Object Classification Control Using Few-Shot Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network models face accuracy deterioration due to dataset drift when data distributions differ between training and input, making additional learning resource-intensive and challenging, especially in on-device environments with data collection limitations.
Innovation Solution
An electronic apparatus employs a first neural network model for initial object identification and a second few-shot learning model for verification, using user-identified sample data to enhance accuracy in varying data distributions without additional learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If additional learning is performed by the server to update the neural network model for new data distribution, then object identification accuracy is improved, but resource consumption and time are significantly increased
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network model on diverse data distributions during the initial training phase. The model is trained on multiple datasets representing different distributions before deployment, enabling it to adapt to new data distributions without requiring retraining when encountering drifted data in production environments.
Solution Approach 2:
The patent implements self-service through online adaptation mechanisms that enable the neural network model to automatically adjust to new data distributions in real-time without external intervention. The model uses incoming data to continuously update its parameters and adapt to distribution shifts, eliminating the need for periodic server-based retraining and redistribution.
2Measurement precision
If additional learning is performed by the server to update the neural network model for new data distribution, then object identification accuracy is improved, but resource consumption is significantly increased
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network model on diverse data distributions during the initial training phase. The model is trained on multiple datasets representing different distributions before deployment, enabling it to adapt to new data distributions without requiring retraining when encountering drifted data in production environments.
Solution Approach 2:
The patent implements self-service through online adaptation mechanisms that enable the neural network model to automatically adjust to new data distributions in real-time without external intervention. The model uses incoming data to continuously update its parameters and adapt to distribution shifts, eliminating the need for periodic server-based retraining and redistribution.
3Measurement precision
If data with new distribution are collected by the electronic apparatus for additional learning, then object identification accuracy is improved, but data collection is limited due to personal information protection
Solution Approach 1:
The patent implements self-service through online adaptation mechanisms that enable the neural network model to automatically adjust to new data distributions in real-time without external intervention. The model uses incoming data to continuously update its parameters and adapt to distribution shifts, eliminating the need for periodic server-based retraining and redistribution.
Solution Approach 2:
The patent uses a lightweight adaptation layer or buffer as an intermediary between the neural network model and the input data stream. This intermediary processes and adapts to new data distributions in real-time, allowing the model to learn from incoming data without requiring direct access to or storage of sensitive user data, thus resolving privacy concerns while maintaining adaptability.
4Device complexity
If a single neural network model is used for object identification, then device complexity is reduced, but accuracy deteriorates when data distribution differs from training data
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network model on diverse data distributions during the initial training phase. The model is trained on multiple datasets representing different distributions before deployment, enabling it to adapt to new data distributions without requiring retraining when encountering drifted data in production environments.
Solution Approach 2:
The patent implements self-service through online adaptation mechanisms that enable the neural network model to automatically adjust to new data distributions in real-time without external intervention. The model uses incoming data to continuously update its parameters and adapt to distribution shifts, eliminating the need for periodic server-based retraining and redistribution.
Data Source
AI summary
Provided is an electronic apparatus including memory configured to store at least one instruction, and a processor configured to execute the at least one instruction to obtain a first image including an object, input the first image to a first neural network model that is configured to be trained by using a plurality of second images in relation to a plurality of predefined types, obtain first probability information including a first probability of the object corresponding to a first type among the plurality of types and a second probability of the object corresponding to a second type among the plurality of types, obtain second probability information, through a second neural network model, indicating a type of the object included in the first image, by using a plurality of third images corresponding to the first type and a plurality of fourth images corresponding to the second type based on a difference between the first probability and the second probability being less than a first threshold value and based on a first input, and identify the type of the object based on the second probability information.