Network Measurements and Traffic Monitoring

The importance of mobile Internet access in our society is steadily
growing, as more physical objects get connected (Internet of Things) and an
increasing share of our daily activities rely on smartphones and apps.
Ubiquitous and reliable connectivity is provided to the end-users by a “network
of networks”, i.e., a global system composed of the interconnection of multiple
IP network infrastructures, wired and wireless, owned and managed by different
players (network operators). The scale and complexity of the network
infrastructure is growing together with the requirements for reliability and
performances. Owing to the ever-evolving nature of such network
infrastructures, the risk of failures and performance degradation caused by
accidental faults and/or deliberate attacks is not decreasing. As with any
complex infrastructure, the network needs to be continuously monitored in order
to promptly identify (and resolve) problems, failures and any sort of anomalous
behavior. 

The course focuses on methods and techniques to measure the
performances of a packet-oriented network in order to reveal performance
degradation and faults in wide-area networks. Particular focus is given to
wireless mobile networks (3G/4G).

The goal of the course is to provide the student with an
understanding of the most important aspects that must be considered in all
phases of the monitoring process, from the design of the data collection
methodology to the representation, analysis and interpretation of the
measurement data. Some fundamental theoretical aspects of data analytics will
be discussed along with more practical “do and don’t” rules. During the course,
several case-study problems from real-world measurement activities and research
problems will be critically examined
and discussed in the classroom, in order to distill general guidelines and data
properties in a bottom-up fashion.

The course is useful also for PhD students that are working in
different research areas. In fact, the analysis of data from a dynamic,
heterogeneous and complex system like “the Internet” leads naturally to
encounter a number of concepts and system-level aspects that are very general
and apply to many other engineered systems.

In summary, the methods and skills acquired by the students during
the project will benefit the students along multiple directions:

1.     
Students will gain a more
in-depth understanding of the dynamics at play in real-world networks.

2.    Based on the methods and skills
acquired during the course, the students will be able to engage in research
activities related to measuring and analyzing datasets from computer networks.

3.    Several notions and ideas
presented during the course with reference to packet networks are actually much
more general, and apply mutatis mutandis
to other classes of large-scale systems in other domains.

Examination: verbal colloquium and project assignment.

Tema predmeta so sodobni algoritmi za učenje iz podatkovnih tokov. Učili se bomo o odprtih izzivih na področju (inkrementalni modeli za nadzorovano učenje, stiskanje podatkov, odkrivanje spremembe v porazdelitvi toka (concept drift), gručenje iz podatkov, specializirane statistike za vrednotenje uspešnosti). S pridobljenim znanjem bo študent sposoben uporabljati svoje znanje o strojnem učenju pri aplikacijah, ki so povezane z obilico vsakdanjih podatkov (finančne in bančne transakcije, vremenski podatki, senzorski podatki itd.).

Predmet bo organiziran kot kombinacija predavanj in laboratorijskih vaj (te bodo izvedene z uporabo statističnega paketa R). V okviru vaj bodo študenti znanje aplicirali na izbranem problemu, ki je lahko direktno povezan tudi s tematiko doktorske naloge. V preostanku semestra bo organizirano tudi medsebojno tekmovanje za izdelavo najbolj točnega napovednega modela na podanih podatkih.

Izbrana poglavja iz umetne inteligence 2: Advanced Topics in Network Science (ANTS)

Networks or graphs are ubiquitous in everyday life. Examples include online social networks, the Web, wiring of a neural system, references between WikiLeaks cables, Supervizor, terrorist affiliations, LPP bus map, plumbing systems and your brain. Many such real-world networks reveal characteristic patterns of connectedness that are far from regular or random. Networks have thus been a prominent tool for investigating real-world systems since the 18th century. However, while small networks can be drawn by hand and analyzed by a naked eye, real-world networks require specialized computer algorithms, techniques and models. This led to the emergence of a new scientific field about 15 years ago denoted network science.

The course will first introduce the language of networks and review the fundamental concepts and techniques for the analysis of large real-world networks. In the main part of the course, the students will learn about selected advanced topics in network science with special emphasis on the practical applicability of the presented approaches. The topics will include node metrics, groups and patterns, large-scale network structure, network sampling, comparison, modeling, mining, inference, visualization and dynamics. The last part of the course will be devoted to invited talks on network science from the perspective of mathematicians, physicists, social scientists and other. 

The objective of the course is not to give a comprehensive theoretical discussion or in-depth review on any of the topics, but to present a rich set of network science tools that students could use in their own PhD work. The latter will be the main part of the coursework.

Except for a clearly identified PhD topic, there are no specific prerequisites for the course. However, the students will benefit from a solid knowledge in graph theory, probability theory and linear algebra, good programming skills in a language of their choice, and familiarity with research work and scientific writing.

The course is offered in the summer semester starting on February 29th, 2016 and lasts for fourteen weeks. Lectures and practice will be held in either English or Slovene.

For more see eUcilnica.

Študentje se bodo pri
predmetu seznanili z različnimi nalogami napovedovanja strukturiranih izhodnih
spremenljivk in z različnimi pristopi za njihovo reševanje. Spoznali bodo nekaj
najsodobnejših orodij za reševanje te vrste nalog in jih uporabili na praktičnih
problemih. Naučili se bodo tudi uporabljati metode napovedne analize strukturiranih
podatkov v kontekstu lastnega raziskovalnega dela. Predictive Analytics for Structured Data

The course will introduce the students to
different tasks of structured output prediction and describe a variety of
approaches for solving such tasks. The students will get to know some
state-of-the-art tools for solving such tasks and examples of their use in
practice. Within the course, the students will learn to apply predictive analytics
methods for structured data in the context of their research.

 The course will cover the following topics:

  1. The
    different tasks of structured output prediction, such as multi-target
    classification/ regression and (hierarchical) multi-label classification.
  2. Predictive
    clustering methods (tree and rule-based) for structured output prediction.
  3. Ontologies
    for data mining and their use for describing structured output prediction.
  4. Ensemble
    methods for structured output prediction (tree and rule ensembles).
  5. Applications
    of structured output prediction to different practical problems, from
    areas such as environmental/ life sciences and image annotation/ retrieval.