cv

Research Assistant

Indiana University–Purdue University Indianapolis, Indianapolis (IN)

• Classification of mineral signatures in hyperspectral images from the CRISM Mars experiment using Bayesian models, deep models, SVM, random forests
• Segmentation of different cell and nerve structures in neurohistological TEM images using U-Net
• Investigating Bayesian models for problems with imbalanced classes and zero-shot learning in segmentation and hyperspectral classification
• Automated tagging and summarization for medical notes using recurrent neural networks and Transformer models
• Design and implement an Android app giving context-based message suggestions to blind/visual impaired people

Research Scientist Intern

Meta, Redmond (WA)

• Modelling uncertainty in deep eye tracking models.
• Develop out-of-distribution approaches to identify anomalous training sequences

Advanced Research Engineer

STMicroelectronics, Agrate (MB), Italy

• Design and development of the X-Cube-AI backend for porting neural networks form different toolboxes to STM platforms
• Development in ANSI C and training in Keras (Python) of deep learning solutions for sensor data classification and activity recognition on microcontrollers
• Design and training of binary-valued neural networks for activity recognition
• Development, implementation and optimization of computer vision algorithms for visual search and object detection
• Meeting with customers and collecting requirements on solutions using machine learning and AI on microcontrollers and sensors
• Supervised interns on the development of algorithms for object detection, Visual Search and prediction on Cyber Physical Systems
• Person in charge of the deliverables assigned to ST Microelectronics in the European project COSSIM H2020 (Feb 2015-Feb 2018)

Research Assistant

Politecnico di Milano, Milan, Italy

• Person in charge (under supervision of the research advisor) of the 8 deliverables assigned to Politecnico di Milano in the European project ASTUTE ARTEMIS (Feb 2011-Feb 2014)
• Development and implementation of computer vision algorithms targeting the MPEG CDVS standard
• Development of several MPEG CDVS demonstrators based on Visual Search algorithms in C++, Linux (x86), Windows (x86), MacOS (x86), Android 4 (ARM, Nexus 4, Nexus 7, Samsung S2, S3, S4), with Python and helper scripts
• Supervised MSc Students theses about: 1) using trifocal tensor on omnidirectional images to implement a visual odometry algorithm; 2) implementing robust estimation techniques to detect geometric planes in point clouds

Research contractor

STMicroelectronics, Agrate (MB), Italy

Development and implementation of algorithms related to the MPEG CDVS standard

• Improvement of the code quality in the CDVS Test Model, accepted in the official Test Model 3.1
• Improvement of the localization precision of CDVS by modifying the RANSAC algorithm, accepted in Test Model 3.0
• Optimization of a interest point detector with three times the speed of the reference CDVS algorithm and with comparable accuracy and filed a ST/PoliMi joint patent (Emanuele Plebani, Danilo Pau)
• Collaborated and provided support to ST Imaging Division on adapting algorithms for low precision implementations
• Won the prize for Best Demo in the GTTI MMSP 2014 (Italian Chapter IEEE Signal Processing Society) conference presenting Visual Search demos on x86 and Android platforms
• Contributed in starting the MPEG CDVA standard for search on video streams and object classification, overview of the state of the art and available datasets
• Contributed with 32 input documents to MPEG between July 2012 and April 2014
• Supervised student Daniele Miatto (ST intern) on implementing a client-server architecture based on CDVS Test Model and porting it to Android/ARM
• Supervised graduate and post-graduate students on: 1) parallelizing the optimized interest point detector with Intel TBB, porting the code on Android/ARM and developing a hybrid Visual Search/Visual Odometry algorithm; 2) testing interest point detection algorithms and writing a white paper on the CDVS standard (officially accepted)
• Co-supervised graduate and post-graduate students on: 1) developing a CDVS extension on color images and building a test dataset 2) on developing a compression scheme for dense HOG descriptors and porting the model on C++

Java Developer

Overdev s.r.l., Terno d'Isola (BG), Italy

Development of a Java JEE application for a small manufacturing company

• Design of a relational database in the JBoss application server
• Optimization of database queries for generation of summary reports

Doctor of Philosophy - PhD

Purdue University

Computer Science

Master of Science (M.Sc.)

Politecnico di Milano

Computer Engineering

COSSIM

COSSIM (“Novel, Comprehensible, Ultra-Fast, Security-Aware CPS Simulator”) addresses the need for simulation tools and models for system design and analysis with an open source framework for efficient and accurate Cyber Physical System simulation.

MPEG Compact Descriptors for Video Analysis

MPEG CDVA is exploring the opportunity of standardizing technologies that will enable inter-operable and efficient of video search systems architectures and applications. In particular, it is seeking technologies for video content search and classification.

MPEG Compact Descriptors for Visual Search

MPEG is considering standardizing technologies that will enable efficient and inter-operable design of visual search applications. In particular it is seeking technologies for visual content matching in images or video. Visual content matching includes matching of views of objects, landmarks, and printed documents that is robust to partial occlusions as well as changes in vantage point, camera parameters, and lighting conditions.

ASTUTE

The project aims at the development of an advanced pro-active HMI interfaces and reasoning engine systems for improving the way humans deal with complex information flows, which without assistance would saturate theirs performance and decision-making capabilities. Politecnico di Milano and its Italian partners will apply ASTUTE's results in the automotive field, in order to improve the driver's experience.