Introducing the Dendrify framework for incorporating dendrites to spiking neural networks

Pagkalos, Michalis; Chavlis, Spyridon; Poirazi, Panayiota

doi:10.1038/s41467-022-35747-8

Cited by 30 publications

(9 citation statements)

References 102 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While we find that the input-output computation of L5PTs under this condition can be captured in very simple models, this is likely not the case in general. Previous studies have shown how to convert multi-compartmental models into deep artificial neural networks [14], simplified conductance based models [15,16] or stacks of linear-nonlinear units [17], which maintain high accuracy throughout a wide range of input conditions. All of these models are highly complex -for example Beniaguev et al (2021) find that a 7-layer convolutional network is necessary to capture the immense computational power of a single neuron.…”

Section: Discussionmentioning

confidence: 99%

Interpretable modelling of input-output computations in cortex

Bast,

Fruengel,

de Kock

et al. 2023

Preprint

View full text Add to dashboard Cite

Neurons receive input from thousands of synapses, which they transform into action potentials (APs) via their complex dendrites. How the dendritic location of these inputs, their timing, strength, and presynaptic origin impact AP output remains generally unknown. Here we demonstrate how to reveal which properties of the input causally underlie AP output, and how this input-output computation is influenced by the morphology and biophysical properties of the dendrites. For this purpose, we derive analytically tractable, interpretable models of the input-output computation that layer 5 pyramidal tract neurons (L5PTs) – the major output cell type of the cerebral cortex – perform upon sensory stimulation. We find that this input-output computation is preserved across L5PTs despite morphological and biophysical diversity. We show that three features are sufficient to explainin vivoobserved sensory responses and receptive fields of L5PTs with high accuracy: the count of active excitatory versus inhibitory synapses preceding the response, their spatial distribution on the dendrites, and the AP history. Based on this analytically tractable and interpretable description of the input-output computation, we show how to dissect the contributions of different input populations in thalamus and cortex to sensory responses of L5PTs. Thus, our approach provides a roadmap for revealing cellular input-output computations across differentin vivoconditions.Author SummaryRevealing how synaptic inputs drive action potential output is one of the major challenges in neuroscience research. An increasing number of approaches therefore seek to combine detailed measurements at synaptic, cellular and network scales into biologically realistic brain models. Indeed, such models have started to make empirically testable predictions about the inputs that underliein vivoobserved activity patterns. However, the enormous complexity of these models generally prevents the derivation of interpretable descriptions that explain how neurons transform synaptic input into action potential output, and how these input-output computations depend on synaptic, cellular and network properties. Here we introduce an approach to reveal input-output computations that neurons in the cerebral cortex perform upon sensory stimulation. We reduce a realistic multi-scale cortex model to the minimal description that accounts forin vivoobserved responses. Thereby, we identify the input-output computation that these cortical neurons perform under thisin vivocondition, and we show that this computation is preserved across neurons despite morphological and biophysical diversity. Our approach provides analytically tractable and interpretable descriptions of neuronal input-output computations during specificin vivoconditions.

show abstract

Section: Discussionmentioning

confidence: 99%

Interpretable modelling of input-output computations in cortex

Bast,

Fruengel,

de Kock

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The central auditory neuron in our model consists of a soma electrically connected to 10 uniform dendritic subunits. We implemented the model using the Dendrify framework and routines [Pagkalos et al, 2023] for creating compartmental neuronal models with dendritic properties. Each compartment was modelled using an integrate-and-fire model as in Eq.…”

Section: Methodsmentioning

confidence: 99%

Flexible integration of natural stimuli by auditory cortical neurons

Ang,

Clopath,

Kozlov

2024

Preprint

View full text Add to dashboard Cite

Neurons have rich input-output functions for processing and combining their inputs. Although many experiments characterize these functions by directly activating synaptic inputs on dendrites in vitro, the integration of spatiotemporal inputs representing real-world stimuli is less well studied. Using ethologically relevant stimuli, we study neuronal integration in relation to Boolean AND and OR operations thought to be important for pattern recognition. We recorded single-unit responses in the mouse auditory cortex to pairs of ultrasonic mouse vocalization (USV) syllables. We observed a range of integration responses, spanning the sublinear to supralinear regimes, with many responses resembling the MAX-like function, an instantiation of the OR operation. Integration was more MAX-like for strongly activating features, and more AND-like for spectrally distinct inputs. Importantly, single neurons could implement more than one integration function, in contrast to artificial networks which typically fix activation functions across all units and inputs. To understand the mechanism underlying the flexibility and heterogeneity in neuronal integration, we modelled how dendritic properties could influence the integration of inputs with complex spectrotemporal structure. Our results link nonlinear integration in dendrites to single-neuron computations for pattern recognition.

show abstract

“…The work of artificial neural networks in the context of the modeling process is presented in the work of Pagkalos M., Chavlis S., Poirazi P. [43]. In their research, the authors consider the method of computational modeling as an indispensable tool for visualizing data.…”

Section: Neural Network Modeling As a Tool For Analyzing Language Unitsmentioning

confidence: 99%

Analysis of the Theoretical Foundations of Neural Network Modeling of Language Unit Recognition

Dovhan¹

2023

Scientific Space: Integration of Traditional and Innovative Processes

View full text Add to dashboard Cite

In the context of linguistic science and the integration of the mathematical paradigm into humanitarian discourse, the analysis and processing of natural language (in particular, neural network modeling of language categories) is an important and urgent task. Research by a number of authors and a series of experiments show that artificial neural networks of various types and kinds with different parameterizations can significantly optimize linguistic research: accelerate, deepen, integrate into various scientific fields, etc. At the same time, the use of artificial neural networks in linguistics is an important area of work, as well as a powerful and productive tool for a number of relevant studies, which, however, requires careful analysis and development of implementation strategies. The purpose of the article is to analyze the features of neural network modeling of language units recognition as an effective method of cognition within the anthropocentric paradigm of research. The solution of such research tasks determines the logic of presentation of the studied material in the article: introduction, systematization of achievements in the theory and practice of modeling as a universal tool of scientific cognition in general and theoretical justification of neural network modeling in the context of linguistic paradigm. Methodology of the study is based of the method of analyzing scientific research was updated, which led to the search and analysis of scientific publications related to neural network modeling (in particular, language units). We analyzed more than 60 recent scientific studies and publications covering aspects of the problem under study, which we evaluated based on their relevance, methodological specificity, and scientific novelty. Thus, the research methods outlined in this article allowed us to conduct a thorough analysis of the state and prospects for the development of the theoretical foundations of neural network modeling of language unit recognition. The analysis of the latest scientific research and publications has made it possible to determine the role and place of artificial neural networks of various types and their specifications in the process of modeling language units. The above made it possible to identify the main trends in working with text data aimed at improving the quality of their processing, generation, etc. Results of the survey showed that one of the core tasks of modern linguistic science is to understand the language polysystem, the peculiarities of its structure and the nature of its functioning. In addition, the discursive nature of language practices in the context of data interconnection is also important for documenting language structures and verbal practices that are socially determined. That is why the use of neural networks as a tool for conducting local linguistic and integrated scientific research involving the mathematical paradigm is gaining popularity. Neural network modeling of language categories in this context is the basic basis for such research. Practical implications. The actualization of neural network modeling of linguistic categories is not only one of the means of studying linguistic polysystems, but also an objective criterion of checking the truth of linguistic knowledge. Value/originality. Neural network modeling of language unit recognition has a significant impact on the development, improvement, and evolution of modern linguistic research. The effectiveness of neural network modeling of language categories creates opportunities for a deeper understanding of the studied linguistic objects, phenomena and processes, encouraging linguists to develop various linguistic models that could solve practical linguistic problems (information retrieval, machine translation, natural language processing, knowledge extraction and localization from text, etc.).

show abstract

Introducing the Dendrify framework for incorporating dendrites to spiking neural networks

Cited by 30 publications

References 102 publications

Interpretable modelling of input-output computations in cortex

Interpretable modelling of input-output computations in cortex

Flexible integration of natural stimuli by auditory cortical neurons

Analysis of the Theoretical Foundations of Neural Network Modeling of Language Unit Recognition

Contact Info

Product

Resources

About