Skip to content

A collection of projects introducing neural networks and data analysis concepts. From search and genetic algorithms to autoencoders and VAEs.

Notifications You must be signed in to change notification settings

AxelPreitiT/nn-intro

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

72.27 - Sistemas de Inteligencia Artificial

Group 07

Projects

TP00: Introduction to Data Analysis

Utilizing pandas and Plotly, this Python project assesses Pokemon capture outcomes by considering diverse parameters and Pokeball selections. Through simulations and Plotly-generated graphs, it analyzes factors like Pokemon health, level, and ball efficacy, offering insights into optimal capture strategies.

TP01: Sokoban Puzzle Solver with Search Methods

This project revolves around the creation of a Sokoban puzzle solver through the implementation of various search algorithms, including BFS, DFS, Greedy and A*, while using admissible heuristics to optimize solution strategies.

TP02: RPG Character Optimization with Genetic Algorithms

This project is dedicated to optimizing character configurations for a role-playing game (RPG) using genetic algorithms. Characters belong to four distinct classes: Warriors, Archers, Defenders, and Assassins, each with specific performance objectives for attack and defense.
The genetic algorithm implementation includes various genetic operators (crossover and mutation), selection and replacement strategies, and termination criteria. All algorithm parameters are specified in an external configuration file for flexibility and experimentation.

TP03: Simple and Multilayer Perceptron

This project involves implementing simple and multilayer perceptrons to solve various problems. It explores logical operations, classification tasks, and digit recognition using neural networks. The project aims to understand the capabilities of these neural models and their generalization performance.
This project provides a comprehensive exploration of neural networks, from basic single-layer perceptrons to more advanced multilayer architectures. It aims to enhance understanding of neural network capabilities and their application in real-world problem-solving scenarios.

TP04: Unsupervised Learning

This project focuses on unsupervised learning and comprises various exercises involving Kohonen networks, Principal Component Analysis (PCA), and pattern recognition using the Hopfield model.
The exercises explore clustering countries based on characteristics, computing principal components, and recognizing noisy patterns

TP05: Autoencoder and Variational Autoencoder (VAE)

This project involved using the previous implementations of Multilayer Perceptron to implement an autoencoder, with a latent variable dimension of size 2, for reproducing 35 pixel letters with at most one pixel off. Then we redefined the architecture to build a variational autoencoder that was used to generate generate new emojis through the continuos distribuition generated by the the decoder section of the VAE. Both models were trained without external libraries beyond numpy, writing ourselves the backpropagation algorithm and the ADAM optimizer.

Members

Releases

No releases published

Packages

No packages published

Languages

  • Python 99.1%
  • Other 0.9%