Application Segmentation Models With Log-Based Memory Protection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models consume substantial memory due to their complex architectures and large parameter sets, leading to challenges in deploying them in resource-limited environments, with current memory reduction techniques often compromising model accuracy, computational efficiency, and requiring complex processes or hardware incompatibilities.
Innovation Solution
Implementing memory surge protection processes that include obtaining log data, determining app-segments and user-groups via machine learning, and performing data purging and batch processing to reduce memory usage, utilizing selective algorithms and batch processing across multiple machines.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If traditional memory reduction techniques (pruning, quantization, weight sharing) are applied to machine learning models, then memory consumption is reduced, but model accuracy deteriorates
Solution Approach 1:
The patent segments the log data into manageable chunks and processes them in batches, dividing the large dataset into smaller sub-datasets that can be handled individually. This segmentation allows the system to reduce memory usage by processing only the necessary portions of data at a time while maintaining processing accuracy through systematic batch evaluation.
Solution Approach 2:
The patent performs preliminary actions by pre-processing and filtering log data before it enters the main processing pipeline. Entries are pre-filtered based on criteria such as transaction thresholds and user interaction patterns, removing unnecessary data beforehand to reduce memory consumption during subsequent processing stages while preserving essential information for accurate model training.
2Quantity of substance
If aggressive memory reduction techniques are applied to machine learning models, then memory consumption is reduced, but computational efficiency and latency performance deteriorate
Solution Approach 1:
The patent implements dynamic memory management where the system adjusts its processing behavior based on real-time memory availability and data characteristics. The batch processing size and filtering thresholds are dynamically adjusted to balance memory consumption with processing speed, allowing the system to maintain high computational efficiency while reducing memory usage through adaptive rather than static memory reduction strategies.
Solution Approach 2:
The patent changes processing parameters such as batch sizes, filtering thresholds, and data sampling rates to optimize the balance between memory consumption and computational efficiency. By adjusting these parameters dynamically based on system state and data characteristics, the system achieves memory reduction without significant penalties to processing speed or model training efficiency.
3Quantity of substance
If machine learning models are deployed in resource-limited environments, then memory consumption is reduced, but model performance and accuracy deteriorate
Solution Approach 1:
The patent extracts and removes unnecessary data entries from the log data through systematic filtering and purging processes. By taking out redundant, rarely used, or low-value entries based on predefined criteria, the system reduces the overall data volume and memory requirements while preserving the essential patterns and relationships needed for maintaining model performance in resource-constrained environments.
Solution Approach 2:
The patent creates simplified representations or copies of the original data and models that capture the essential characteristics while requiring less memory. Through techniques like data sampling, feature selection, and model quantization, the system generates reduced versions of the full model and data that maintain sufficient accuracy for effective operation in resource-limited settings.
4Quantity of substance
If data purging is performed to reduce memory usage, then memory consumption is reduced, but information loss may occur
Solution Approach 1:
The patent implements feedback mechanisms that continuously monitor data quality, model performance, and memory usage. The purging process is guided by feedback from model training results and performance metrics, allowing the system to adjust filtering criteria to preserve information that is critical for maintaining model accuracy while removing data that can be safely discarded based on observed performance patterns.
Solution Approach 2:
The patent applies partial purging strategies where not all data is removed, but rather a carefully selected portion is filtered based on importance thresholds. By using partial action with conservative filtering criteria, the system reduces memory consumption while maintaining sufficient information content, avoiding excessive purging that would result in significant information loss.
Data Source
AI summary
Systems and methods for memory surge protection for application segmentation models include obtaining log data for a plurality of users of an enterprise where the log data relates to usage of a plurality of applications by the plurality of users and user metadata; determining a memory usage estimation based on the log data; determining i) app-segments that are groupings of application of the plurality of applications and ii) user-groups that are groupings of users of the plurality of users, based on the log data and the memory usage estimation; and providing access policy of the plurality of applications based on the user-groups and the app-segments.