17850807. SPARSITY FOR NEURAL NETWORK MODELS BASED ON SPARSITY ATTRIBUTES simplified abstract (Microsoft Technology Licensing, LLC)
SPARSITY FOR NEURAL NETWORK MODELS BASED ON SPARSITY ATTRIBUTES
Organization Name
Microsoft Technology Licensing, LLC
Inventor(s)
Ningxin Zheng of Shanghai (CN)
SPARSITY FOR NEURAL NETWORK MODELS BASED ON SPARSITY ATTRIBUTES - A simplified explanation of the abstract
This abstract first appeared for US patent application 17850807 titled 'SPARSITY FOR NEURAL NETWORK MODELS BASED ON SPARSITY ATTRIBUTES
Simplified Explanation
The present disclosure relates to systems and methods for introducing sparsity in neural network models based on sparsity attributes.
- The system receives a definition of a neural network model, which includes a set of tensors and sparsity attribute values for elements of a tensor.
- The sparsity attribute values for the tensor are then propagated to a subset of tensors to create a modified definition of the neural network model.
- Based on this modified definition, the system generates the actual neural network model.
Potential Applications
This technology can have various applications in the field of artificial intelligence and machine learning, including:
- Efficient deployment of neural network models on resource-constrained devices, such as mobile phones or IoT devices.
- Accelerating the training and inference processes of neural networks by reducing the number of computations required.
- Enabling faster and more efficient processing of large-scale datasets in areas like computer vision, natural language processing, and speech recognition.
Problems Solved
The technology addresses several challenges in neural network models:
- Overcoming the computational and memory limitations of deploying complex neural networks on devices with limited resources.
- Reducing the time and energy consumption associated with training and running neural networks.
- Improving the scalability and efficiency of processing large amounts of data in neural network applications.
Benefits
The use of sparsity attributes in neural network models offers several benefits:
- Improved model efficiency by reducing the number of non-zero elements in tensors, leading to faster computations and reduced memory requirements.
- Enhanced model interpretability by identifying and focusing on the most important elements within tensors.
- Increased flexibility in deploying neural network models on various devices and platforms, including edge devices and cloud environments.
Original Abstract Submitted
Embodiments of the present disclosure include systems and methods for providing sparsity for neural network models based on sparsity attributes. A first neural network model definition is received. The first neural network model definition specifies a neural network model comprising a set of tensors and a set of sparsity attribute values for elements of a tensor in the set of tensors. The set of sparsity attribute values for the tensor are propagated to elements of a subset of the set of tensors to form a second neural network model definition. The neural network model is generated based on the second neural network model definition.
