Structural network measures reveal the emergence of heavy-tailed degree distributions in lottery ticket multilayer perceptrons

Neural Networks, Год журнала: 2025, Номер 187, С. 107308 - 107308

Опубликована: Март 12, 2025

Artificial neural networks (ANNs) were originally modeled after their biological counterparts, but have since conceptually diverged in many ways. The resulting network architectures are not well understood, and furthermore, we lack the quantitative tools to characterize structures. Network science provides an ideal mathematical framework with which systems of interacting components, has transformed our understanding across domains, including mammalian brain. Yet, little been done bring ANNs. In this work, propose that leverage adapt methods measure both global- local-level characteristics Specifically, focus on structures efficient multilayer perceptrons as a case study, sparse systematically pruned such they share real-world networks. We use adapted metrics show pruning process leads emergence spanning subnetwork (lottery ticket perceptrons) complex architecture. This exhibits global local characteristics, heavy-tailed nodal degree distributions dominant weighted pathways, mirror patterns observed human neuronal connectivity. Furthermore, alterations precede catastrophic decay performance is heavily pruned. science-driven approach analysis artificial serves valuable tool establish improve fidelity, increase interpretability, assess Significance Statement become increasingly complex, often diverging from counterparts To design plausible "brain-like" architectures, whether advance neuroscience research or explainability, it essential these optimally resemble counterparts. offer information about interconnected systems, brain, attracted much attention for analyzing Here, present significance work: •We analyze structural demonstrate organizational similar those brain emerge through alone. convergence features compelling evidence optimality processing capabilities. •Our significant first step towards science-based networks, potential shed light fidelity

Язык: Английский

---

NeuroSimWorm: A multisensory framework for modeling and simulating neural circuits of Caenorhabditis elegans

Neurocomputing, Год журнала: 2025, Номер unknown, С. 130055 - 130055

Опубликована: Апрель 1, 2025

Язык: Английский

---

Knowledge-Enhanced Large Language Models for Ideation to Implementation: A New Paradigm in Product Design

Applied Soft Computing, Год журнала: 2025, Номер unknown, С. 113147 - 113147

Опубликована: Апрель 1, 2025

Язык: Английский

---

A learning orientation detection system and its application to grayscale images

Knowledge-Based Systems, Год журнала: 2024, Номер unknown, С. 112901 - 112901

Опубликована: Дек. 1, 2024

Язык: Английский

---

Learning feature relationships in CNN model via relational embedding convolution layer

Neural Networks, Год журнала: 2024, Номер 179, С. 106510 - 106510

Опубликована: Июль 6, 2024

Язык: Английский

---