Islab at 08:57, 4 October 2021

2021-10-04T08:57:49Z

Eren: Created page with "{{StudentProjectTemplate |Summary=In this project, the candidate is supposed to implement a deep graph network that receives a set of graphs as input and returns the predicted..."

2020-09-21T21:26:10Z

Created page with "{{StudentProjectTemplate |Summary=In this project, the candidate is supposed to implement a deep graph network that receives a set of graphs as input and returns the predicted..."

New page

{{StudentProjectTemplate
|Summary=In this project, the candidate is supposed to implement a deep graph network that receives a set of graphs as input and returns the predicted next upcoming graph(s).
|Keywords=Deep Graph Networks
|Prerequisites=A solid background in Deep Learning. Experience in computer vision is a big plus.
|Supervisor=Eren Erdal Aksoy,
|Level=Master
|Status=Open
}}
In this project, the candidate is supposed to implement a deep graph network that receives a set of graphs as input and returns the predicted next upcoming graph(s). The main research question we will address is about the scalability of graph networks in action prediction and anticipation tasks. Therefore, the implemented network will be used for both action classification and anticipation tasks. In this regard, the proposed topic is more software related.
The candidate will first go through the literature and search for any related work. Next, a novel variational-autoencoder-based graph network will be implemented. The network bottleneck will be used for the classification task. The network will be then extended with an additional branch for the anticipation task. Since there exist various graph datasets, the candidate will employ them for network training without spending time with data annotation.

@@ Line 1: / Line 1: @@
 {{StudentProjectTemplate
 |Summary=In this project, the candidate is supposed to implement a deep graph network that receives a set of graphs as input and returns the predicted next upcoming graph(s).
 |Keywords=Deep Graph Networks
 |Prerequisites=A solid background in Deep Learning. Experience in computer vision is a big plus.
 |Supervisor=Eren Erdal Aksoy,
 |Level=Master
 |Status=Open
 }}
-In this project, the candidate is supposed to implement a deep graph network that receives a set of graphs as input and returns the predicted next upcoming graph(s). The main research question we will address is about the scalability of graph networks in action prediction and anticipation tasks. Therefore, the implemented network will be used for both action classification and anticipation tasks. In this regard, the proposed topic is more software related.
+In this project, the candidate is supposed to implement a deep graph network that receives a set of graphs as input and returns the predicted next upcoming graph(s). The main research question we will address is about the scalability of graph networks in action prediction and anticipation tasks. Therefore, the implemented network will be used for both action classification and anticipation tasks. There exist one MSc study with publicly available source code.
-The candidate will first go through the literature and search for any related work. Next, a novel variational-autoencoder-based graph network will be implemented. The network bottleneck will be used for the classification task. The network will be then extended with an additional branch for the anticipation task. Since there exist various graph datasets, the candidate will employ them for network training without spending time with data annotation.
+The candidate will build his work on this study. For instance, the network in this study will be extended with an additional branch for the anticipation task. Since there exist various graph datasets, the candidate will employ them for network training without spending time with data annotation.

Deep Graph Networks for Future Graph Prediction - Revision history

Islab at 08:57, 4 October 2021

Eren: Created page with "{{StudentProjectTemplate |Summary=In this project, the candidate is supposed to implement a deep graph network that receives a set of graphs as input and returns the predicted..."