Doing a master thesis about Unsupervised Object Segmentation

Master Thesis at DTU

Co-founded a start up in Denmark that deploy Predictive Maintenance Solution in Industries using AI and Machine Learning

Co-Founder

From the beginning of my studies in data science i used pandas because it's a very good library for data manipulation. Indeed it's not the perfect library when handling big big data because it doesn't support parallelism and put everything in ram but it is good for data until 1-2 milions of rows i suppose. I've solved the parallel problem by myself implementing a decorator function that auto apply a transformation column function using all the cores available.
I can't say easily all the experiences with this library because any time i've to build a machine learning model i used pandas, other times simply to make plots and to visualize some stats, in some way i suppose like most people use excel

Built complex Neural Networks for Unsupervised Object Segmentation (e.g. MONet), Variational AutoEncoder for anomaly detection, Deep Reinforcment Q-Learner for a Card Game called Briscola

Simple is beautiful

Michele De Vita

The Paper “Generalization and Robustness Implications in Object-Centric Learning” is about object-centric learning which is a type of machine learning that focuses on learning representations of objects in images or videos. The paper discusses the generalization and robustness implications of object-centric learning in an out-of-sample context. I implemented the models MONet, Slot-Attention and GENESIS and I reproduced their result. Furthermore, I designed the algorithms to assess the mask accuracy procedure and the degree of disentanglement of the latent space

Personal details

Michele V.

Summary

Technical skills

Work Experience

Personal Projects