17932941. CONSTRAINED MASKING FOR SPARSIFICATION IN MACHINE LEARNING simplified abstract (QUALCOMM Incorporated)
Contents
- 1 CONSTRAINED MASKING FOR SPARSIFICATION IN MACHINE LEARNING
- 1.1 Organization Name
- 1.2 Inventor(s)
- 1.3 CONSTRAINED MASKING FOR SPARSIFICATION IN MACHINE LEARNING - A simplified explanation of the abstract
- 1.4 Simplified Explanation
- 1.5 Potential Applications
- 1.6 Problems Solved
- 1.7 Benefits
- 1.8 Potential Commercial Applications
- 1.9 Possible Prior Art
- 1.10 Original Abstract Submitted
CONSTRAINED MASKING FOR SPARSIFICATION IN MACHINE LEARNING
Organization Name
Inventor(s)
Debasmit Das of San Diego CA (US)
Jamie Menjay Lin of San Diego CA (US)
CONSTRAINED MASKING FOR SPARSIFICATION IN MACHINE LEARNING - A simplified explanation of the abstract
This abstract first appeared for US patent application 17932941 titled 'CONSTRAINED MASKING FOR SPARSIFICATION IN MACHINE LEARNING
Simplified Explanation
Certain aspects of the present disclosure provide techniques and apparatus for feature masking. A feature tensor is accessed in a neural network, and a feature mask is generated by processing the feature tensor using a masking subnetwork, where the masking subnetwork was trained based at least in part on a polarization constraint and an activation constraint to generate feature masks. A masked feature tensor is generated based on the feature tensor and the feature mask, and an output inference is generated using the neural network based at least in part on the masked feature tensor.
- Feature masking techniques and apparatus are provided in the patent application.
- A feature mask is generated by processing a feature tensor using a masking subnetwork trained with polarization and activation constraints.
- A masked feature tensor is created based on the feature tensor and the feature mask for generating output inferences.
Potential Applications
The technology could be applied in:
- Image recognition systems
- Natural language processing
- Autonomous vehicles
Problems Solved
- Enhances privacy by masking sensitive features in data
- Improves model interpretability by focusing on relevant features
- Helps in reducing overfitting in neural networks
Benefits
- Increased data privacy and security
- Enhanced model interpretability
- Improved generalization of neural networks
Potential Commercial Applications
Optimized for:
- Healthcare data analysis
- Financial fraud detection
- Customer sentiment analysis
Possible Prior Art
There may be prior art related to:
- Feature selection techniques in machine learning
- Privacy-preserving data processing methods
Unanswered Questions
How does this technology compare to existing feature masking methods in terms of computational efficiency?
The patent application does not provide a direct comparison with existing feature masking methods in terms of computational efficiency. Further research or experimentation may be needed to determine the efficiency of this technology compared to others.
What are the potential limitations or drawbacks of using this feature masking technique in real-world applications?
The patent application does not address potential limitations or drawbacks of using this feature masking technique in real-world applications. It would be important to investigate any possible constraints or challenges that may arise when implementing this technology in practical scenarios.
Original Abstract Submitted
Certain aspects of the present disclosure provide techniques and apparatus for feature masking. A feature tensor is accessed in a neural network, and a feature mask is generated by processing the feature tensor using a masking subnetwork, where the masking subnetwork was trained based at least in part on a polarization constraint and an activation constraint to generate feature masks. A masked feature tensor is generated based on the feature tensor and the feature mask, and an output inference is generated using the neural network based at least in part on the masked feature tensor.
