1. Modulverzeichnis (MTS)
  2. Module
  3. Modulbeschreibung anzeigen

Architecture of Machine Learning Systems

Zur Modulseite PDF generieren

#41078 / #1

SoSe 2023 - WiSe 2024/25

English

Architecture of Machine Learning Systems

6

Böhm, Matthias

Benotet

Mündliche Prüfung

English

Zugehörigkeit


Fakultät IV

Institut für Softwaretechnik und Theoretische Informatik

34352900 FG Big Data Engineering

Keine Angabe

Kontakt


Keine Angabe

Böhm, Matthias

matthias.boehm@tu-berlin.de

Keine Angabe

Lernergebnisse

Machine learning (ML) applications profoundly transform our lives, and many domains such as health care, finance, media, transportation, production, and information technology itself. In a narrow sense, ML systems are software systems underpinning theses ML applications. However, in a broad sense, ML systems comprise the entire systems from ML applications, over the compiler/runtime stack, to the underlying heterogeneous hardware devices. The overall objective of this module is to provide a holistic overview of the architecture, internals, and important concepts of modern (large-scale) ML systems. To this end, students acquire important competencies related to ML systems: O1: Understanding of the characteristics of ML systems and alternatives, which enables a better evaluation and usage of such systems O2: Understanding of effective techniques related to ML systems, which enables building and extending ML and data systems

Lehrinhalte

This module covers the architecture and essential concepts of modern machine learning (ML) systems for both local and large-scale machine learning. These architectures include systems for data-parallel execution, parameter servers, ML lifecycle systems, and the integration of ML into database systems. The covered topics focus both on a microscopic view of internal compilation, execution, and data management techniques, as well as a macroscopic view of end-to-end ML pipelines. In detail, the module covers the following topics which also reflect the lecture calendar (with a separate 90-120min lecture per topic): A: Overview and ML System Internals 01 Introduction and Overview 02 Languages, Architectures, and System Landscape 03 Size Inference, Rewrites, and Operator Selection 04 Operator Fusion and Runtime Adaptation 05 Data- and Task-Parallel Execution 06 Parameter Servers 07 Hybrid Execution and HW Accelerators 08 Caching, Partitioning, Indexing and Compression B: ML Lifecycle Systems 09 Data Acquisition, Cleaning, and Preparation 10 Model Selection and Management 11 Model Debugging, Fairness, and Explainability 12 Model Serving Systems and Techniques The module contains both lectures and exercises/programming projects which are offered as separate courses in order to increase flexibility.

Modulbestandteile

Compulsory area

Die folgenden Veranstaltungen sind für das Modul obligatorisch:

LehrveranstaltungenArtNummerTurnusSpracheSWS ISIS VVZ
Architecture of Machine Learning SystemsUESoSeen2
Architecture of Machine Learning SystemsVLSoSeen3
ISIS-Kurssuche nach Veranstaltungen des Moduls

Arbeitsaufwand und Leistungspunkte

Architecture of Machine Learning Systems (UE):

AufwandbeschreibungMultiplikatorStundenGesamt
Attendance Discussion Rounds3.02.0h6.0h
Exercise Prototype Implementation1.080.0h80.0h
86.0h(~3 LP)

Architecture of Machine Learning Systems (VL):

AufwandbeschreibungMultiplikatorStundenGesamt
Attendance Lectures15.03.0h45.0h
Pre/post processing Lectures15.01.0h15.0h
Exam Preparation1.030.0h30.0h
90.0h(~3 LP)
Der Aufwand des Moduls summiert sich zu 176.0 Stunden. Damit umfasst das Modul 6 Leistungspunkte.

Beschreibung der Lehr- und Lernformen

This module combines traditional lectures (on ML system internals and ML lifecycle systems), supporting examples of open source ML systems, as well as practical exercises/projects on related topics. For the practical part, teams of 1-3 students can pick one of two alternatives: a) Contribution of a unique feature to an open-source ML system (from a list of project proposals), or b) Alternative exercise on data-centric ML pipelines from data preparation, over model building, tuning, and parallelization, to model debugging

Voraussetzungen für die Teilnahme / Prüfung

Wünschenswerte Voraussetzungen für die Teilnahme an den Lehrveranstaltungen:

Completed bachelor degree Completed basic courses on applied machine learning, data management, and distributed systems

Verpflichtende Voraussetzungen für die Modulprüfungsanmeldung:

Voraussetzung
Leistungsnachweis »Architecture of Machine Learning Systems - Programmieraufgabe«

Abschluss des Moduls

Benotung

Benotet

Prüfungsform

Oral exam

Sprache(n)

English

Dauer/Umfang

45min

Dauer des Moduls

Für Belegung und Abschluss des Moduls ist folgende Semesteranzahl veranschlagt:
1 Semester.

Dieses Modul kann in folgenden Semestern begonnen werden:
Sommersemester.

Maximale teilnehmende Personen

Die maximale Teilnehmerzahl beträgt 60.

Anmeldeformalitäten

Registration in ISIS and exercise/project selection within the first 4 weeks of the semester.

Literaturhinweise, Skripte

Skript in Papierform

Verfügbarkeit:  nicht verfügbar

 

Skript in elektronischer Form

Verfügbarkeit:  verfügbar
Zusätzliche Informationen:

 

Literatur

Empfohlene Literatur
Matthias Boehm, Arun Kumar, Jun Yang: Data Management in Machine Learning Systems. Synthesis Lectures on Data Management, Morgan & Claypool Publishers 2019

Zugeordnete Studiengänge


Diese Modulversion wird in folgenden Studiengängen verwendet:

Studiengang / StuPOStuPOsVerwendungenErste VerwendungLetzte Verwendung
Dieses Modul findet in keinem Studiengang Verwendung.

Studierende anderer Studiengänge können dieses Modul ohne Kapazitätsprüfung belegen.

Sonstiges

Keine Angabe