Hasso-Plattner-Institut
Prof. Dr. Felix Naumann
  
 

Knowledge Graphs

Supervisors: Dr. Ralf Krestel

Organisation: We are planning to have pysical meetings if possible.
If necessary, we will switch to a virtual setup using BBB.

The introductory, first session is open to all and will take place on Tuesday, April 12th at 09:15 in F.E-06 (and/or, if needed, on BBB).
Afterwards, you can register for the seminar. If more than 9 students register, we will choose randomly.

This seminar's focus is on knowledge graphs (KG) in general and on three common knowlege graph problems in particular. We will learn about KG construction, KG completion using embeddings, and KG refinement based on current state-of-the-art deep learning approaches. In the first half of the semester, each students will read and present one paper solving one of the three tasks. In the second half, the students will work in teams of three to create/complete/refine a knowledge graph. We collaborate with the Wildenstein Plattner Institute in New York, which provides training data, namely a large collection of art-historic documents out of which a KG should be created.

Schedule

Time: 09:15
Location: Campus 3, F-E-06 and BBB https://bbb.hpi.uni-potsdam.de/b/ral-1ib-9my-ol2

DateTopic
April 13thOrganization & Introduction to Knowlege Graphs (KGs)
April 20thInitial Meeting & Introduction to KG Construction
May 4th3 Student Presentations & Introduction to KG Embedding/Completion
May 18th3 Student Presentations & Introduction to KG Refinement
June 1st3 Student Presentations & Introduction to Project Work
July 13thTeam Presentations & Closing of Seminar

(still updated, subject to change)

Grading

  • 50% Paper Presentation
  • 50% Team Project

(subject to change)

Prerequisites

  • Basic knowledge of machine learning is necessary (What is cross-validation? What is an objective function? What is classification? What are labels? etc.).
  • Having prior knowledge about deep learning is an advantage but not required (What is backpropagation? What is a CNN? etc.).
  • Having some experience with Python is recommended.