Neural Structured Learning - Part 2: Training with natural graphs

No video

Neural Structured Learning - Part 2: Training with natural graphs

Рет қаралды 19,988

Күн бұрын

Welcome to episode 2 of our short series on Neural Structured Learning! In this episode, Software Engineer Arjun Gopalan discusses what natural graphs are and how they can be used to train neural networks.
Graphs are powerful data structures and can be used to represent data that most of us interact with every day. In this video, you’ll learn how to leverage graph-based data to effectively train neural networks for any machine learning task.
Links:
Get started with Neural Structured Learning → goo.gle/2sOz6DE
Hands-on tutorial → goo.gle/37pAD27
Coding TensorFlow → goo.gle/Coding...
Subscribe to the TensorFlow channel → goo.gle/Tensor...

Пікірлер: 6

@hayleecs4223 4 жыл бұрын

I wanted to solve a simple classification problem, but the documentation is not as simple as the code shown in the video. "make_datasets" function does not even exist. Doc examples are so problem specific that I cannot understand what to apply to solve my problem. Poorly documented...

@bhuvaneshs.k638 4 жыл бұрын

Thqs for this video...

@hegdelabs 4 жыл бұрын

Thanks for the video. Very informative video . Just wanted to know , during the inference is it possible to obtain the inferred graph as an output ? For example, in the form of adjacency matrix ?

@meifishK 4 жыл бұрын

It seems the loss function they use is to add a summed "distance difference" from the target sample (i) to one of its neighbors (j) besides the regular loss function. So the additional term is a summation of 1-to-1 relationship (i and every j in the neighbor). Therefore, the individual difference information (from i to individual j) was not kept. So I guess the 1-to-many adjacency information can not be kept in this scenario. More details about the loss they innovate: medium.com/tensorflow/introducing-neural-structured-learning-in-tensorflow-5a802efd7afd