Cristiano Malossi

Overview

Title

Principal Research Scientist and Manager

Location

IBM Research Europe - Zurich Zurich, Switzerland

Bio

Cristiano Malossi is Principal Research Scientist and Manager of the AI Automation Group at the IBM Research Laboratory in Zurich. From 2020, Cristiano leads global research and innovation strategy around Enterprise Visual Inspection, with focus on inspection of Large-Scale Infrastructures. Cristiano’s team owns the design, development, and productization of a scalable AI cloud service for detection of small and rare defects in large high-resolution images. A DEMO of the service is available to interested customer, on-request.

Between 2017 and 2019 Cristiano’s has led a global research team around neural network automation. In 2018, Cristiano’s team released on the IBM Cloud the first IBM engine for automation of neural network synthesis (NeuNetS) applicable to image and text classification.

In the earlier stages of his career at IBM, Cristiano has been responsible of the development of energy-aware computing algorithms, as part of the Exa2Green project. Later, between 2017 and 2020, Cristiano has been coordinator of the Open Transprecision Computing (OPRECOMP) project, with focus on low-power/low-energy computing paradigms, based on approximation and transprecision computing - a new computing paradigm created by the project that combines approximation and automation. Cristiano has also written and participated to many other EU projects, including VIMMP, IM-SAFE, and Romeo.

Cristiano is a recipient of the 2016 IPDPS Best Paper Award and the 2015 ACM Gordon Bell Prize. Since 2015 he is also member of ACM, has been ACM Distinguished Speaker, and has served in the technical program committee of top conferences, including SC, ISC, IPDPS, AAAI, NeurIPS, ICML, and DATE.

Before IBM, Cristiano graduated from the Swiss Federal Institute of Technology in Lausanne (EPFL) in Lausanne with a PhD in applied mathematics. In 2013, his thesis on parallel algorithms and mathematical methods for the numerical simulation of cardiovascular problems granted him the IBM Research Prize for Scientific Computing. Cristiano has also a B.Sc. in Aerospace Engineering and a M.Sc. in Aeronautical Engineering from the Politecnico di Milano (Italy).

Cristiano research interests include enterprise visual inspection, acceleration and new computing paradigms for machine learning and deep learning, AI lifecycle automation for enterprise data, AI systems design and user experience, high performance computing, transprecision & energy-aware computing and - from his academic education - CFD, FEM, and aircraft design.

Honors and Awards:

2022 Best Paper Award at CVCIE @ ECCV2022 workshop
2019 & 2022 - ACM Distinguished Speaker
2016 - IEEE/ACM IPDPS Best Paper Award
2015 - IBM Pat Goldberg Memorial Best Paper Award
2015 - ACM Gordon Bell Prize
2013 - IBM Research Prize for Computational Science (for the PhD thesis)

Research in the News:

Swiss-EU Success Story: OPRECOMP (SwissCore - December 2020)
Why Smarter Roads, Bridges, and Tunnels are good for Economies and Societies (Youtube - October 2020)
OPRECOMP, Transprecision computing for energy efficiency (Open Access Government - October 2020)
Artificial intelligence, drones and sensors set to save our crumbling infrastructure (Medium.com - December 2019)
Mit KI und Drohnen auf der Suche nach Brückenschäden (Computerworld - December 2019)
AI for AI: in the middle of the future (Migros Magazin Cover, 3-millions printed copies - May 2019)
Radical computing rethink to save time and energy (EC Research and Innovation Success Stories - February 2019)
NeuNetS: Automating Neural Network Model Synthesis for Broader Adoption of AI (IBM Blog - December 2018)
TAPAS: Frugally Predicting the Accuracy of a Neural Network Prior to Training (IBM Blog - December 2018)
Restoring Balance in Machine Learning Datasets (IBM Blog - October 2018)
Come funzionano le reti neurali (MaddMaths! - October 2017)
The future belongs to cognitive systems (SIX Connect - May 2017)
Gordon Bell Prize Winners Simulate Earth's Mantle (IBM Systems Magazine - November 2016)
Data Centric Systems, la frontiera del supercalcolo (01net. - 6 May 2016)
Trade talk: Serial solver (Nature Careers Q&A - 14 April 2016)
Finding job satisfaction in high performance computing (Naturejobs blog - 13 April 2016)
SC15 Gordon Bell Prize Winners (PR Newswire; IBM Blog; HPCWire - 20 November 2015)
Meet an IBM Researcher (IBM Blog - 6 November 2015)
IBM Research Prize for Computational Science (EPFL News - 10 October 2013)

Selected Publications:

F. Scheidegger, L. Benini, C. Bekas, A. C. I. Malossi. Constrained deep neural network architecture search for IoT devices accounting hardware calibration. NeurIPS - Thirty-third Conference on Neural Information Processing Systems, 2019. (Acceptance rate 21.2% over 6743 reviewed submissions)
R. Istrate, F. Scheidegger, G. Mariani, D. S. Nikolopoulos, C. Bekas, A. C. I. Malossi. TAPAS: Train-less Accuracy Predictor for Architecture Search. AAAI, 2019. (Acceptance rate 16.2% over 7095 reviewed submissions)
P. W. J. Staar, P. K. Barkoutsos, R. Istrate, A. C. I. Malossi, I. Tavernelli, N. Moll, H. Giefers, C. Hagleitner, C. Bekas, A. Curioni. Stochastic Matrix-Function Estimators: Scalable Big-Data Kernels with High Performance. IEEE International Parallel and Distributed Processing Symposium (IPDPS), pp. 812-821, 2016. (Best Paper Winner)
J. Rudi, A. C. I. Malossi, T. Isaac, G. Stadler, M. Gurnis, P. W. J. Staar, Y. Ineichen, C. Bekas, A. Curioni, O. Ghattas.An Extreme-scale Implicit Solver for Complex PDEs: Highly Heterogeneous Flow in Earth's Mantle. Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, pp. 1-12, ACM, 2015. (Winner of the ACM Gordon Bell Price at SC15)
A. C. I. Malossi, P. J. Blanco, P. Crosetto, S. Deparis, A. Quarteroni. Implicit coupling of one-dimensional and three-dimensional blood flow models with compliant vessels. Multiscale Modeling & Simulation 11(2), 474-506, SIAM, 2013.
A. C. I. Malossi, P. J. Blanco, S. Deparis. A two-level time step technique for the partitioned solution of one-dimensional arterial networks. Computer Methods in Applied Mechanics and Engineering 237-240, 212-226, Elsevier, 2012.