New software that integrates dendritic properties into neural network models

summary: A new software framework integrates dendritic properties and mechanisms into large-scale neural network models.

source: from now on

FORTH-IMBB researchers have developed an innovative computational tool to uncover the role of dendrites. A complex neural brain structure that plays an important role in information processing.

New software allows important dendritic properties to be incorporated into neural network models. This work has important applications both in understanding brain function and in artificial intelligence.

Deciphering the secrets of the brain is considered one of the most important scientific endeavors of the 21st century. A better understanding of the mechanisms underlying brain function will contribute to research in the treatment of neurological diseases as well as in the field of artificial intelligence, which has become an integral part of our society.

in a recent publication Nature CommunicationsResearch Director of the Institute for Molecular Biology and Biotechnology (IMBB) under the Foundation for Research and Technology-Hellas (FOTH), Dr. Panayiota Poirazi’s team presents a new software framework that can integrate dendrites and their keys. Convert mechanisms into large-scale neural network models.

Dendrites are branching extensions of nerve cells that morphologically resemble the branches of a tree (hence the name). Its main function is to receive information (in the form of electrical or chemical signals) from other neurons and transmit it to the cell body.

In the decades following their discovery, their role in information processing remained unknown due to technical limitations in research.

However, recent studies have shown that dendrites possess rich mechanisms to perform complex mathematical calculations independently of the main neuron. At the same time, dendrites are equally important for the plasticity of the nervous system, the ability of the brain to adapt and change to its environment. This process plays a dominant role in complex brain functions such as learning, memory, decision-making and cognition.

Although we now largely understand the contribution of dendrites to the behavior of single neurons, their implications at the network level or whole brain regions remain unexplored.

A handful of studies have identified a correlation between dendritic complexity and various cognitive markers, and dendritic spines are known to decrease in aging or neurodegenerative diseases such as Alzheimer’s.

Additionally, research in AI has already benefited from using dendrite mechanisms as a source of inspiration for developing new algorithms that are improved and more efficient.

However, many open questions remain, and Dr. Poirazi’s team hopes that the tools they are developing will ease the work of those trying to understand the role of dendrites in brain function.

This study introduces a novel software framework that allows even naive users to build dendrite-like neural models in a simple and efficient manner, minimizing computational complexity. These computational models help explain the role of dendrites in complex brain functions while facilitating the integration of dendrites into neuromorphic devices, a type of neural-inspired artificial intelligence architecture.

This effort was led by Dr. Michalis Pangalos. Candidate in the Department of Biology at the University of Crete in collaboration with Dr. Spiros Chavlis, postdoctoral fellow at IMBB under the supervision of Dr. Poirazi.

Introducing AI and neuroscience research news

Author: public affairs office
source: from now on
contact: Public Affairs Office – Force
video: Image is in public domain

ORIGINAL RESEARCH: open access.
Michalis Pagkalos et al. Nature Communications

---

abstract

Introducing the Dendrify framework for integrating dendrites into spiking neural networks

Computational modeling is indispensable for understanding how intracellular neuronal function influences circuit processing.

However, the role of dendritic computation in network-level tasks remains largely unstudied. This is partly because existing tools do not allow for the development of realistic and efficient network models describing dendrites.

Although current spiking neural networks are efficient, they are generally very simplistic, overlooking essential dendritic properties. Conversely, circuit models with morphologically detailed neuron models are computationally expensive, making them impractical for large-scale network simulations.

To bridge the gap between these two extremes and facilitate the adoption of dendritic features in spiking neural networks, we introduce Dendrify, an open-source Python package based on Brian 2. Dendrify simplifies, through simple commands, automatically reduced compartmental neuron models while still retaining biologically relevant dendritic and synaptic integration properties.

These models strike a good balance between flexibility, performance and biological accuracy, paving the way for the development of more robust neuromorphic systems while exploring dendritic contributions to network-level function.