Aging Clinical and Experimental Research, Год журнала: 2023, Номер 35(11), С. 2363 - 2397
Опубликована: Сен. 8, 2023
Abstract The increasing access to health data worldwide is driving a resurgence in machine learning research, including data-hungry deep algorithms. More computationally efficient algorithms now offer unique opportunities enhance diagnosis, risk stratification, and individualised approaches patient management. Such are particularly relevant for the management of older patients, group that characterised by complex multimorbidity patterns significant interindividual variability homeostatic capacity, organ function, response treatment. Clinical tools utilise determine optimal choice treatment slowly gaining necessary approval from governing bodies being implemented into healthcare, with implications virtually all medical disciplines during next phase digital medicine. Beyond obtaining regulatory approval, crucial element implementing these trust support people use them. In this context, an increased understanding clinicians artificial intelligence provides appreciation possible benefits, risks, uncertainties, improves chances successful adoption. This review broad taxonomy algorithms, followed more detailed description each algorithm class, their purpose capabilities, examples applications, geriatric Additional focus given on clinical challenges involved relying devices reduced interpretability progress made counteracting latter via development explainable learning.
Язык: Английский
Процитировано
50Nonlinear manifolds underlie neural population activity during behaviour
bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2023, Номер unknown
Опубликована: Июль 19, 2023
There is rich variety in the activity of single neurons recorded during behaviour. Yet, these diverse neuron responses can be well described by relatively few patterns neural co-modulation. The study such low-dimensional structure population has provided important insights into how brain generates Virtually all studies have used linear dimensionality reduction techniques to estimate population-wide co-modulation patterns, constraining them a flat “neural manifold”. Here, we hypothesised that since nonlinear and make thousands distributed recurrent connections likely amplify nonlinearities, manifolds should intrinsically nonlinear. Combining recordings from monkey, mouse, human motor cortex, mouse striatum, show that: 1) are nonlinear; 2) their nonlinearity becomes more evident complex tasks require varied patterns; 3) manifold varies across architecturally distinct regions. Simulations using network models confirmed proposed relationship between circuit connectivity nonlinearity, including differences Thus, underlying generation behaviour inherently nonlinear, properly accounting for nonlinearities will critical as neuroscientists move towards studying numerous regions involved increasingly naturalistic behaviours.
Язык: Английский
Процитировано
28Random vector functional link networks for function approximation on manifolds
Frontiers in Applied Mathematics and Statistics, Год журнала: 2024, Номер 10
Опубликована: Апрель 17, 2024
The learning speed of feed-forward neural networks is notoriously slow and has presented a bottleneck in deep applications for several decades. For instance, gradient-based algorithms, which are used extensively to train networks, tend work slowly when all the network parameters must be iteratively tuned. To counter this, both researchers practitioners have tried introducing randomness reduce requirement. Based on original construction Igelnik Pao, single layer neural-networks with random input-to-hidden weights biases seen success practice, but necessary theoretical justification lacking. In this study, we begin fill gap. We then extend result non-asymptotic setting using concentration inequality Monte-Carlo integral approximations. provide (corrected) rigorous proof that Pao universal approximator continuous functions compact domains, approximation error squared decaying asymptotically like O (1/ n ) number nodes. setting, proving one can achieve any desired high probability provided sufficiently large. further adapt randomized architecture approximate smooth, submanifolds Euclidean space, providing guarantees asymptotic forms. Finally, illustrate our results manifolds numerical experiments.
Язык: Английский
Процитировано
13Structured Dynamics in the Algorithmic Agent
Entropy, Год журнала: 2025, Номер 27(1), С. 90 - 90
Опубликована: Янв. 19, 2025
In the Kolmogorov Theory of Consciousness, algorithmic agents utilize inferred compressive models to track coarse-grained data produced by simplified world models, capturing regularities that structure subjective experience and guide action planning. Here, we study dynamical aspects this framework examining how requirement tracking natural drives structural properties agent. We first formalize notion a generative model using language symmetry from group theory, specifically employing Lie pseudogroups describe continuous transformations characterize invariance in data. Then, adopting generic neural network as proxy for agent system drawing parallels Noether’s theorem physics, demonstrate forces mirror model. This dual constraint on agent’s constitutive parameters repertoire enforces hierarchical organization consistent with manifold hypothesis network. Our findings bridge perspectives information theory (Kolmogorov complexity, modeling), (group theory), dynamics (conservation laws, reduced manifolds), offering insights into correlates agenthood structured systems, well design artificial intelligence computational brain.
Язык: Английский
Процитировано
1Downstream network transformations dissociate neural activity from causal functional contributions
Scientific Reports, Год журнала: 2024, Номер 14(1)
Опубликована: Янв. 24, 2024
Abstract Neuroscientists rely on distributed spatio-temporal patterns of neural activity to understand how units contribute cognitive functions and behavior. However, the extent which reliably indicates a unit's causal contribution behavior is not well understood. To address this issue, we provide systematic multi-site perturbation framework that captures time-varying contributions elements collectively produced outcome. Applying our intuitive toy examples artificial networks revealed recorded may be generally informative their due transformations within network. Overall, findings emphasize limitations inferring mechanisms from activities offer rigorous lesioning for elucidating contributions.
Язык: Английский
Процитировано
6Neuronal signature of spatial decision-making during navigation by freely moving rats by using calcium imaging
Proceedings of the National Academy of Sciences, Год журнала: 2022, Номер 119(44)
Опубликована: Окт. 24, 2022
A challenge in spatial memory is understanding how place cell firing contributes to decision-making navigation. recency task was created which freely moving rats first became familiar with a context over several days and thereafter were required encode then selectively recall one of three specific locations within it that chosen be rewarded day. Calcium imaging used record from more than 1,000 cells area CA1 the hippocampus five during exploration, sample, choice phases daily task. The key finding neural activity startbox rose steadily short period prior entry arena this selective population predictive changing goal on correct trials but not animals made errors. Single-cell measures converged idea prospective coding can involved navigational decision-making.
Язык: Английский
Процитировано
19Learning visual stimulus-evoked EEG manifold for neural image classification
Neurocomputing, Год журнала: 2024, Номер 588, С. 127654 - 127654
Опубликована: Апрель 15, 2024
Visual neural decoding, namely the ability to interpret external visual stimuli from patterns of brain activity, is a challenging task in neuroscience research. Recent studies have focused on characterizing activity across multiple neurons that can be described terms population-level features. In this study, we combine spatial, spectral, and temporal features achieve manifold classification capable characterize perception simulate working memory human brain. We treat spatio-temporal spectral information separately by means custom deep learning architectures based Riemann two-dimensional EEG spectrogram representation. addition, CNN-based model used classify stimulus-evoked signals while viewing 11-class (i.e., all-black plus 0-9 digit images) MindBigData MNIST dataset. The effectiveness proposed integration strategy evaluated signal task, achieving an overall accuracy 86%, comparable state-of-the-art benchmarks.
Язык: Английский
Процитировано
4Information Geometry and Manifold Learning: A Novel Framework for Analyzing Alzheimer’s Disease MRI Data
Diagnostics, Год журнала: 2025, Номер 15(2), С. 153 - 153
Опубликована: Янв. 10, 2025
Background: Alzheimer’s disease is a progressive neurological condition marked by decline in cognitive abilities. Early diagnosis crucial but challenging due to overlapping symptoms among impairment stages, necessitating non-invasive, reliable diagnostic tools. Methods: We applied information geometry and manifold learning analyze grayscale MRI scans classified into No Impairment, Very Mild, Moderate Impairment. Preprocessed images were reduced via Principal Component Analysis (retaining 95% variance) converted statistical manifolds using estimated mean vectors covariance matrices. Geodesic distances, computed with the Fisher Information metric, quantified class differences. Graph Neural Networks, including Convolutional Networks (GCN), Attention (GAT), GraphSAGE, utilized categorize levels graph-based representations of data. Results: Significant differences structures observed, increased variability stronger feature correlations at higher levels. distances between Impairment Mild (58.68, p<0.001) (58.28, are statistically significant. GCN GraphSAGE achieve perfect classification accuracy (precision, recall, F1-Score: 1.0), correctly identifying all instances across classes. GAT attains an overall 59.61%, variable performance Conclusions: Integrating geometry, learning, GNNs effectively differentiates AD stages from The strong indicates their potential assist clinicians early identification tracking progression.
Язык: Английский
Процитировано
0Charting the Course of Leadership Through the Digital Era
Advances in logistics, operations, and management science book series, Год журнала: 2025, Номер unknown, С. 119 - 156
Опубликована: Янв. 24, 2025
This chapter aims to provide actionable strategies for empowering leaders in leveraging technology amplify their leadership efficacy contemporary business environments by employing extensive literature review and deep learning models, a method Artificial Intelligence (AI). It investigates the effectiveness of three approaches—Transformational, Transactional, Servant Leadership—in meeting high-performance expectations within organizational contexts. The performance these styles based on scores have been analyzed using data analysis techniques neural network including Generative Adversarial Networks (GANs) Variational Autoencoders (VAEs). Furthermore, Deep Convolutional (DCGANs), utilized visualize complex dynamics. findings from this comprehensive valuable insights into strengths limitations each approach, guiding strategic development initiatives decision-making processes.
Язык: Английский
Процитировано
0Exploring the role of dimensionality transformation in episodic memory
Trends in Cognitive Sciences, Год журнала: 2025, Номер unknown
Опубликована: Фев. 1, 2025
Episodic memory must accomplish two adversarial goals: encoding and storing a multitude of experiences without exceeding the finite neuronal structure brain, recalling memories in vivid detail. Dimensionality reduction expansion ('dimensionality transformation') enable brain to meet these demands. Reduction compresses sensory input into simplified, storable codes, while reconstructs details. Although processes are essential memory, their neural mechanisms for episodic remain unclear. Drawing on recent insights from cognitive psychology, systems neuroscience, neuroanatomy, we propose accounts how dimensionality transformation occurs brain: structurally (via corticohippocampal pathways) functionally (through oscillations). By examining cross-species evidence, highlight that may support identify crucial questions future research.
Язык: Английский
Процитировано
0