Home Social networking Exploring Graphical Neural Networks – Analytics India Magazine

Exploring Graphical Neural Networks – Analytics India Magazine

6
0

Data Scientists at CRED, Ravi Kumar and Samiran Roy explained the essence of the use of graphical neural networks and how the emerging technology is being used by CRED during the recent Deep Learning DevCon 2021. The duo explained how graphical neural networks should be modeled and the key factors separating graphical data from traditional tabular data used in neural network models.

CRED is an online payment application related to credit cards. The app was created by Kunal Shah, the founder of the FreeCharge company. The CRED application aims to automate the use of credit cards. The app also offers plenty of rewards for use in the form of CRED coins, which can later be redeemed for cash or various offers.

Register for our upcoming Masterclass>>

In the first few minutes, Ravi Kumar explained what graphical analysis and graphical neural networks are and discussed issues with traditional neural networks currently in use in the market.

“Graphics are a general language for describing and analyzing entities with relationships and interactions. In a graph network, nodes are entities that define a user, a merchant, or the like. Edges describe the relationship between two nodes and properties define information associated with nodes or edges, ”he said.

Image source: DevCon 2021

Looking for a job change? Let us help you.

“Real-world data is dynamic and keeps growing over time. A graph database brings deeper context to the data being processed and provides high value to the relationship between features. Tabular data becomes scarce as data grows, ”said Kumar, explaining why graphical databases are used dynamically.

There are different types of graph networks; some examples include Wiki networks, theft networks, underground networks and social networks. Additionally, he demonstrated how charts represent data and the difference between two of the most popular types of representation, RDF (Resource Description Frame) and LPG (Labeled Property Chart). In RDF, vertices and edges in the graph network have no internal structure, while in LPG, vertices and edges in the graph have internal structure and properties. RDF does not support the same pair of nodes and relationships more than once, but LPG does.

Image source: DevCon 2021

Ravi then developed the fundamental problems of traditional neural networks, such as interactivity between data points, the disadvantages of logical separation of nodes, the problems encountered when nodes are beyond 3, and more.

“In general machine learning, we can assume that identifiers are data points, but in the case of graphical neural networks, we can’t say the same,” Ravi added, continuing the discussion.

Samiran Roy then resumed the presentation and deepened the subject, explaining the operation and framework of a graphic neural network. A graph neural network consists of three main blocks: the Edge block, the node block, and the global block. These blocks work together with aggregation functions towards the required target goal.

Samiran said, “From a data science perspective, we have the ability to build functionality into any of the graphics network components. We can integrate features at the edge level, at the node level or at the graph level.

He also described how points in a graph network influence each other and what methods can be used to overcome problems in a graph network. During his presentation, he explained the interactions between network blocks and their aggregators in graph networks and their different variants.

See also
Facebook Introduces New Image Generation Model Called Instance-Packed GAN

Variants of graphical neural networks include the following types:

  • Complete GN block
  • Independent recurring block
  • Message transmission neural network
  • Non-local neural network
  • Relational network
  • Deep set

Speaking of the problems of graph neural networks in practice, Roy added, “The graphs we encounter in the real world are heterogeneous graphs; there are several types of knots. The aggregate functions used should be of utmost importance when defining our target output. When working with large graphs, we don’t need to calculate for all edges and nodes; instead, we can downsample nodes and use it to train our graphical neural networks. “

Graphical neural networks have several use cases, such as classifying social media users and predicting molecular property, to name a few.

Current use cases for graph neural networks at CRED include:

  • Product targeting model
  • Community detection
  • Graph completion
  • Sponsorship propensity ranking

The use cases explored by CRED incorporate models for downstream use cases and create user-user or user-merchant affinity models.

Graphic neural networks are nascent but are evolving rapidly as a field. Today, industry data requires a lot of research on graph networks, as current research is repeated on the same standard data sets. The flexibility of defining features and targets allows graphical neural networks to stand out from other types of typical neural networks.


Join our Discord server. Be part of an engaging online community. Join here.


Subscribe to our newsletter

Receive the latest updates and relevant offers by sharing your email.

Victor Dey

Victor Dey

Victor is an aspiring Data Scientist and holds a Master of Science in Data Science & Big Data Analytics. He is a researcher, data science influencer and former college football player. A great connoisseur of new developments in data science and artificial intelligence, he is committed to developing the data science community.


Source link

LEAVE A REPLY

Please enter your comment!
Please enter your name here