Neural Networks and Applications

A Local Hebbian Rule Based Model of Invariant Object Representation and Classification.

This paper introduces a biologically realistic local learning rule that enables recurrent networks to achieve invariant object representation and classification without error backpropagation.

Structured random receptive fields enable informative sensory encodings

The paper models neuronal receptive fields as random samples from distributions, showing improved performance in sensory encoding tasks.

Biological connectomes as a representation for the architecture of artificial neural networks

The paper explores how biological connectomes can inform artificial neural network design, showing that architectural statistics from biological circuits provide a useful prior for machine learning tasks.

Differential geometry methods for constructing manifold-targeted recurrent neural networks.

This paper introduces a method using differential geometry to construct recurrent neural networks with dynamics targeted to a specific manifold, improving the design of task-solving neural dynamics.

Recurrent neural networks learn robust representations by dynamically balancing compression and expansion

This study shows how recurrent neural networks balance compression and expansion of activity dimensionality during learning to solve discrimination tasks effectively.

Operative dimensions in unconstrained connectivity of recurrent neural networks

This paper identifies low-dimensional 'operative dimensions' in RNNs that significantly influence dynamics, showing that much of the high-dimensional connectivity is functionally irrelevant.

Developmental and evolutionary constraints on olfactory circuit selection

The paper predicts neural circuit structure across species by optimizing a three-layer neural network to achieve efficient learning, recovering mammalian 3/2 and invertebrate 7/2 scaling laws.

Active Predictive Coding Networks: A Neural Solution to the Problem of Learning Reference Frames and Part-Whole Hierarchies

This paper introduces APCNs, a neural network framework that learns reference frames and parses visual scenes into part-whole hierarchies, demonstrating compositional representation and transfer learning on standard datasets.

Neural circuits for dynamics-based segmentation of time series

The paper introduces two biologically plausible algorithms for segmenting time series based on underlying dynamics, showing good performance on autoregressive models and voice recordings.

Confidence-controlled Hebbian learning efficiently extracts category membership from stimuli encoded in view of a categorization task

The paper introduces a confidence-controlled Hebbian learning rule that efficiently extracts category membership from optimized stimulus encoding layers, outperforming reward-modulated Hebbian learning.

Gated recurrence enables simple and accurate sequence prediction in stochastic, changing, and structured environments

The paper introduces a recurrent neural network architecture that enables accurate sequence prediction in stochastic and structured environments through gating, lateral connections, and recurrent weight training.

Reset Networks: Emergent Topography by Composition of Convolutional Neural Networks

The paper introduces a Reset model composed of convolutional neural networks that self-organize into clusters and exhibit topographic organization similar to parietal cortex number representation.

The Information Theory of Developmental Pruning: Optimizing Global Network Architecture Using Local Synaptic Rules

The paper shows how local synaptic pruning rules can optimize global network architecture in a deep Boltzmann machine model of sensory encoding.

Weight Perturbation Learning Performs Similarly or Better than Node Perturbation on Broad Classes of Temporally Extended Tasks

This paper compares weight perturbation and node perturbation learning rules on temporally extended tasks, finding that weight perturbation can perform as well or better under certain conditions.

A correspondence between normalization strategies in artificial and biological neural networks

This paper explores parallels between normalization strategies in artificial and biological neural networks, arguing for functional equivalence and proposing a new algorithm inspired by synaptic scaling.

Autoencoder networks extract latent variables and encode these variables in their connectomes

The paper uses autoencoder networks to show that regularization can resolve ambiguity in connectivity weights, allowing latent variable extraction through nonlinear dimensionality reduction.

Dendritic normalisation improves learning in sparsely connected artificial neural networks

The paper introduces dendritic normalisation to improve learning in sparsely connected neural networks, showing significant performance gains over existing methods.

Artificial Neurovascular Network (ANVN) to Study the Accuracy Vs. Efficiency Trade-Off in an Energy Dependent Neural Network

The paper constructs an Artificial Neurovascular Network (ANVN) to study how training both neural and vascular networks affects performance and energy efficiency.

The Gradient Clusteron: A model neuron that learnsvia dendritic nonlinearities, structural plasticity, and gradient descent

The paper introduces the gradient clusteron (G-clusteron), a model neuron that uses gradient descent and structural plasticity to solve classification tasks, achieving comparable performance to logistic regression on MNIST.

Complex Computation from Developmental Priors

This paper introduces a developmental framework for machine learning that mimics neuroevolutionary principles to improve task performance and metalearning capabilities.

Encoding innate ability through a genomic bottleneck

The paper explores how neural circuits can be compressed through a genomic bottleneck, showing that this compression can enhance transfer learning and suggest implications for AI design.

A rapid and efficient learning rule for biological neural circuits

This paper introduces DGNs, a biologically plausible learning rule that outperforms backpropagation in efficiency and forgetting resistance, inspired by cerebellar structure and function.

Neural state space alignment for magnitude generalization in humans and recurrent networks

This study shows that both humans and recurrent neural networks develop aligned neural number lines to generalize magnitude concepts across contexts.

Bio-instantiated recurrent neural networks

This paper explores bio-instantiated recurrent neural networks, examining their performance in working memory tasks and demonstrating how empirical neural network data can be used to construct biologically realistic topologies.

Efficiency of local learning rules in threshold-linear associative networks

This paper demonstrates that Hebbian learning in threshold-linear associative networks can achieve higher storage capacity than binary networks and surpass theoretical bounds through pattern sparsification.

Learning to live with Dale's principle: ANNs with separate excitatory and inhibitory units

Splitting network units into excitatory and inhibitory populations need not cost learning: initialize inhibition to centre and standardize excitatory activity, and scale inhibitory updates by Fisher information.

The covariance perceptron: A new paradigm for classification and processing of time series in recurrent neuronal networks

The paper introduces a new classification framework for time series data using covariance patterns in recurrent neuronal networks, showing it can effectively capture second-order statistical features.

Efficient Similarity-Preserving Unsupervised Learning using Modular Sparse Distributed Codes and Novelty-Contingent Noise

The paper introduces a novel unsupervised learning algorithm using modular sparse distributed codes that preserves similarity and enables fixed-time learning and retrieval.

Nonlinear Reconfiguration of Network Edges, Topology and Information Content During an Artifical Learning Task

This study examines how network structure and information content change during an artificial learning task, identifying three phases of network reconfiguration.

Developing a novel recurrent neural network architecture with fewer parameters and good learning performance

This paper develops a novel RNN architecture, YamRNN, with fewer parameters and faster learning performance compared to LSTM and GRU on sequence classification tasks.