Tesi etd-11172015-154738 |
Link copiato negli appunti
Tipo di tesi
Tesi di laurea magistrale
Autore
BASSU, CESARE
URN
etd-11172015-154738
Titolo
Estimation of human image processing time using machine learning tools
Dipartimento
INFORMATICA
Corso di studi
INFORMATICA PER L'ECONOMIA E PER L'AZIENDA (BUSINESS INFORMATICS)
Relatori
relatore Prof. Micheli, Alessio
Parole chiave
- data mining
- decision forest
- estimation
- machine learning
- regression
Data inizio appello
04/12/2015
Consultabilità
Non consultabile
Data di rilascio
04/12/2085
Riassunto
In this thesis it is shown the work produced in the period from April to October 2015 in the company d601, located in Aarhus, the second city of Denmark. d60 is specialized both in software production and in Business Intelligence for big customers (mainly Danish, but also international ones).
In this thesis the work has been done onto data coming from another com- pany, which is specialized in image processing. The customer company is formed by several employees that work with the Photoshop software to edit daily a huge amount of images (about 11.000). The main offered service is to remove the background but requests also can be different, like applying additional effects (e.g. add shadows or merge different images).
The goal of the project is to use the data to estimate the processing time of images' sets for different purposes, and it has been done by developing a regression model that taking in input data concerning the desired output, customers and workers information, is able to predict the processing time. The idea is that the model predicts the processing time for each image and afterwards, by summing these values up, the customer has an estimation of the processing time for a group of images.
This model can be used in two different ways: the first one is for the orders time estimation, which means that when an order arrives the company can use the prediction model and schedule the job looking at the prediction, and the second one is for the workers' bonus. At the end of the month the company is able to estimate for each worker the processing time needed for the work and compares it to the effective time. This estimation is used to calculate the bonus in the salary proportionally to the difference (if the estimation is higher than the effective time). We validated the model using some specifically designed benchmark, giving a trust measure together with the prediction, and the customer can schedule the work using the results.
Since the customer company has several ofices in different countries we decided to develop a cloud solution using Azure Machine Learning: a new tool included in the Microsoft Azure suite. When the model is created Microsoft Azure Machine Learning transforms it into a web service, so it can be reached for the predictions using the APIs. We developed a SSIS (SQL Server Integra- tion Service) package that automatically uploads the data, uses the APIs and downloads the predictions.
La tesi discute l'implementazione e valutazione sperimentale di un modello di Machine Learning presso l'azienda d60(Aarhus, Danimarca) che stima il tempo di processamento delle immagini elaborate da un'azienda specializzata nell'editing.
Tale risultato è stato ottenuto tramite il servizio cloud Microsoft Azure Machine Learning per generare le predizioni del modello.
In this thesis the work has been done onto data coming from another com- pany, which is specialized in image processing. The customer company is formed by several employees that work with the Photoshop software to edit daily a huge amount of images (about 11.000). The main offered service is to remove the background but requests also can be different, like applying additional effects (e.g. add shadows or merge different images).
The goal of the project is to use the data to estimate the processing time of images' sets for different purposes, and it has been done by developing a regression model that taking in input data concerning the desired output, customers and workers information, is able to predict the processing time. The idea is that the model predicts the processing time for each image and afterwards, by summing these values up, the customer has an estimation of the processing time for a group of images.
This model can be used in two different ways: the first one is for the orders time estimation, which means that when an order arrives the company can use the prediction model and schedule the job looking at the prediction, and the second one is for the workers' bonus. At the end of the month the company is able to estimate for each worker the processing time needed for the work and compares it to the effective time. This estimation is used to calculate the bonus in the salary proportionally to the difference (if the estimation is higher than the effective time). We validated the model using some specifically designed benchmark, giving a trust measure together with the prediction, and the customer can schedule the work using the results.
Since the customer company has several ofices in different countries we decided to develop a cloud solution using Azure Machine Learning: a new tool included in the Microsoft Azure suite. When the model is created Microsoft Azure Machine Learning transforms it into a web service, so it can be reached for the predictions using the APIs. We developed a SSIS (SQL Server Integra- tion Service) package that automatically uploads the data, uses the APIs and downloads the predictions.
La tesi discute l'implementazione e valutazione sperimentale di un modello di Machine Learning presso l'azienda d60(Aarhus, Danimarca) che stima il tempo di processamento delle immagini elaborate da un'azienda specializzata nell'editing.
Tale risultato è stato ottenuto tramite il servizio cloud Microsoft Azure Machine Learning per generare le predizioni del modello.
File
| Nome file | Dimensione |
|---|---|
Tesi non consultabile. |
|