradigm, to assess linear memory capacities neural-network reservoirs extracted from anatomical connectivity data in lifespan cohort 636 individuals. The computational capacity emerges as robust marker aging, being associated resting-state functional activity, white matter integrity, locus coeruleus signal intensity, and performance. We replicate our findings an independent 154 young 72 old By linking cognition, open new pathways employ investigate aging age-related disorders.

Language: Английский

Citations

1
Spatially embedded recurrent neural networks reveal widespread links between structural and functional neuroscience findings DOI Creative Commons
Jascha Achterberg, Danyal Akarca,

Daniel Strouse

et al.

Nature Machine Intelligence, Journal Year: 2023, Volume and Issue: 5(12), P. 1369 - 1381

Published: Nov. 20, 2023

Abstract Brain networks exist within the confines of resource limitations. As a result, brain network must overcome metabolic costs growing and sustaining its physical space, while simultaneously implementing required information processing. Here, to observe effect these processes, we introduce spatially embedded recurrent neural (seRNN). seRNNs learn basic task-related inferences existing three-dimensional Euclidean where communication constituent neurons is constrained by sparse connectome. We find that converge on structural functional features are also commonly found in primate cerebral cortices. Specifically, they solving using modular small-world networks, which functionally similar units configure themselves utilize an energetically efficient mixed-selective code. Because emerge unison, reveal how many common motifs strongly intertwined can be attributed biological optimization processes. incorporate biophysical constraints fully artificial system serve as bridge between research communities move neuroscientific understanding forwards.

Language: Английский

Citations

22
Bio-instantiated recurrent neural networks: Integrating neurobiology-based network topology in artificial networks DOI Creative Commons
Alexandros Goulas, Fabrizio Damicelli, Claus C. Hilgetag

et al.

Neural Networks, Journal Year: 2021, Volume and Issue: 142, P. 608 - 618

Published: July 24, 2021

Biological neuronal networks (BNNs) are a source of inspiration and analogy making for researchers that focus on artificial (ANNs). Moreover, neuroscientists increasingly use ANNs as model the brain. Despite certain similarities between these two types networks, important differences can be discerned. First, biological neural sculpted by evolution constraints it entails, whereas engineered to solve particular tasks. Second, network topology systems, apart from some analogies drawn, exhibits pronounced differences. Here, we examine strategies construct recurrent (RNNs) instantiate brains different species. We refer such RNNs bio-instantiated. investigate performance bio-instantiated in terms of: (i) prediction itself, is, capacity minimize cost function at hand test data, (ii) speed training, how fast during training reaches its optimal performance. working memory tasks where task-relevant information must tracked sequence events unfolds time. highlight used with found BNNs, without sacrificing observe no enhancement when compared randomly wired RNNs, our approach demonstrates empirical data constructing thus, facilitating further experimentation biologically realistic topologies, contexts aspect is desired.

Language: Английский

Citations

32
Reservoir Computing Using Measurement-Controlled Quantum Dynamics DOI Open Access

A. H. Abbas,

Ivan S. Maksymov

Electronics, Journal Year: 2024, Volume and Issue: 13(6), P. 1164 - 1164

Published: March 21, 2024

Physical reservoir computing (RC) is a machine learning algorithm that employs the dynamics of physical system to forecast highly nonlinear and chaotic phenomena. In this paper, we introduce quantum RC probed atom in cavity. The experiences coherent driving at particular rate, leading measurement-controlled evolution. proposed can make fast reliable forecasts using small number artificial neurons compared with traditional algorithm. We theoretically validate operation reservoir, demonstrating its potential be used error-tolerant applications, where approximate approaches may feasible conditions limited computational energy resources.

Language: Английский

Citations

4
A biologically inspired computational model of human ventral temporal cortex DOI
Yiyuan Zhang, Ke Zhou,

Pinglei Bao

et al.

Neural Networks, Journal Year: 2024, Volume and Issue: 178, P. 106437 - 106437

Published: June 13, 2024

Language: Английский

Citations

4
Systematic perturbation of an artificial neural network: A step towards quantifying causal contributions in the brain DOI Creative Commons
Kayson Fakhar, Claus C. Hilgetag

PLoS Computational Biology, Journal Year: 2022, Volume and Issue: 18(6), P. e1010250 - e1010250

Published: June 17, 2022

Lesion inference analysis is a fundamental approach for characterizing the causal contributions of neural elements to brain function. This has gained new prominence through arrival modern perturbation techniques with unprecedented levels spatiotemporal precision. While inferences drawn from perturbations are conceptually powerful, they face methodological difficulties. Particularly, challenged disentangle true involved elements, since often functions arise coalitions distributed, interacting and localized have unknown global consequences. To elucidate these limitations, we systematically exhaustively lesioned small artificial network (ANN) playing classic arcade game. We determined functional all nodes links, contrasting results sequential single-element simultaneous multiple elements. found that lesioning individual one at time, produced biased results. By contrast, multi-site lesion captured crucial details were missed by single-site lesions. conclude even seemingly simple ANNs show surprising complexity needs be addressed multi-lesioning coherent characterization.

Language: Английский

Citations

17
Dynamical measures of developing neuroelectric fields in emerging consciousness DOI Creative Commons
William J. Bosl, Jenny Capua-Shenkar

Current Opinion in Behavioral Sciences, Journal Year: 2025, Volume and Issue: 61, P. 101480 - 101480

Published: Jan. 10, 2025

Language: Английский

Citations

0
Asymmetrically connected reservoir networks learn better DOI Creative Commons
Shailendra K. Rathor,

Martin Ziegler,

Jörg Schumacher

et al.

Physical review. E, Journal Year: 2025, Volume and Issue: 111(1)

Published: Jan. 24, 2025

We show that connectivity within the high-dimensional recurrent layer of a reservoir network is crucial for its performance. To this end, we systematically investigate impact on performance, i.e., examine symmetry and structure in relation to computational power. Reservoirs with random asymmetric connections are found perform better an exemplary Mackey-Glass time series than all structured reservoirs, including biologically inspired connectivities, such as small-world topologies. This result quantified by information processing capacity different topologies which becomes highest randomly connected networks. Published American Physical Society 2025

Language: Английский

Citations

0