Nicola Pezzotti

Research Scientist and Engineer in
Artificial Intelligence and Visual Analytics

Follow me on GitHub


Research Scientist :: Philips Research :: November 2018 - Present

I develop cutting edge AI technology to improve people’s lives. I set up research collaborations, with internal and external partners, to address challenging problems in healthcare.

Responsibilities in 2019:

  • Set up research collaborations with several external parties, including LUMC, UvA, AUMC, Radboud UMC and NKI for the participation in the FastMRI challenge. The two teams won all challenge tracks
  • Lead the AI development for the FastMRI challenge, in a close collaboration with the Philips BIU MR and LUMC. Our team won all multi-coil tracks in the challenge
  • Joined the Kickstart-AI initiative, to set up a joint-appointment program between universities and industrial partners
  • Joined the program committee of the Philips conference on AI

Research Intern :: Google :: February 2018 - June 2018

I work with Alexander Mordvintsev, the creator of Google’s DeepDream project, on the interpretability of Deep Neural Networks, where I published two research papers. I developed a scalable manifold-learning optimization approach that is used for analyzing the output of deep neural networks directly in the web browser. My work is released as a library in the TensorFlow.js family, it featured on the Google AI Blog and among Google AI’s research effort in 2019. Moreover, I published a article on neural network interpretability and generative properties.

Visiting Researcher :: INRIA/AVIZ :: April 2017 - June 2017

I worked with professor Jean-Daniel Fekete on the development of the Progressive Visual Analytics paradigm for the analysis of large data collections. This work powered analytics system for the analysis of deep learning models and large networks.

PhD Student :: TU Delft :: September 2014 - October 2018

My research consists in the development of scalable manifold-learning algorithms for the analysis of extremely large high-dimensional data, such as medical datasets, social-networks and deep neural networks. My algorithms and systems were presented in the most important visual analytics venues and are used by medical researchers for the analysis of real-world data. Thanks to the Hierarchical Stochastic Neighbor Embedding (HSNE) algorithm that I developed, we were able to identify previously unknown immune-system cell types. This result was achieved by scaling up the number of cells that could be analyzed with a manifold-learning algorithm from a few thousands to several millions. HSNE is also the cornerstone for DeepEyes, a system for the visual analysis of deep neural networks during training that I developed. My algorithms are implemented in C++ and are released as part of the High-Dimensional-Inspector library.

Software Engineer :: Open Technologies S.R.L. :: July 2011 - Aug 2014

I was responsible of the development of the high-end real-time scanner Insight3 I optimized the algorithms developed during my Research Fellowship and developed several Arduino-based systems for the on-board control of Insight3. Furthermore, I contributed to the development of the computational-geometry module of the Open Technologies S.R.L. proprietary library and I was in charge of the control versioning and the release of the company’s main software.

Research Fellow :: University of Brescia :: Oct 2011 - Oct 2012

I developed proprietary algorithms for the real-time computation of implicit surfaces on the GPU. These algorithms are designed to work with {off-the-shelf} and real-time scanning devices like the Microsoft Kinect and the PrimeSense Carmine/Capri. Furthermore, I devised a proprietary passive stereo system that led to the development of the Insight3 high-quality real-time scanner. Due to the strict real-time requirements all the developed algorithms were implemented in C++, CUDA and Thrust.

Master Thesis Intern :: Open Technologies S.R.L. :: Jan 2011 - Jul 2011

During my master thesis I worked on the development of fast and automatic tools for the alignment of 3D data such as point clouds, meshes and range images. This work was done in collaboration with the company Open Technologies S.R.L. I graduated with a final grade of 110/110.

Bachelor Thesis Intern :: G2L S.R.L. :: Jul 2008 - Dec 2008

During my bachelor thesis I developed a library for interprocess communication between real-time applications working in Linux-Xenomai and other Linux applications. This work was done in collaboration with the company G2L S.R.L.