Multi-task Deep Learning Model Training via Batch Data Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning models are designed and trained for single tasks, leading to inefficiencies in storage and difficulty in learning multiple tasks, as well as challenges in acquiring appropriate training data for multi-task models.
Innovation Solution
A method for training a multi-task integrated deep learning model that generates training data in batches using multi-data conversion kernels, allowing the model to perform various visual intelligence tasks such as dehazing, super-resolution, and denoising.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple separate deep learning models are designed for different visual intelligence tasks, then each model can be optimized for its specific task, but the storage space requirement increases significantly
Solution Approach 1:
The patent combines multiple separate deep learning models into a single multi-task integrated model that shares common layers and parameters. This merging approach reduces the total number of parameters and storage requirements while maintaining the capability to perform multiple visual intelligence tasks such as dehazing, super-resolution, and denoising simultaneously
Solution Approach 2:
The integrated deep learning model is designed with universal functionality to handle multiple different visual intelligence tasks through a unified architecture. The model uses task-specific modules that can be selectively activated based on the input task, allowing one model to replace multiple specialized models
2Quantity of substance
If a multi-task integrated deep learning model is designed to perform various tasks, then storage efficiency is improved, but the model has difficulty in learning and acquiring appropriate training data
Solution Approach 1:
The training process is segmented into task-specific training phases where the model is trained on individual visual intelligence tasks separately before being integrated. This segmentation allows for acquiring and processing training data task-by-task, making the overall training process more manageable despite the multi-task nature of the model
Solution Approach 2:
The patent employs preliminary data conversion kernels that pre-process visual data into task-specific formats before the main training process. These kernels generate appropriate training data for each task (such as adding haze for dehazing training or adding noise for denoising training) in advance, facilitating easier acquisition and preparation of diverse training data
3Measurement precision
If separate models are trained for each visual intelligence task, then training data can be task-specific and optimized, but the overall training time and computational resources increase
Solution Approach 1:
By merging multiple task-specific training processes into a unified multi-task training framework, the model can learn multiple tasks simultaneously in a single training run. This combines the benefits of task-specific optimization with the efficiency of consolidated training, reducing total training time compared to sequential training of separate models
Data Source
AI summary
There is provided a training method of a multi-task integrated deep learning model. A multi-task integrated deep learning model training method according to an embodiment may generate training data for a plurality of visual intelligence tasks from visual data in a batch, and may train a multi-task integrated deep learning model which performs a plurality of visual intelligence tasks by using the generated training data. Accordingly, training data for training an integrated deep learning model which performs various visual intelligence tasks is generated in a batch through multi-data conversion kernels, so that appropriate training data for performing multiple tasks may be easily obtained and effective training of a multi-task integrated deep learning model is possible.