Skip to content. | Skip to navigation

Personal tools
You are here: Home Theses Accelerating KGlove Graph Embedding


Prof. Dr. S. Decker
RWTH Aachen
Informatik 5
Ahornstr. 55
D-52056 Aachen
Tel +49/241/8021501
Fax +49/241/8022321

How to find us

Annual Reports





Accelerating KGlove Graph Embedding

Thesis type
  • Bachelor
Student Abdulrahman Altabba
Status Finished
Presentation on 20. Feb 2018 00:00
Presentation room ?
Add presentation to calendar vCal
File download

Lately several methods for embedding graphs nodes into a vector space have been proposed. These embeddings can then used to train other machine learning models. Learning these embeddings is typically done using CPUs. In this thesis the student would look into the use of other hardware, like GPUs and distributed computation options to speed up the learning process. The challenge is that algorithms working on graphs have typically a bad memory locality. Hence, existing algorithms might need profound modification in order to use them on GPUs or in a distributed fashion.

The exponential growth of data and the need of processing it requires efficient algorithms that are capable of gradually interpreting this immense amount of new information gathered in all kind of structures. In this thesis we list several approaches to accelerate a recent graph embedding method and deploy it in tuning the hyper-parameters for the whole graph embedding model taking a graph as an input and delivering a set of embeddings for the instances in this graph. We provide some theoretical background and thoughts and explain the details of accelerating the underlying GloVe training model and how to overcome some related issues. Eventually we build a random search model that is used to determine the suitable hyper-parameters for an efficient graph embedding model.


The student need knowledge of algorithms and data structures. Experience programming GPUs is beneficial.

Document Actions