Trainable Module Training With Output-Deviation Label Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing training methods for trainable modules in autonomous vehicles are hindered by the inconsistency and errors in human-labeled datasets, leading to inaccurate decision-making in classification tasks.
Innovation Solution
A method is developed to pre-train multiple variations of trainable modules using diverse subsets of datasets, assess uncertainty through deviations in output values, and adjust the training process to prioritize accurate datasets, minimizing the impact of incorrect labels by optimizing parameters and updating labels as needed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If human-labeled training datasets are used for training trainable modules, then the training process can be completed with available data, but the accuracy of classification tasks deteriorates due to inconsistencies and errors in labels
Solution Approach 1:
The patent introduces an uncertainty assessment mechanism as an intermediary between the training data and the trainable module. This mechanism evaluates the reliability of training examples and selectively weights or excludes uncertain data points during training, thereby mediating between using available human-labeled data and maintaining classification accuracy
Solution Approach 2:
The patent dynamically adjusts training parameters based on uncertainty assessments. Training examples with high uncertainty are either excluded from training or assigned lower weights, while examples with low uncertainty are prioritized. This parameter adjustment resolves the contradiction by adapting the training process to data quality rather than treating all data equally
2Measurement precision
If multiple variations of trainable modules are pre-trained to assess uncertainty, then the accuracy of classification improves through better data evaluation, but the device complexity increases due to multiple module variations
Solution Approach 1:
The patent segments the training process into distinct phases: pre-training multiple variations, assessing uncertainty, and selective re-training. This segmentation allows the system to use multiple module variations only when necessary for uncertainty assessment, rather than maintaining all variations simultaneously in production, thus managing complexity
Solution Approach 2:
The patent performs preliminary actions by pre-training multiple module variations before the main training process. These variations are used solely for assessing the uncertainty of training labels. Once uncertainty is assessed and clean data is identified, the system proceeds with training using only the necessary components, avoiding the need to maintain all variations throughout the entire process
3Quantity of substance
If training datasets with incorrect labels are used, then the quantity of training data increases, but the reliability of the trainable module deteriorates due to learning from erroneous information
Solution Approach 1:
The patent applies local quality by assessing and treating individual training examples differently based on their specific uncertainty characteristics. Rather than discarding all data or treating it uniformly, the system identifies and prioritizes high-quality local regions (certain training examples) while downweighting or excluding uncertain ones, thus maintaining reliability while utilizing available data volume
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
The invention relates to a method (100) for training a trainable module (1), having the following steps: a plurality of modification processes (1a-1c) of the trainable module (1), said modification processes differing from one another to such a degree that the modification processes do not completely lead over to one another during a progressive learning process, are each pre-trained (110) at least using a respective sub-quantity of learning data sets (2); • learning input variable values (11a) of at least one learning data set (2) are fed (120) to all of the modification processes (1a-1c) as input variables (11); • the degree of uncertainty (13b) of output variable values (13) is ascertained from the deviation of the output variable values (13), into which each of the learning input variable values (11a) are converted by the modification processes (1a-1c), from one another and is assigned (130) to the learning data set (2) as the degree of uncertainty (2a) thereof; and • the uncertainty (2a) is used to ascertain (140) an evaluation (2b) of the learning data set (2), said evaluation being a measurement of how much the assignment of the learning output variable values (13a) relate to the learning input variable values (11a) in the learning data set (2). The invention also relates to a method (200) in which the trainable module is further operated (220) and actuates a system (50, 60, 70) using an actuation signal (5).