Multitask learning and learning-to-learn: a statistical learning perspective
Prof. Massimiliano Pontil , Istituto Italiano di Technologia & University College London
Abstract:
A fundamental limitation of standard machine learning methods is the
cost incurred by the preparation of the large training samples
required for good generalization. A potential remedy is offered by
multi-task learning: in many cases, while individual sample sizes are
rather small, there are samples to represent a large number of
learning tasks, which share some constraining or generative property.
If this property is sufficiently simple it should allow for better
learning of the individual tasks despite their small individual sample
sizes. The course review a wide class of multi-task learning methods,
describe techniques to solve the underlying optimization problems and
present an analysis of the generalization performance of these
learning methods which provides a proof of the superiority of
multi-task learning under specific conditions. We also highlight the
link between MTL and learning-to-learn (or meta-learning) and reviews
recent efforts in this direction.
Software security across abstraction layers (SS)
Abstract:
The execution infrastructure for today's software applications is
typically structured in a layered architecture where each layer offers
abstractions that are supposed to hide some of the implementation
details of lower layers.
Many security problems that occur in software applications however
are very much cross-layer: attacks against applications that are
running on top of the execution infrastructure often rely at least
to some extent on implementation details of one or more lower
layers.
The objective of this course is to study the problem of software
security across these various abstraction layers. The course starts
with an overview of the field of software security, and the definition
of the core concepts of the field. Next, it covers the security of
software systems against attackers that have access to lower
abstraction layers.
The course will cover both attack techniques as well as defense
techniques. On the attack side, we give an overview of influential
attack techniques ranging from classic buffer overflows to the most
recent micro-architectural attacks like Spectre, Meltdown and
Foreshadow. On the defense side, the emphasis will be on techniques
for which strong formal security guarantees can be given, as well
as on the limitations of such formal guarantees in the field of
security.
Internet of things: a data oriented approach
Abstract:
We will introduce the enabling technologies, protocols,
software architectures and applications for the development of the
Internet of Things (IoT) paradigm. After a short introduction to
pervasive computing and the emerging fields of applications (e.g.
Industry 4.0, domotics, intelligent transportation systems, wearable
devices, etc.) the course will provide an illustration of the enabling
components, technologies and standards of IoT systems. These will
include the wireless communications among end-devices and towards
infrastructures, data processing, languages for the development of
applications and prototypes (e.g. Arduino, STM32 Nucleo, etc.).
Specifically, the components of a typical IoT system will be illustrated
by following a data-oriented approach: from sensor to gateway (data
generation on behalf of sensors to data transmission in WSAN, WPAN,
WLAN), from gateway to cloud (up to cloud streaming and storages), and
from cloud to applications (including the data processing and
integration in complex software systems). A short illustration on open
issues and bottlenecks will conclude the seminar.