Neural Networks

Experiments with neural network technologies

Studies, experiments and examples about deep learning machine models based on different topologies of neural networks: multilayer perceptrons, convolutional and recurrent layers, long-short-term-memory cells. Applications of neural networks to fit mathematical objects, to analyze texts, images, sounds and videos, to search for recurrent patterns in numerical series, to solve differencial equations. Code strictly original written in Python 3 with TensorFlow and/or PyTorch or in Julia with Flux, working and freely available on GitHub.

All Posts about Neural Networks

Pruning of neural networks with TensorFlow: The pruning of the weights of a neural net puts to zero the insignificant weights of the model in phase of training with the purpose to obtain a sure level of sparsity in such way to render the model more easy compressible.
Differential Equations and Neural Networks: This page is an index of posts posts of this website that deal with some specific topic related to differential equations and neural networks.
Experiments with Neural ODEs in Python with TensorFlowDiffEq: Examples of usages of Neural ODEs implemented in Python using TensorFlow 2.x and TensorFlowDiffEq.
Experiments with Neural ODEs in Julia: Examples of usages of Neural ODEs implemented in Julia using the packages DifferentialEquations, Flux, DiffEqFlux of the Julia ecosystem.
Ordinary differential equation solvers in Python: Examples of use of some ordinary differential equation solvers in Python implemented by libraries frequently used in scientific applications in general and expecially in machine learning and deep learning.
Ordinary differential equation solvers in Julia: Examples of use of some ordinary differential equation solvers in Julia implemented by the DifferentialEquations.jl native package of Julia ecosystem.
Common tools for function fitting: This post describes the usage of some general tools for the generation and visualization of datasets that represent or approximate mathematical objects such as functions, curves and surfaces; such tools are used by other programs described in other posts on this website.
Forecast of a univariate equally spaced time series with TensorFlow: Forecast of univariate and equally spaced time series via various neural network taxonomies implemented with TensorFlow without writing code but only via command line.
Fitting with highly configurable multi layer perceptrons: Fitting with highly configurable multi layer perceptrons (MLP) of functions, curves and surfaces with TensorFlow and PyTorch.
One-variable real-valued function fitting with TensorFlow: Highly configurable multilayer perceptron (MLP), implemented in TensorFlow, that fits a curve of a real-valued continuous and limited function defined in a closed interval of the reals.
One-variable real-valued function fitting with PyTorch: Highly configurable multilayer perceptron (MLP), implemented in PyTorch, that fits a curve of a real-valued continuous and limited function defined in a closed interval of the reals.
Parametric curve on plane fitting with TensorFlow: Highly configurable multilayer perceptron (MLP), implemented in TensorFlow, that fits a continuous and limited parametric curve on plane with the parameter belonging to a closed interval of the reals.
Parametric curve on plane fitting with PyTorch: Highly configurable multilayer perceptron (MLP), implemented in PyTorch, that fits a continuous and limited parametric curve on plane with the parameter belonging to a closed interval of the reals.
Parametric curve in space fitting with TensorFlow: Highly configurable multilayer perceptron (MLP), implemented in TensorFlow, that fits a continuous and limited parametric curve in space with the parameter belonging to a closed interval of the reals.
Parametric curve in space fitting with PyTorch: Highly configurable multilayer perceptron (MLP), implemented in PyTorch, that fits a continuous and limited parametric curve in space with the parameter belonging to a closed interval of the reals.
Two-variables real-valued function fitting with TensorFlow: Highly configurable multilayer perceptron (MLP), implemented in TensorFlow<, that fits a surface of a real-valued continuous and limited two variables real function defined over a rectangle.
Two-variables real-valued function fitting with PyTorch: Highly configurable multilayer perceptron (MLP), implemented in PyTorch, that fits a surface of a real-valued continuous and limited two variables real function defined over a rectangle.