Computational Neuroscience Commons^™

Coarse-Graining Spiking Reservoirs: Reducing Reservoir Size While Preserving Critical Dynamics, Tucker X. Mastin, Christof Teuscher

Northeast Journal of Complex Systems (NEJCS)

We propose an extension of renormalization into the domain of spiking neural networks, thereby providing a novel framework for coarse-graining neural networks without disrupting their critical properties. The proposed coarse-graining technique merges neurons and synaptic connections based on a graph-theoretic distance derived from synaptic weight strength and is configured to effectively prune the reservoir size while preserving the scale-free spiking dynamics indicative of criticality. Criticality in spiking neural networks may provide information-theoretic advantages by optimizing information processing and sensitivity to input. Using time-series prediction benchmarks, we demonstrate that networks operating at criticality exhibit up to 32% higher prediction accuracy before …

Topodino: Self-Supervised Topological Representation Learning For Neuronal Morphologies, Yasser Binbisher

Master's Theses

Neuronal cell types are categorized by transcriptomic identity, yet their morphological heterogeneity defies this classification. In response, researchers have adopted unsupervised graph representation learning as a tool to reveal morphological variation within single-class transcriptomic types. However, the complex geometry of neuronal morphology—especially long axons and dense dendrites—challenges graph neural networks, which struggle with message propagation across extended structures. To mitigate this, current approaches enforce sub-sampling on neuronal graphs and omit axons entirely, sacrificing critical biological features for computational efficiency. To overcome this trade-off, this thesis introduces TopoDINO, a self-supervised, topology-aware representation learning model designed to preserve the full hierarchical organization …

Robust And Generalizable Representations In The Hippocampus, Hung-Tu Chen

Dartmouth College Ph.D Dissertations

An intelligent system must balance generalizing across similar experiences with maintaining the distinctiveness of each experience. This thesis explores how the hippocampus manages this balance through its neural representations to support adaptive behavior. In Chapter 1, I provide an overview of key hippocampal phenomena that contribute to this process, including remapping, splitter signal, and replay. In Chapter 2, I challenge the concept of random remapping by showing that it is possible to predict, better than chance, how a given experience will be encoded in the hippocampus across different subjects. This suggests that encoding of related experiences, which was previously thought …

Genomic Data Science Approaches For Understanding Human Diseases, Snehal Shah

All Dissertations

The intricate interplay of genetic predisposition, environmental influences, and lifestyle acts as the multifactorial landscape of diseases. Understanding this complexity presents a significant challenge. Molecular insights into disease mechanisms, particularly the interactions of DNA, RNA, and proteins with environmental and lifestyle factors, have revolutionized disease diagnosis, prognosis, and treatment. High-throughput technologies, such as next-generation sequencing, generate large amounts of molecular data, holding a wealth of knowledge. These datasets unveil the roles of genes and their interactions with various factors through analysis, shedding light on previously unknown molecular mechanisms underlying disease pathogenesis. Furthermore, they facilitate the discovery of biomarkers crucial for …

Fusing Classic Motion Energy Models And Deep Learning For Coarse-To-Fine Moving Object Segmentation, Matthias Tangemann, Matthias Kümmerer, Matthias Bethge

MODVIS Workshop

Classic motion energy models are able to predict a wide range of physiological and behavioral aspects of motion perception in humans. Whether these models can be used as a basis for higher-level tasks, such as moving object segmentation, has however hardly been explored yet. Here, we present a model that combines a motion energy representation with recent computer vision approaches for figure-ground segmentation of naturalistic stimuli. We find that unlike established motion segmentation models but similar to humans, our model generalizes to random-dot stimuli when only trained on RGB videos.

Analysis And Computation Of Constrained Sparse Coding On Emerging Non-Von Neumann Devices, Kyle Henke

Mathematics & Statistics ETDs

This dissertation seeks to understand how different formulations of the neurally inspired Locally Competitive Algorithm (LCA) represent and solve optimization problems. By studying these networks mathematically through the lens of dynamical and gradient systems, the goal is to discern how neural computations converge and link this knowledge to theoretical neuroscience and artificial intelligence (AI). Both classical computers and advanced emerging hardware are employed in this study. The contributions of this work include:

1. Theoretical Work: A comprehensive convergence analysis for networks using both generic Rectified Linear Unit (ReLU) and Rectified Sigmoid activation functions. Exploration of techniques to address the binary …

When Brain Meets Artificial Intelligence, Lu Zhang

Computer Science and Engineering Dissertations - Archive

When we review the history of development of artificial intelligence (AI), we will find that brain science plays a pivotal role in fostering breakthroughs in AI, such as artificial neural networks (ANNs). Today, AI has made remarkable strides, particularly with the emergence of large language models (LLMs), surpassing expectations and achieving human-level performance in certain tasks. Nonetheless, an insurmountable gap remains between AI and human intelligence. It is urgent to establish a bridge between brain science and AI, promoting their mutual enhancement and collaborations. This involve establishing connections from brain science to AI (brain-inspired AI), and reversely, from AI to …

Self-Supervised Pretraining And Transfer Learning On Fmri Data With Transformers, Sean Paulsen

Dartmouth College Ph.D Dissertations

Transfer learning is a machine learning technique founded on the idea that knowledge acquired by a model during “pretraining” on a source task can be transferred to the learning of a target task. Successful transfer learning can result in improved performance, faster convergence, and reduced demand for data. This technique is particularly desirable for the task of brain decoding in the domain of functional magnetic resonance imaging (fMRI), wherein even the most modern machine learning methods can struggle to decode labelled features of brain images. This challenge is due to the highly complex underlying signal, physical and neurological differences between …

Neural Tabula Rasa: Foundations For Realistic Memories And Learning, Patrick R. Perrine

Master's Theses

Understanding how neural systems perform memorization and inductive learning tasks are of key interest in the field of computational neuroscience. Similarly, inductive learning tasks are the focus within the field of machine learning, which has seen rapid growth and innovation utilizing feedforward neural networks. However, there have also been concerns regarding the precipitous nature of such efforts, specifically in the area of deep learning. As a result, we revisit the foundation of the artificial neural network to better incorporate current knowledge of the brain from computational neuroscience. More specifically, a random graph was chosen to model a neural system. This …

Deep Hybrid Modeling Of Neuronal Dynamics Using Generative Adversarial Networks, Soheil Saghafi

Dissertations

Mechanistic modeling and machine learning methods are powerful techniques for approximating biological systems and making accurate predictions from data. However, when used in isolation these approaches suffer from distinct shortcomings: model and parameter uncertainty limit mechanistic modeling, whereas machine learning methods disregard the underlying biophysical mechanisms. This dissertation constructs Deep Hybrid Models that address these shortcomings by combining deep learning with mechanistic modeling. In particular, this dissertation uses Generative Adversarial Networks (GANs) to provide an inverse mapping of data to mechanistic models and identifies the distributions of mechanistic model parameters coherent to the data.

Chapter 1 provides background information on …

A Dynamical Model Of Binding In Visual Cortex During Incremental Grouping And Search, Daniel Schmid, Daniel A. Braun, Heiko Neumann

MODVIS Workshop

Binding of visual information is crucial for several perceptual tasks. To incrementally group an object, elements in a space-feature neighborhood need to be bound together starting from an attended location (Roelfsema, TICS, 2005). To perform visual search, candidate locations and cued features must be evaluated conjunctively to retrieve a target (Treisman&Gormican, Psychol Rev, 1988). Despite different requirements on binding, both tasks are solved by the same neural substrate. In a model of perceptual decision-making, we give a mechanistic explanation for how this can be achieved. The architecture consists of a visual cortex module and a higher-order thalamic module. While the …

Multimodal Neuron Classification Based On Morphology And Electrophysiology, Aqib Ahmad

Graduate Theses, Dissertations, and Problem Reports (ETD)

Categorizing neurons into different types to understand neural circuits and ultimately brain function is a major challenge in neuroscience. While electrical properties are critical in defining a neuron, its morphology is equally important. Advancements in single-cell analysis methods have allowed neuroscientists to simultaneously capture multiple data modalities from a neuron. We propose a method to classify neurons using both morphological structure and electrophysiology. Current approaches are based on a limited analysis of morphological features. We propose to use a new graph neural network to learn representations that more comprehensively account for the complexity of the shape of neuronal structures. In …

Causal Inference In Psychology And Neuroscience: From Association To Causation, Dehua Liang

Computational and Data Sciences (PhD) Dissertations

In psychology and neuroscience, inferring causality in non-experimental studies is almost taboo, because data in these studies, e.g., survey data and resting-state neuroimaging data, are often contaminated by unmeasured confounders. Psychologists and neuroscientists are often cautious about their results, and reluctant to make false claims about causality in non-experimental studies. Therefore, they adopt less stringent statistical analysis techniques that can only infer associational relations. However, the ambiguity about causality in traditional statistical analysis creates much confusion in interpreting analytical results - some studies make implicit causal claims about their results using words such as “impacts”, “lead to” and “affects”. This …

Contrastive Learning For Unsupervised Auditory Texture Models, Christina Trexler

Computer Science and Computer Engineering Undergraduate Honors Theses

Sounds with a high level of stationarity, also known as sound textures, have perceptually relevant features which can be captured by stimulus-computable models. This makes texture-like sounds, such as those made by rain, wind, and fire, an appealing test case for understanding the underlying mechanisms of auditory recognition. Previous auditory texture models typically measured statistics from auditory filter bank representations, and the statistics they used were somewhat ad-hoc, hand-engineered through a process of trial and error. Here, we investigate whether a better auditory texture representation can be obtained via contrastive learning, taking advantage of the stationarity of auditory textures to …

Is The Selective Tuning Model Of Visual Attention Still Relevant?, John K. Tsotsos

MODVIS Workshop

No abstract provided.

Efficient Reduced Bias Genetic Algorithm For Generic Community Detection Objectives, Aditya Karnam Gururaj Rao

Theses

The problem of community structure identification has been an extensively investigated area for biology, physics, social sciences, and computer science in recent years for studying the properties of networks representing complex relationships. Most traditional methods, such as K-means and hierarchical clustering, are based on the assumption that communities have spherical configurations. Lately, Genetic Algorithms (GA) are being utilized for efficient community detection without imposing sphericity. GAs are machine learning methods which mimic natural selection and scale with the complexity of the network. However, traditional GA approaches employ a representation method that dramatically increases the solution space to be searched by …

Deep Learning-Based Framework For Autism Functional Mri Image Classification, Xin Yang, Saman Sarraf, Ning Zhang

Journal of the Arkansas Academy of Science

The purpose of this paper is to introduce deep learning-based framework LeNet-5 architecture and implement the experiments for functional MRI image classification of Autism spectrum disorder. We implement our experiments under the NVIDIA deep learning GPU Training Systems (DIGITS). By using the Convolutional Neural Network (CNN) LeNet-5 architecture, we successfully classified functional MRI image of Autism spectrum disorder from normal controls. The results show that we obtained satisfactory results for both sensitivity and specificity.

Hpcnmf: A High-Performance Toolbox For Non-Negative Matrix Factorization, Karthik Devarajan, Guoli Wang

COBRA Preprint Series

Non-negative matrix factorization (NMF) is a widely used machine learning algorithm for dimension reduction of large-scale data. It has found successful applications in a variety of fields such as computational biology, neuroscience, natural language processing, information retrieval, image processing and speech recognition. In bioinformatics, for example, it has been used to extract patterns and profiles from genomic and text-mining data as well as in protein sequence and structure analysis. While the scientific performance of NMF is very promising in dealing with high dimensional data sets and complex data structures, its computational cost is high and sometimes could be critical for …

Object Recognition And Visual Search With A Physiologically Grounded Model Of Visual Attention, Frederik Beuth, Fred H. Hamker

MODVIS Workshop

Visual attention models can explain a rich set of physiological data (Reynolds & Heeger, 2009, Neuron), but can rarely link these findings to real-world tasks. Here, we would like to narrow this gap with a novel, physiologically grounded model of visual attention by demonstrating its objects recognition abilities in noisy scenes.

To base the model on physiological data, we used a recently developed microcircuit model of visual attention (Beuth & Hamker, in revision, Vision Res) which explains a large set of attention experiments, e.g. biased competition, modulation of contrast response functions, tuning curves, and surround suppression. Objects are represented by …

Modeling Visual Features To Recognize Biological Motion: A Developmental Approach, Giulio Sandini, Nicoletta Noceti, Alessia Vignolo, Alessandra Sciutti, Francesco Rea, Alessandro Verri, Francesca Odone

MODVIS Workshop

In this work we deal with the problem of designing and developing computational vision models – comparable to the early stages of the human development – using coarse low-level information.

More specifically, we consider a binary classification setting to characterize biological movements with respect to non-biological dynamic events. To this purpose, our model builds on top of the optical flow estimation, and abstract the representation to simulate the limited amount of visual information available at birth. We take inspiration from known biological motion regularities explained by the Two-Thirds Power Law, and design a motion representation that includes different low-level features, …

Learning Emotions: A Software Engine For Simulating Realistic Emotion In Artificial Agents, Douglas Code

Senior Independent Study Theses

This paper outlines a software framework for the simulation of dynamic emotions in simulated agents. This framework acts as a domain-independent, black-box solution for giving actors in games or simulations realistic emotional reactions to events. The emotion management engine provided by the framework uses a modified Fuzzy Logic Adaptive Model of Emotions (FLAME) model, which lets it manage both appraisal of events in relation to an individual’s emotional state, and learning mechanisms through which an individual’s emotional responses to a particular event or object can change over time. In addition to the FLAME model, the engine draws on the design …