GPU-Accelerated Knowledge Graph Analytics Kernels for All-Pairs Paths
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing knowledge mining methods for large-scale biomedical knowledge graphs, such as those derived from PubMed, are inefficient and lack scalability, precision, and provide insufficient explanations for discovered relationships, particularly when using natural language models and all-pairs shortest path algorithms in high-performance computing environments.
Innovation Solution
The DSNAPSHOT algorithm, a GPU-accelerated, distributed-memory parallel version of the Floyd-Warshall algorithm, is implemented to calculate the shortest path between all pairs of entities in a biomedical knowledge graph, utilizing a message passing interface and semiring operations to enhance efficiency and scalability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all-pairs shortest path algorithms are used to discover relationships in biomedical knowledge graphs, then the ability to uncover meaningful connections between entities is improved, but the computational time and resources required increase significantly
Solution Approach 1:
The knowledge graph is divided into multiple partitions or blocks, allowing the all-pairs shortest path algorithm to be applied to smaller subgraphs independently. This segmentation reduces the computational complexity from O(n³) for the entire graph to multiple smaller O(k³) operations where k << n, significantly reducing execution time while maintaining relationship discovery accuracy through subsequent merging of results.
Solution Approach 2:
The system pre-computes and caches shortest path results for frequently queried entity pairs or commonly accessed subgraphs. When a query is received, the system first checks the cache for pre-computed results before executing the full algorithm, thereby reducing repeated computational overhead and accelerating query response time without sacrificing accuracy.
2Extent of automation
If natural language models are used for knowledge mining, then the ability to extract relationships from unstructured text is improved, but the interpretability and explanation of discovered connections deteriorate
Solution Approach 1:
The system introduces an intermediary layer that bridges the black-box natural language model and the interpretable knowledge graph structure. This intermediary extracts structured relationship triples from NLP outputs and maps them to the knowledge graph, preserving both the automated extraction capability of NLP models and the interpretability of explicit graph relationships through standardized entity-relation-entity representations.
Solution Approach 2:
The system transforms the continuous probability outputs of natural language models into discrete, interpretable relationship categories by applying thresholding and clustering techniques. This parameter transformation converts opaque confidence scores into structured relationship types with clear semantic meanings, maintaining automated extraction while restoring explanation capability through categorical relationship labels.
3Quantity of substance
If the knowledge graph is expanded to include more biomedical concepts and relationships, then the comprehensiveness of knowledge mining is improved, but the computational complexity and processing requirements increase
Solution Approach 1:
The system applies different processing qualities to different regions of the knowledge graph based on their importance and access frequency. Frequently accessed or medically critical subgraphs are processed with higher precision and maintained in optimized memory structures, while less critical regions use approximate or compressed representations. This local differentiation maintains comprehensiveness while reducing overall computational complexity by focusing resources where they provide maximum value.
Data Source
AI summary
Data mining large-scale corpora of scholarly publications, such as the full biomedical literature, which may consist of tens of millions of papers spanning decades of research. The present disclosure provides a Distributed Accelerated Semiring All-Pairs Shortest Path (DSNAPSHOT) algorithm for computing shortest paths of a knowledge graph using distributed-memory parallel computers accelerated by GPUs. DSNAPSHOT implementations can analyze connected input graphs with millions of vertices using a large number graphics processing units (e.g., the 24,576 GPUs of the Oak Ridge National Laboratory's Summit supercomputer system). DSNAPSHOT provides sustained performance of about 136*1015 floating-point operations per second (136 petaflop/s) at a parallel efficiency of about 90% under weak scaling and, in absolute speed, 70% of the performance given our computation (in the single-precision tropical semiring or “min-plus” algebra). DSNAPSHOT may enable mining of scholarly knowledge corpora when embedded and integrated into artificial intelligence-driven natural language processing workflows at scale.