Neural Architecture Search Using Knowledge Database Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural architecture search methods are inefficient due to their failure to consider relevant starting conditions and ineffective neural network architecture parameters, leading to longer search times and reduced accuracy.
Innovation Solution
The proposed system initiates neural architecture searches using starting conditions with a higher probability of relevance, identified through a network knowledge database, and labels ineffective parameters to aid in machine learning analysis, thereby reducing search time and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional neural architecture search methods are used without considering starting conditions and parameter effectiveness, then the search process is simpler to implement, but the search time increases and accuracy decreases
Solution Approach 1:
The system performs preliminary actions by querying a network knowledge database before initiating the neural architecture search to identify relevant starting conditions and ineffective parameters. This pre-processing step filters and prepares optimized search parameters in advance, ensuring that the subsequent search process starts with high-quality inputs that reduce both time and improve accuracy.
Solution Approach 2:
The system implements feedback mechanisms by continuously querying the network knowledge database with historical search results and performance data. This feedback loop allows the system to learn from past searches, refine starting conditions, and adjust ineffective parameters, thereby improving search accuracy over time while reducing redundant exploration.
2Productivity
If traditional neural architecture search methods are used without considering starting conditions and parameter effectiveness, then the system complexity is lower, but the productivity decreases
Solution Approach 1:
The system introduces an intermediary component - the network knowledge database - that mediates between the search requirements and the actual neural architecture search process. This intermediary stores and provides pre-analyzed starting conditions and ineffective parameters, allowing the search system to benefit from sophisticated analysis without requiring the search algorithm itself to be overly complex.
Solution Approach 2:
The system implements self-service by automatically querying and utilizing the network knowledge database without requiring manual intervention. The system autonomously identifies relevant starting conditions and ineffective parameters, enabling high productivity through automated knowledge retrieval and application, while the complexity is managed through systematic database operations rather than complex search algorithms.
Data Source
AI summary
Methods, apparatus, systems, and articles of manufacture are disclosed to improve neural architecture searches. An example apparatus includes similarity verification circuitry to identify candidate networks based on a combination of a target platform type, a target workload type to be executed by the target platform type, and historical benchmark metrics corresponding to the candidate networks, the candidate networks associated with performance metrics. The example apparatus also includes likelihood verification circuitry to categorize (a) a first set of the candidate networks based on a first one of the performance metrics corresponding to first tier values, and (b) a second set of the candidate networks based on a second one of the performance metrics corresponding to second tier values, and extract first features corresponding to the first set of the candidate networks and extract second features corresponding to the second set of the candidate networks. The example apparatus also includes network analysis circuitry to improve network analysis efficiency by providing the first features and the second features to a network analyzer to identify particular ones of the candidate networks.