Learning Apparatus for Dynamic Model Training Parameter Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for training machine learning models are cumbersome and time-consuming due to the need to comprehensively train models for multiple training parameter values, requiring extensive processing time.
Innovation Solution
A learning apparatus that acquires and iteratively trains machine learning models based on reference sequence data, adjusting training parameter values based on inference performance differences, using a slimmable neural network architecture to optimize model performance across different parameter settings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If comprehensive training is performed for each of multiple training parameter values, then model performance can be optimized, but processing time and operational complexity increase significantly
Solution Approach 1:
The patent performs preliminary training of a reference model once before the actual training process. The reference sequence obtained from this preliminary training is then reused multiple times during iterative training with different parameter values, eliminating the need to re-train the reference model and significantly reducing processing time.
Solution Approach 2:
The patent creates a reference sequence from a reference model training and copies this sequence for use in multiple training iterations with different parameter values. Instead of重新 training the reference model each time, the same reference sequence is reused, reducing computational overhead while maintaining performance optimization capabilities.
2Manufacturing precision
If comprehensive training is performed for each of multiple training parameter values, then model performance can be optimized, but operational complexity increases
Solution Approach 1:
The reference model is trained in advance before the actual training process begins. This preliminary action separates the reference sequence generation from the iterative training process, simplifying the operational flow by eliminating repeated reference model training steps and reducing operational complexity.
Solution Approach 2:
The reference sequence is copied and reused across multiple training iterations with different parameter values. This copying approach simplifies the training process by eliminating the need to re-generate reference sequences, making the operation more efficient and less complex.
3Productivity
If iterative learning is performed with dynamic parameter adjustment, then processing time is reduced, but training complexity increases
Solution Approach 1:
The patent implements feedback by calculating the difference between the target sequence and reference sequence during iterative training, and using this difference to dynamically adjust training parameter values. This feedback mechanism enables automatic parameter optimization without requiring complex manual intervention, improving processing speed while managing training complexity through systematic control.
Data Source
AI summary
According to one embodiment, a learning apparatus includes a processing circuit. The processing circuit acquires first sequence data representing transition of inference performance according to a training progress of a first model trained in accordance with a first training parameter value concerning a specific training condition. The processing circuit performs iterative learning of a second model in accordance with a second training parameter value concerning the specific training condition and changes the second training parameter value based on the inference performance of the second model and the first sequence data in a training process of the second model.