18047335. SYSTEMS AND METHODS FOR DATA CORRECTION simplified abstract (Adobe Inc.)

SYSTEMS AND METHODS FOR DATA CORRECTION

Organization Name

Adobe Inc.

Inventor(s)

Varun Manjunatha of Newton MA (US)

Sarthak Jain of Boston MA (US)

Rajiv Bhawanji Jain of Falls Church VA (US)

Ani Nenkova Nenkova of Philadelphia PA (US)

Christopher Alan Tensmeyer of Fulton MD (US)

Franck Dernoncourt of Seattle WA (US)

Quan Hung Tran of San Jose CA (US)

Ruchi Deshpande of Belmont CA (US)

SYSTEMS AND METHODS FOR DATA CORRECTION - A simplified explanation of the abstract

This abstract first appeared for US patent application 18047335 titled 'SYSTEMS AND METHODS FOR DATA CORRECTION

The patent application focuses on a system and method for data correction, specifically in identifying false labels generated by a neural network and correcting them based on the influence of training labels.

• The system identifies false labels among predicted labels for different parts of an input sample.
• It computes the influence of each training label on the false label by approximating a change in conditional loss for the neural network.
• Based on the computed influence, it identifies the part of a training sample and the corresponding source label.
• The training set is then modified using this information to obtain a corrected training set.

Potential Applications: - This technology can be applied in various fields where accurate data labeling is crucial, such as image recognition, natural language processing, and medical diagnosis.

Problems Solved: - Addresses the issue of false labels in neural network predictions, improving the overall accuracy and reliability of the system.

Benefits: - Enhances the performance of neural networks by correcting false labels and improving the quality of training data. - Increases the trustworthiness of AI systems by ensuring more accurate predictions and classifications.

Commercial Applications: - "Enhanced Data Correction System for Neural Networks: Improving Accuracy and Reliability in AI Applications" - This technology can be utilized in industries such as healthcare, finance, and autonomous vehicles to enhance decision-making processes and reduce errors.

Prior Art: - Previous methods of data correction in neural networks may not have focused on identifying false labels based on the influence of training labels.

Frequently Updated Research: - Stay updated on advancements in neural network training methods and data correction techniques to further enhance the accuracy and efficiency of AI systems.

Questions about Data Correction System for Neural Networks:

Question 1: How does this technology compare to existing methods of data correction in neural networks? Answer: This technology stands out by identifying false labels based on the influence of training labels, leading to more accurate corrections and improved performance.

Question 2: Can this system be integrated into existing AI models easily? Answer: Yes, this system can be integrated into various neural network architectures with minimal modifications, making it versatile and adaptable to different applications.

Original Abstract Submitted

One aspect of systems and methods for data correction includes identifying a false label from among predicted labels corresponding to different parts of an input sample, wherein the predicted labels are generated by a neural network trained based on a training set comprising training samples and training labels corresponding to parts of the training samples; computing an influence of each of the training labels on the false label by approximating a change in a conditional loss for the neural network corresponding to each of the training labels; identifying a part of a training sample of the training samples and a corresponding source label from among the training labels based on the computed influence; and modifying the training set based on the identified part of the training sample and the corresponding source label to obtain a corrected training set.