Ilaria Scarabottolo

I am a third-year Ph.D. student at Università della Svizzera italiana (Lugano, Switzerland), under the supervision of Prof. Laura Pozzi.

I am curious and interested about everything related to applied Mathematics, and my education and life choices brought me to the fascinating field of Approximate Computing and, in particular, to the design of approximate hardware for embedded systems.

My research work has mostly focused on logic simplification for the design of inexact, energy-efficient circuits, and on the identification of generic algorithms to derive error models of faulty hardware. I have also worked on fault-detection in distributed systems and signal processing, mainly during my master thesis at Politecnico di Milano (Italy).

More recently, I have also approached the boolean satisfiability problem applied to logic simplification for approximate circuits, as well as a probabilistic approach for error model derivation.

Research

Since the beginning of my school career I have been attracted to engineering subjects and to the idea of contributing to a sustainable and efficient technological development. My research direction stems from this deep interest and lands in the field of Approximate Computing, a rising computing paradigm that promotes informed trade-offs between energy efficiency and accuracy.

energy efficiency vs accuracy Indeed, a vast number of applications do not need fully-precise computations to produce satisfying results. A few of many examples are media processing applications, where a small quality loss can be afforded due to the impossibility for human users to perceive it; wireless sensor networks, where data can be already corrupted due to noise and redundant acquisitions; recommender systems that do not have a single golden answer; heuristic and probabilistic algorithms that deal with average values and do not even aim at obtaining fully precise results. On the other hand, the majority of those applications is power-hungry and energy consumption has to be carefully controlled.

My research project aims at obtaining energy-efficient inexact circuits, starting from their exact counterparts and given an error level tolerance. These hardware components can be employed in any error-resilient application, entailing remarkable gains in energy savings by slightly reducing the accuracy level.

During the first years of PhD, I have focused on developing algorithms to automatically synthesize such circuits, without needing a-priori knowledge of the specific problem. Along with identifying most efficient faulty circuits, I have studied methodologies to derive error-models for circuits or larger systems, such as Neural Netwok or FPGAs. I am currently working on integrating probabilistic analysis to such algorithms, since they represent an interesting and valid opportunity to further improve energy efficiency. Finally, I am also exploring different ways of approaching error-modeling problems, by expressing them in form of satisfiability problems.

Publications

Partition and Propagate: an Error Derivation Algorithm for the Design of Approximate Circuits

Ilaria Scarabottolo, Giovanni Ansaloni, George Constantinides and Laura Pozzi

This work aims at deriving an exhaustive circuit error-model, in terms of identification of the influence of each circuit gate (or component) on the final output. The maximum error that can be seen on the circuit primary output if a gate is removed is obtained through a novel partitiong algorithm: by partitioning the original graph representing the circuit into sub-circuits, and by studying each subgraph truth table separately, it is possible to efficiently obtain an accurate estimate of such influence. The novel partitioning algorithm developed exploits the graph topology and monotonicity of subgraphs to improve the result accuracy.

DAC - June 2019 graph partition

Circuit Carving: A Methodology for the Design of Approximate Hardware

Ilaria Scarabottolo, Giovanni Ansaloni and Laura Pozzi

A novel algorithm that aims at synthesising the most efficient circuits given a predefined error threshold, which denotes the maximum tolerated error for that specific application. By considering a circuit netlist, along with the given error threshold, Circuit Carving performs a binary-tree search on the graph nodes to find the maximum subgraph (i.e., subcircuit) that can be eliminated from the circuit without overcoming the abovementioned threshold. Pruning criteria and, in particular, a definition of closure property for subgraphs are exploited to bound the exploration.

DATE - March 2018 binary tree search

A spectrum-based adaptive sampling algorithm for smart sensing

Ilaria Scarabottolo, Cesare Alippi and Manuel Roveri

This work is the subject of my master thesis and presents a novel algorithm based on change-detection tests applied on signal frequency bands to identify anomalies in the signal spectrum and adapt the sampling frequency accordingly. This approach can be employed to monitor changes in a phenomenon observed with wireless sensors, where energy savings are vital for a sustainable infrastructure. A first version of the algorithm has also been tested on programmable boards by STMicroelectronics, confirming its efficiency and validity when exposed to real sensors measuring temperature variations.

IEEE SmartWorld - July 2017 frequency bands for change detection

Work-in-Progress: A Partitioning Strategy for exploring Error-Resilience in Circuits

Ilaria Scarabottolo, Giovanni Ansaloni and Laura Pozzi

Early stage of a more detailed work aimed at finding error-models of circuits through a design-and-conquer approach.

CASES - October 2018

CV

PhD student at Università della Svizzera italiana, Faculty of Informatics

Lugano, Switzerland

Supervisor: Prof. Laura Pozzi.

Research area: approximate computing, design of highly efficient inexact hardware.

September 2016 - Present

Visiting PhD student at Imperial College London

London, United Kingdom

Host supervisor: Prof. George Constantinides.

Research area: approximate computing, boolean satisfiability for hardware error-modeling.

February 2019 - Present

Master degree in Computer Science Engineering at Politecnico di Milano

Milano, Italy

Thesis: “A spectrum-based adaptive sampling mechanism for energy conservation in Cyber-Physical Systems”. Later published in the IEEE Smartworld 2018 Conference transactions.

Thesis supervisors: Prof. Cesare Alippi, Prof. Manuel Roveri.

Final grade summa cum laude.

July 2016

Bachelor degree in Mathematical Engineering at Politecnico di Milano

Milano, Italy

Thesis: “Algorithmics: the spirit of computing”.

Final grade summa cum laude.

September 2013

Double degree at École Centrale Paris

Paris, France

Double master degree international program T.I.M.E. (Top Industrial Managers for Europe).

August 2011 - September 2013

Diploma: Scientific high school Alessandro Volta

Milano, Italy

Final grade: 98/100
Italian winner of EUCYS Contest (European Union Contest for Young Scientists) in fourth year.

July 2009

Teaching

Automata and Formal Languages

Teaching Assistant

Course resposible: Prof. Laura Pozzi

Fall semester 2017 and 2018

Software Atélier: Human-Computer Interaction

Teaching Assistant

Course responsible: dr. M. Landoni

Spring semester 2016 and 2017

Awards & Grants

Academic Activities

As a girl in computer science, I have organized and participated to several events for scientific career-promotion among young girls, strongly believing that soon these events will become unnecessary.