Interactives Maschinelles Lernen für die Medizinische Informatik: Wann brauchen wir den Human-in-the-Loop?

Publikation: Beitrag in einer FachzeitschriftArtikelForschungBegutachtung

Abstract

Maschinelles Lernen (ML) ist der am schnellsten wachsende Bereich in der Informatik und die medizinische Informatik gehört zu den größten Herausforderungen der Zukunft. ML soll Algorithmen entwickeln, die sich im Laufe der Zeit verbessern und für Prognosen genutzt werden können. Die meisten ML Forscher konzentrieren sich auf automatisches Maschinelles Lernen (aML), wo große Fortschritte geleistet wurden: beispielsweise in der Spracherkennung, Empfehlungsdienstsysteme oder bei autonomen Fahrzeugen (Stichwort: „Google car“). Solche vollautomatischen Ansätze profitieren sehr stark von großen Datenmengen (Stichwort: „Big Data“) mit vielen Trainingsdaten. Im Gesundheitswesen werden wir jedoch oft mit einer kleinen Anzahl von Daten konfrontiert oder mit seltenen Ereignissen, wo aML-Ansätze nur unzureichende Ergebnisse liefern. Hier kann interaktives Maschinelles Lernen (iML) Hilfestellung bieten, d.h. nicht auf menschliches Domänenwissen zu verzichten, sondern vielmehr menschliche Intelligenz und ML zu kombinieren. Solche Ansätze haben ihre Wurzeln im Verstärkungslernen, Präferenzlernen und im so genannten Aktiven Lernen. Interaktives Maschinelles Lernen (iML) ist ein relativ neuer Ansatz und noch kein sehr geläufiger Begriff. Es handelt sich dabei um Algorithmen, die mit – teils menschlichen – Agenten interagieren und durch diese Interaktion ihr Lernverhalten optimieren können. Ein solcher „Human-in-the-Loop“ Ansatz kann vorteilhaft sein bei der Lösung schwerer Probleme, wie beispielsweise im Subspace Clustering, bei der Faltung von Proteinen oder bei der k-Anonymisierung von Gesundheitsdaten, wo menschliches Wissen helfen kann, einen exponentiellen Suchraum durch heuristische Auswahl der Daten drastisch zu reduzieren. Solche menschliche Agenten können daher in der Lernphase helfen die Komplexität zu reduzieren und daher Probleme lösen die andernfalls NP-schwer wären.
Titel in ÜbersetzungInteractives Maschinelles Lernen für die Medizinische Informatik: Wann brauchen wir den Human-in-the-Loop?
Originalspracheenglisch
Seiten (von - bis)119-131
Seitenumfang12
FachzeitschriftBrain Informatics
Jahrgang3
Ausgabenummer2
Frühes Online-Datum31 Mär 2016
DOIs
PublikationsstatusVeröffentlicht - 17 Mai 2016

Fingerprint

Informatics
Learning systems
Health
Learning
Protein folding
Problem-Based Learning
Protein Folding
Recommender systems
Reinforcement learning
Simulation Training
Machine Learning
Speech recognition
Computer science
Cluster Analysis
Computational complexity
Research Personnel

Schlagwörter

  • Maschinelles Lernen
  • Medizinische Informatik

ASJC Scopus subject areas

  • Artificial intelligence

Fields of Expertise

  • Information, Communication & Computing

Treatment code (Nähere Zuordnung)

  • Basic - Fundamental (Grundlagenforschung)

Dies zitieren

Interactive Machine Learning for Health Informatics: When do we need the human-in-the-loop? / Holzinger, Andreas.

in: Brain Informatics, Jahrgang 3, Nr. 2, 17.05.2016, S. 119-131.

Publikation: Beitrag in einer FachzeitschriftArtikelForschungBegutachtung

@article{79aa1ea8d80f4521a5af39b21e2ac0e8,
title = "Interactive Machine Learning for Health Informatics: When do we need the human-in-the-loop?",
abstract = "Machine learning (ML) is the fastest growing field in computer science, and health informatics is among the greatest challenges. The goal of ML is to develop algorithms which can learn and improve over time and can be used for predictions. Most ML researchers concentrate on automatic machine learning (aML), where great advances have been made, for example, in speech recognition, recommender systems, or autonomous vehicles. Automatic approaches greatly benefit from big data with many training sets. However, in the health domain, sometimes we are confronted with a small number of data sets or rare events, where aML-approaches suffer of insufficient training samples. Here interactive machine learning (iML) may be of help, having its roots in reinforcement learning, preference learning, and active learning. The term iML is not yet well used, so we define it as “algorithms that can interact with agents and can optimize their learning behavior through these interactions, where the agents can also be human.” This “human-in-the-loop” can be beneficial in solving computationally hard problems, e.g., subspace clustering, protein folding, or k-anonymization of health data, where human expertise can help to reduce an exponential search space through heuristic selection of samples. Therefore, what would otherwise be an NP-hard problem, reduces greatly in complexity through the input and the assistance of a human agent involved in the learning phase.",
keywords = "Machine Learning, Health Informatics, Maschinelles Lernen, Medizinische Informatik",
author = "Andreas Holzinger",
note = "http://www.springer.com/computer/ai/journal/40708",
year = "2016",
month = "5",
day = "17",
doi = "10.1007/s40708-016-0042-6",
language = "English",
volume = "3",
pages = "119--131",
journal = "Brain Informatics",
issn = "2198-4018",
publisher = "SpringerOpen",
number = "2",

}

TY - JOUR

T1 - Interactive Machine Learning for Health Informatics: When do we need the human-in-the-loop?

AU - Holzinger, Andreas

N1 - http://www.springer.com/computer/ai/journal/40708

PY - 2016/5/17

Y1 - 2016/5/17

N2 - Machine learning (ML) is the fastest growing field in computer science, and health informatics is among the greatest challenges. The goal of ML is to develop algorithms which can learn and improve over time and can be used for predictions. Most ML researchers concentrate on automatic machine learning (aML), where great advances have been made, for example, in speech recognition, recommender systems, or autonomous vehicles. Automatic approaches greatly benefit from big data with many training sets. However, in the health domain, sometimes we are confronted with a small number of data sets or rare events, where aML-approaches suffer of insufficient training samples. Here interactive machine learning (iML) may be of help, having its roots in reinforcement learning, preference learning, and active learning. The term iML is not yet well used, so we define it as “algorithms that can interact with agents and can optimize their learning behavior through these interactions, where the agents can also be human.” This “human-in-the-loop” can be beneficial in solving computationally hard problems, e.g., subspace clustering, protein folding, or k-anonymization of health data, where human expertise can help to reduce an exponential search space through heuristic selection of samples. Therefore, what would otherwise be an NP-hard problem, reduces greatly in complexity through the input and the assistance of a human agent involved in the learning phase.

AB - Machine learning (ML) is the fastest growing field in computer science, and health informatics is among the greatest challenges. The goal of ML is to develop algorithms which can learn and improve over time and can be used for predictions. Most ML researchers concentrate on automatic machine learning (aML), where great advances have been made, for example, in speech recognition, recommender systems, or autonomous vehicles. Automatic approaches greatly benefit from big data with many training sets. However, in the health domain, sometimes we are confronted with a small number of data sets or rare events, where aML-approaches suffer of insufficient training samples. Here interactive machine learning (iML) may be of help, having its roots in reinforcement learning, preference learning, and active learning. The term iML is not yet well used, so we define it as “algorithms that can interact with agents and can optimize their learning behavior through these interactions, where the agents can also be human.” This “human-in-the-loop” can be beneficial in solving computationally hard problems, e.g., subspace clustering, protein folding, or k-anonymization of health data, where human expertise can help to reduce an exponential search space through heuristic selection of samples. Therefore, what would otherwise be an NP-hard problem, reduces greatly in complexity through the input and the assistance of a human agent involved in the learning phase.

KW - Machine Learning

KW - Health Informatics

KW - Maschinelles Lernen

KW - Medizinische Informatik

UR - http://rd.springer.com/journal/40708

U2 - 10.1007/s40708-016-0042-6

DO - 10.1007/s40708-016-0042-6

M3 - Article

VL - 3

SP - 119

EP - 131

JO - Brain Informatics

JF - Brain Informatics

SN - 2198-4018

IS - 2

ER -