PORTALE DELLA DIDATTICA

Ricerca CERCA
  KEYWORD

XAI methods for image classification

azienda Tesi esterna in azienda    


Riferimenti FABRIZIO LAMBERTI

Riferimenti esterni Lia Morra

Gruppi di ricerca GR-09 - GRAphics and INtelligent Systems - GRAINS

Tipo tesi TESI IN AZIENDA

Descrizione One of the main objectives to eXplainable Artificial Intelligence (XAI) is to provide effective explanations for black-box classifiers. State-of-the-art XAI approaches for image classification typically produce masks or heatmaps, e.g. to highlight which pixels contribute to the classification. Multiple thesis topics are available to extend this approach so that image classification models can be explained in a deeper, more detailed way. Specifically, the research activities will entail either: (a) combining existing methods with generative models, for instance using inpainting on masked images or studying the impact of GAN dissection (https://arxiv.org/pdf/1811.10597.pdf) or (b) extending state-of-the-art methods to encompass and explain high-level features or components, in addition to the final classification. Strong programming (Python) and analytical skills are required. Keras/Tensorflow expertise is preferred and will be acquired. The thesis will be co-supervised by the ISI Foundation.

Vedi anche  http://grains.polito.it/work.php


Scadenza validita proposta 06/02/2021      PROPONI LA TUA CANDIDATURA




© Politecnico di Torino
Corso Duca degli Abruzzi, 24 - 10129 Torino, ITALY
Contatti