Neural Networks and Applications

Reverse-engineering Recurrent Neural Network solutions to a hierarchical inference task for mice

This paper studies how RNNs solve a hierarchical inference task with two latent variables and disparate timescales, finding they learn behavior similar to Bayesian baselines and develop interpretable models through knowledge distillation.

Adaptive Tuning Curve Widths Improve Sample Efficient Learning

This paper introduces a learning algorithm that achieves high sample efficiency using bio-plausible mechanisms, showing that adaptive tuning curve widths improve both sample efficiency and final precision.

One-time learning in a biologically-inspired Salience-affected Artificial Neural Network (SANN)

This paper introduces a Salience Affected Artificial Neural Network (SANN) that enables one-time learning through salience tagging, improving classification accuracy on MNIST and Fashion MNIST datasets.

Separability and Geometry of Object Manifolds in Deep Neural Networks

This study examines how classification capacity in deep neural networks improves with hierarchy, showing that changes in object manifold geometry underlie this improvement.

Direct-fit to nature: an evolutionary perspective on biological (and artificial) neural networks

The paper compares evolutionary processes and over-parameterized neural networks, arguing that both use blind fitting to solve complex tasks without explicit representation.

Meta-Learning Biologically Plausible Semi-Supervised Update Rules

This paper proposes a meta-learning approach to design biologically plausible synaptic update rules for semi-supervised learning, showing effectiveness on classification tasks.

Computations, optimization and tuning of deep feedforward neural networks

The paper provides an overview of computations, optimization, and tuning in deep feedforward neural networks, using a small network to explain generalizations for larger networks.

Natural representation of composite data with replicated autoencoders

This paper introduces a method using replicated autoencoders optimized for local entropy to infer compositional features from noisy data, showing improved performance on synthetic and protein sequence data.

A Critique of Pure Learning: What Artificial Neural Networks can Learn from Animal Brains

This paper argues that animal learning is not primarily driven by sophisticated algorithms but by innate, structured brain connectivity, suggesting a new direction for improving artificial neural networks.

The development, recognition, and learning mechanisms of an animal-like neural network

This paper introduces an animal-like neural network (RRN) that mimics biological recognition mechanisms, showing it can maintain object invariance and learn without disrupting existing functions.

Local online learning in recurrent networks with random feedback

This paper introduces a biologically plausible learning rule for recurrent neural networks that uses local information and random feedback to train networks for time-dependent tasks.

Fundamental bounds on learning performance in neural circuits

The paper shows that neural circuits have an optimal size for learning performance, where larger circuits can either enhance complex behaviors or improve learning of simpler tasks, but beyond a point, increased size impairs learning due to noise.

Unsupervised deep learning on biomedical data with BoltzmannMachines.jl

The paper introduces a Julia package for deep Boltzmann machines to enable unsupervised learning on biomedical data.

Deep Neural Networks In Computational Neuroscience

This paper discusses how deep neural networks can model neural responses and cognitive functions, highlighting their accuracy and potential in computational neuroscience.

Manifold-tiling Localized Receptive Fields are Optimal in Similarity-preserving Neural Networks

The paper shows that similarity-preserving neural networks develop localized receptive fields that tile manifolds, mirroring brain representations like place cells.

Neural classifiers with limited connectivity and recurrent readouts

This paper proposes recurrent attractor neural networks with sparse connectivity to overcome classification limitations in feed-forward networks with limited neuron connections.

Why Does the Neocortex Have Columns, A Theory of Learning the Structure of the World

The paper proposes a network model of neocortical columns and layers that learns object recognition through temporal integration and lateral connections, showing improved performance with multi-column networks.

An approximation of the error back-propagation algorithm in a predictive coding network with local Hebbian synaptic plasticity

This paper shows that a predictive coding network with local Hebbian plasticity can approximate error back-propagation for supervised learning without external control.