Testing Federated Learning on a hard non-iid setting:

learning a model for driver distraction prediction

Candidato: Vincenzo Gargano

Relatori: Marco Botta, Roberto Esposito

Università degli studi di Torino

November 2022

Candidato: Vincenzo Gargano Relatori: MarcoTesting

Botta,Federated
Roberto Esposito
Learning(Università
on a hard non-iid
degli studi
setting:
di Torino)
learning
November
a model
2022
for driver distraction
1/1
 Contents

Candidato: Vincenzo Gargano Relatori: MarcoTesting

Botta,Federated
Roberto Esposito
Learning(Università
on a hard non-iid
degli studi
setting:
di Torino)
learning
November
a model
2022
for driver distraction
2/1
 Driver Distraction

Driver distraction is a major cause of road accidents.

Driver distraction is a complex phenomenon that can be caused by a
wide range of factors. And sometime we’re even not aware of it.

Candidato: Vincenzo Gargano Relatori: MarcoTesting

Botta,Federated
Roberto Esposito
Learning(Università
on a hard non-iid
degli studi
setting:
di Torino)
learning
November
a model
2022
for driver distraction
3/1
 What causes driver distraction?
Visual distractions, such as looking at a billboard or a sign, talking to
a passenger.
Biomechanical where drivers are doing something that involves
moving their body, such as reaching for something, eating or drinking
and much more
Cognitive and Emotional distractions, such as daydreaming or
thinking about something else. (We can’t predict those)
Driver health distractions, such as fatigue, illness, or drugs and
medications effects. (We can’t directly predict those)

Candidato: Vincenzo Gargano Relatori: MarcoTesting

Botta,Federated
Roberto Esposito
Learning(Università
on a hard non-iid
degli studi
setting:
di Torino)
learning
November
a model
2022
for driver distraction
4/1
 How to solve driver distraction?
Approach to driver distraction with ML:
Use a model to predict driver distraction, using data directly from the
drivers. But there is a flaw, the privacy concerns of the end user.

Images/gdpr_new-dawn_fromSymantec.jpg

Candidato: Vincenzo Gargano Relatori: MarcoTesting

Botta,Federated
Roberto Esposito
Learning(Università
on a hard non-iid
degli studi
setting:
di Torino)
learning
November
a model
2022
for driver distraction
5/1
 Introduzione al Federated Learning
Federated Learning Cross-Device (FL) is a machine learning technique that
allows to train a model on distributed data without sharing it.

Images/FL_DiagramUpdated_0.2.drawio.png

Candidato: Vincenzo Gargano Relatori: MarcoTesting

Botta,Federated
Roberto Esposito
Learning(Università
on a hard non-iid
degli studi
setting:
di Torino)
learning
November
a model
2022
for driver distraction
6/1
 Federated Learning Simulation

Candidato: Vincenzo Gargano Relatori: MarcoTesting

Botta,Federated
Roberto Esposito
Learning(Università
on a hard non-iid
degli studi
setting:
di Torino)
learning
November
a model
2022
for driver distraction
7/1
 Risultati

Candidato: Vincenzo Gargano Relatori: MarcoTesting

Botta,Federated
Roberto Esposito
Learning(Università
on a hard non-iid
degli studi
setting:
di Torino)
learning
November
a model
2022
for driver distraction
8/1
Grazie per l’attenzione