18394307. GRAPH NEURAL NETWORK MODEL FOR NEURAL NETWORK SCHEDULING DECISIONS simplified abstract (Intel Corporation)

From WikiPatents
Jump to navigation Jump to search

GRAPH NEURAL NETWORK MODEL FOR NEURAL NETWORK SCHEDULING DECISIONS

Organization Name

Intel Corporation

Inventor(s)

Hamza Yous of Dublin (IE)

Ian Hunter of Carnaross (IE)

Alessandro Palla of Pisa (IT)

GRAPH NEURAL NETWORK MODEL FOR NEURAL NETWORK SCHEDULING DECISIONS - A simplified explanation of the abstract

This abstract first appeared for US patent application 18394307 titled 'GRAPH NEURAL NETWORK MODEL FOR NEURAL NETWORK SCHEDULING DECISIONS

Simplified Explanation

A graph neural network (GNN) model is used in a scheduling process for compiling a deep neural network (DNN). The DNN, and parameter options for scheduling the DNN, are represented as a graph, and the GNN predicts a set of parameters that is expected to have a low cost. Using the GNN-based model, a compiler can produce a schedule for compiling the DNN in a relatively short and predictable amount of time, even for DNNs with many layers and/or many parameter options. For example, the GNN-based model reduces the overhead of exploring every parameter combination and does not exclude combinations from consideration like prior heuristic-based approaches.

  • The innovation involves using a graph neural network (GNN) model to optimize the scheduling process for compiling deep neural networks (DNNs).
  • The GNN predicts a set of parameters that result in a low cost, enabling efficient compilation of DNNs with multiple layers and parameter options.
  • This approach reduces exploration overhead and considers all parameter combinations, unlike previous heuristic-based methods.

Potential Applications

The technology can be applied in:

  • Deep learning model compilation
  • Optimization of scheduling processes in neural network development

Problems Solved

  • Efficient compilation of deep neural networks
  • Reduction of exploration overhead in parameter optimization

Benefits

  • Faster and more predictable compilation of DNNs
  • Improved optimization of scheduling processes
  • Enhanced efficiency in neural network development

Potential Commercial Applications

Optimizing DNN compilation processes for:

  • Tech companies
  • Research institutions

Possible Prior Art

Prior art may include:

  • Heuristic-based approaches for scheduling DNN compilation
  • Traditional optimization methods for parameter selection in neural networks

Unanswered Questions

How does the GNN model compare to traditional heuristic-based approaches in terms of efficiency and accuracy in predicting optimal parameters for DNN compilation?

The article does not provide a direct comparison between the GNN model and traditional heuristic-based approaches in terms of efficiency and accuracy.

What are the potential limitations or challenges in implementing the GNN-based scheduling process for DNN compilation in real-world applications?

The article does not address potential limitations or challenges that may arise in implementing the GNN-based scheduling process for DNN compilation in real-world applications.


Original Abstract Submitted

A graph neural network (GNN) model is used in a scheduling process for compiling a deep neural network (DNN). The DNN, and parameter options for scheduling the DNN, are represented as a graph, and the GNN predicts a set of parameters that is expected to have a low cost. Using the GNN-based model, a compiler can produce a schedule for compiling the DNN in a relatively short and predictable amount of time, even for DNNs with many layers and/or many parameter options. For example, the GNN-based model reduces the overhead of exploring every parameter combination and does not exclude combinations from consideration like prior heuristic-based approaches.

Retrieved from "http://wikipatents.org/index.php?title=18394307._GRAPH_NEURAL_NETWORK_MODEL_FOR_NEURAL_NETWORK_SCHEDULING_DECISIONS_simplified_abstract_(Intel_Corporation)&oldid=58151"