-
Notifications
You must be signed in to change notification settings - Fork 33
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
1 parent
bff166d
commit 36bfb59
Showing
1 changed file
with
17 additions
and
1 deletion.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -1 +1,17 @@ | ||
# Shortest path using graph networks and attention | ||
# Finding shortest paths with Graph Networks | ||
|
||
> In this article we show how a Graph Network with attention read and write can perform shortest path calculations. This network performs this task with 99.91% accuracy after minimal training. | ||
Here at Octavian we believe that graphs are a powerful medium for representing diverse knowledge (for example BenevolentAI uses them to represent pharmaceutical research and knowledge). | ||
|
||
Neural networks are a way to create functions that no human could write. They do this by harnessing the power of large datasets. | ||
|
||
On problems for which we have capable neural models, we can use example inputs and outputs to train the network to learn a function that transforms those inputs into those outputs, and hopefully generalizes to other unseen inputs. | ||
|
||
We need to be able to build neural networks that can learn functions on graphs. Those neural networks need the right inductive biases so that they can reliably learn useful graph functions. With that foundation, we can build powerful neural graph systems. | ||
|
||
Here we present a "Graph network with attention read and write", a simple network that can effectively compute shortest path. It is an example of how to combine different neural network components to make a system that readily learns a classical graph algorithm. | ||
|
||
We present this network both as a novel system in of itself, but more importantly as the basis for further investigation into effective neural graph computation. | ||
|
||
See the [full descripton on Medium]() |