Parallel SGD Model Aggregation for Cyber Attack Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face challenges in efficiently detecting and mitigating cyber attacks, such as denial-of-service (DoS) attacks, due to the sequential nature of stochastic gradient descent (SGD) algorithms, which leads to prolonged computation times and resource-intensive data transmission, affecting the timeliness and scalability of network security.
Innovation Solution
Implementing a parallel implementation of SGD across multiple geographic locations to compute local models and symbolic representations, allowing for the generation of a global model that determines illegitimate data traffic associated with cyber attacks, while conserving resources and reducing latency by transmitting local models and symbolic representations instead of large training data subsets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sequential SGD is used to compute model parameters for cyber attack detection, then measurement precision is improved, but loss of time increases
Solution Approach 1:
The patent divides the training dataset into multiple partitions and assigns each partition to a separate processing node. Each node independently computes model parameters using sequential SGD on its local partition, then the results are aggregated. This segmentation allows parallel execution of otherwise sequential operations, reducing total computation time while maintaining parameter accuracy through proper aggregation of results from all nodes.
2Measurement precision
If centralized processing of large training datasets is performed, then measurement precision is improved, but use of energy increases
Solution Approach 1:
The training dataset is segmented and distributed across multiple processing nodes, each handling a portion of the computation locally. This eliminates the need to move and process entire large datasets through a single centralized system, reducing energy consumption associated with data transmission and centralized processing while maintaining model accuracy through aggregation of results from all distributed nodes.
Solution Approach 2:
The patent transitions from a single-dimensional centralized processing architecture to a multi-dimensional distributed architecture across multiple processing nodes. This dimensional change allows the system to process large datasets by distributing the computational workload across spatial dimensions (multiple nodes) rather than concentrating it in one location, thereby reducing energy consumption while preserving measurement precision.
3Manufacturing precision
If sequential SGD is used for model training, then manufacturing precision is improved, but productivity decreases
Solution Approach 1:
The patent segments the training process into independent parallel tasks executed by multiple processing nodes. Each node performs sequential SGD on its local data partition with the same precision as traditional sequential SGD, ensuring manufacturing precision is maintained. Meanwhile, the parallel execution of multiple segments dramatically increases productivity by generating the final model faster than single-node sequential processing could achieve.
4Productivity
If distributed computing is implemented for parallel SGD, then productivity is improved, but device complexity increases
Solution Approach 1:
The patent implements distributed computing by segmenting the training dataset and assigning partitions to different processing nodes. Each node independently executes sequential SGD on its local data, which simplifies the architecture compared to implementing true parallel SGD. This segmented approach improves productivity through parallel execution while avoiding the complexity of coordinating simultaneous parameter updates across distributed nodes, as each node operates independently on its data partition.
Data Source
AI summary
Described herein is a system transmits and combines local models, that individually include a set of local parameters computed via stochastic gradient descent (SGD), into a global model that includes a set of global model parameters. The local models are computed in parallel at different geographic locations (e.g., different instances of computing infrastructure) along with symbolic representations. Network transmission of the local models and the symbolic representations, rather than transmission of the large training data subsets processed to compute the local models and symbolic representations, conserves resources and decreases latency. The global model can then be used as a model to determine a likelihood that at least a portion of current and/or recently received data traffic is illegitimate data traffic that is associated with a cyber attack. In some instances, the system can implement a remedial action to mitigate the effects of the cyber attack on computing infrastructure.