Reinforcement Learning in Combinatorially-Structured Environments: An Empirical Study hos Københavns Universitet

Natural Language Processing

Forskerne arbejder med metoder til automatisk at behandle, forstå og generere tekst, typisk ved hjælp af statistiske modeller og machine learning. Anvendelser af sådanne metoder omfatter automatisk faktatjek, oversættelser af computere og svar på spørgsmål.

Forskningsfeltet opdeles på DIKU i tre områder:

Core Natural Language Processing Natural Language Understanding Multi-modal Language Processing

Læs mere her: https://di.ku.dk/english/research/nlp/

Engelsk version:

The research section investigates methods to automatically process, understand as well as generate text, typically using statistical models and machine learning. Applications of such methods include automatic fact-checking, machine translation, question answering, and visually grounded language learning.

The section is divided into three areas:

Core Natural Language Processing Natural Language Understanding Multi-modal Language Processing

Read more here: https://di.ku.dk/english/research/nlp/

 

In some Reinforcement Learning (RL) tasks, a learner interacts with an environment with some underlying combinatorial structure. Some recent literature (e.g., [1]) present some algorithmic ideas in order to efficiently and provable exploid such structure to speed-up the learning process.
The goal of this project is to conduct an extensive empirical study of such algorithmic ideas. In particular, we would like to understand, through an empirical study, whether such model-based ideas could be combined with model-free algorithms such as Q-learning.

 

  • Suitable for 1-2 students
  • Good background in mathematics is a plus.

Supervisor: 
Sadegh Talebi (sadegh.talebi@di.ku.dk)

Husk at nævne, at du fandt dette opslag på KU Projekt & Job