Decision Network Processing Indexing Parallel Batches
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for processing decision networks are inefficient, particularly when evaluating large decision trees, sparse or categorical input data, or trees with many false conditions, leading to slow processing and wasteful resource usage.
Innovation Solution
A method involving a processing system that uses an index to selectively map input parameter values to affected decision parameters, initializing values to default, and evaluating only necessary parameters, allowing for efficient evaluation of decision networks by decoupling decision parameter determination from predicate evaluation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If serialized evaluation of decision nodes is performed, then memory access latency is reduced, but processing speed becomes slow due to lack of parallelism
Solution Approach 1:
The patent segments the decision tree evaluation into independent batches of decision nodes that can be processed in parallel. By dividing the tree into multiple batches where nodes in each batch are independent of each other, the system achieves parallelism without requiring random memory access, thus resolving the contradiction between processing speed and memory access latency.
Solution Approach 2:
The patent performs preliminary actions by pre-organizing decision nodes into batches and pre-calculating their dependencies before evaluation. This allows the system to prepare the evaluation structure in advance, enabling efficient parallel processing without incurring memory access latency during the actual evaluation phase.
2Productivity
If parallelized evaluation of all decision nodes is performed, then processing speed increases, but resource efficiency decreases due to evaluation of unnecessary nodes
Solution Approach 1:
The patent extracts and evaluates only the necessary decision nodes for each input record by identifying the specific path from root to leaf. Instead of evaluating all nodes in parallel, the system extracts the minimal subset of nodes required for each evaluation, thereby maintaining high processing efficiency while avoiding computational waste on unnecessary nodes.
Solution Approach 2:
The patent applies partial action by evaluating only the portion of decision nodes that are necessary for each specific input record. Rather than performing excessive evaluation of all nodes, the system selectively evaluates only those nodes that lie on the actual decision path, optimizing both productivity and resource efficiency.
3Ease of manufacture
If conventional processing techniques are used, then implementation simplicity is maintained, but processing performance deteriorates for large decision trees and sparse data
Solution Approach 1:
The patent segments the decision tree into batches of independent nodes, which simplifies the implementation of parallel processing while significantly improving processing performance. This segmentation approach maintains ease of implementation by using standard parallel processing techniques while addressing the performance deterioration issue for large decision trees and sparse data.
Data Source
AI summary
Aspects of the subject disclosure may include, for example, a technique for processing a decision network that includes obtaining an index encoding a mapping from potential values of an input parameter to decision parameters of the network's predicates, wherein the mapping associates potential values of the input parameter with decision parameters affected by those potential values; evaluating decision parameters affected by specified values of the input parameter, including identifying each decision parameter to which the index maps at least one specified values of the input parameter, and setting the values of those decision parameters in accordance with the input parameter's specified values; and analyzing the decision network, including evaluating the predicates of one or more of the decision nodes based on the values of the predicates' decision parameters, and determining, based on the values of the evaluated predicates and a topology of the decision network, that a particular terminal node encodes the network's output. Other embodiments are disclosed.