Parallel Implementation of Machine Learning Algorithms and Neural Networks with Noir - Academic Thesis

Aug 2, 2026

Upload by Rowan

The rapid growth of data in the digital era accelerates innovation in data science, machine learning, and artificial intelligence, but it also increases computational cost, training time, and can exceed the memory or storage limits of single machines. Distributed processing enables parallel execution across multiple computing nodes, motivating the use of the Noir framework for big data. The work implements common machine learning algorithms and dense neural networks in a parallel-distributed setting, evaluating how effectively each algorithm maps to distributed data and whether Noir improves computation speed versus frameworks such as Scikit-learn and TensorFlow without reducing quality or efficiency.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages98

Reading Time4 h 7 min

AccessFree to download

Introduction

State of the art

Research objectives

Background and Problem Statement

Introduction to Noir

Problem Statement

Algorithms

Linear Regression

Logistic Regression

Random Forest

Naive Bayes

Support Vector Machine

Gradient Boosting

Singular Value Decomposition

K-Nearest Neighbors

K-Means

Artificial Neural Networks

Implementation

Linear Regression

Logistic Regression

Random Forest

Naive Bayes

Support Vector Machine

K-Nearest Neighbors

K-Means

Artificial Neural Networks

Parallel Implementation of Machine Learning Algorithms and Neural Networks with Noir Tesi di Laurea Magistrale in Computer Science and Mathematical Engineering Ingegneria Informatica e Ingegneria Matematica Authors: Adalberto Trevisani, Christopher Volpi Student ID: 993218, 991748 Advisor: Prof. Alessandro Margara Co-advisors: Prof. Gianpaolo Cugola, Luca De Martini Academic Year: 2022-23 i Abstract In the current digital era, the exponential growth of data presents both immense opportunities and challenges. This increment in data volume has fueled innovation in data science, machine learning, and artificial intelligence. Machine learning algorithms and neural networks excel with significant data but face challenges such as increased computational cost and extended training times. Moreover, large datasets may exceed the memory or storage capacities of individual machines. To address these challenges, distributed processing paradigms enable parallel execution across multiple computing nodes. The Noir framework, designed for big data processing, serves as the basis for our investigation. Our research focuses on practically implementing well-known machine learning algorithms and dense neural networks in a parallel-distributed context. We aim to assess how effectively each algorithm integrates into a data distribution setup and whether Noir can enhance computation speed compared to popular frameworks like Scikit-learn and TensorFlow, without compromising algorithm quality and efficiency. Keywords: Machine Learning, Parallel Computing, Neural Networks, Distributed Systems, Big Data, Rust Abstract in lingua italiana Nell’attuale era digitale, la crescita esponenziale dei dati presenta sia immense opportunità che sfide significative. Questo aumento nel volume dei dati ha alimentato interesse einnovazione nell’ambito della Data Science e dell’Intelligenza Artificale. Gli algoritmi di machine learning e le reti neurali eccellono con dati significativi, ma affrontano sfide come il crescente costo computazionale e dei tempi di addestramento. Inoltre, grandi quantità didati possono superare le capacità di memoria o di archiviazione delle singole macchine. Per affrontare queste sfide, i paradigmi di "parallel processing" consentono l’esecuzione parallela su più nodi di calcolo. Il framework Noir, progettato per l’elaborazione di big data, serve come base per la nostra indagine. La nostra ricerca si concentra sull’implementazione pratica di noti algoritmi di Machine Learning e reti neurali in un contesto distribuito in parallelo. Abbiamo l’obiettivo di valutare quanto efficacemente ogni algoritmo si integrain un contesto di distribuzione dei dati e se Noir possa migliorare la velocità di calcolorispetto a framework popolari come Scikit-learn e TensorFlow, senza compromettere laqualità ed efficienza degli algoritmi. Keywords: Machine Learning, Calcolo Parallelo, Reti Neurali, Sistemi Distribuiti, Big Data, Rust v Contents Abstract i Abstract in lingua italiana iii Contents v 1 Introduction 1 1.1 State of the art ................................. 2 1.2 Research objectives ............................... 4 2 Background and Problem Statement 5 2.1 Introduction to Noir .............................. 5 2.1.1 Dataflow Paradigm ........................... 5 2.1.2 Noir Architecture and Usage ...................... 6 2.1.3 Key Operators ............................. 8 2.2 Problem Statement ............................... 11 3 Algorithms 13 3.1 Linear Regression ................................ 13 3.1.1 Gradient Descent ............................ 14 3.1.2 Ordinary Least Squares ........................ 14 3.2 Logistic Regression ............................... 16 3.2.1 Multinomial Logistic Regression .................... 16 3.3 Random Forest ................................. 19 3.3.1 Decision Tree .............................. 19 3.3.2 Impurity Indexes .......................

FAQs

What problem does the thesis address?

It addresses the challenge of running machine learning and neural networks efficiently as dataset sizes grow, leading to higher computational cost and longer training times and potential memory limits on single machines.

How does the work use the Noir framework?

Noir is used as the basis for implementing machine learning algorithms and dense neural networks in a parallel-distributed context for big data processing.

What does the thesis evaluate when comparing Noir to other frameworks?

It evaluates how well each algorithm integrates with a distributed data setup and whether Noir can improve computation speed compared with frameworks like Scikit-learn and TensorFlow without compromising algorithm quality and efficiency.

Aug 2, 2026

Upload by Rowan