18114658. MULTI-OBJECTIVE NEURAL ARCHITECTURE SEARCH FRAMEWORK simplified abstract (Samsung Electronics Co., Ltd.)
Contents
- 1 MULTI-OBJECTIVE NEURAL ARCHITECTURE SEARCH FRAMEWORK
- 1.1 Organization Name
- 1.2 Inventor(s)
- 1.3 MULTI-OBJECTIVE NEURAL ARCHITECTURE SEARCH FRAMEWORK - A simplified explanation of the abstract
- 1.4 Simplified Explanation
- 1.5 Potential Applications
- 1.6 Problems Solved
- 1.7 Benefits
- 1.8 Potential Commercial Applications
- 1.9 Possible Prior Art
- 1.10 Original Abstract Submitted
MULTI-OBJECTIVE NEURAL ARCHITECTURE SEARCH FRAMEWORK
Organization Name
Inventor(s)
Mostafa El-khamy of San Diego CA (US)
Kee-Bong Song of San Diego CA (US)
MULTI-OBJECTIVE NEURAL ARCHITECTURE SEARCH FRAMEWORK - A simplified explanation of the abstract
This abstract first appeared for US patent application 18114658 titled 'MULTI-OBJECTIVE NEURAL ARCHITECTURE SEARCH FRAMEWORK
Simplified Explanation
The patent application describes a system and method for conducting a neural architecture search using continuous relaxation of a discrete network search space over operators in a super-network.
- Sampling a discrete network search space
- Determining a differential architecture network from a super-network using continuous relaxation
- Calculating a reward based on proxy accuracy or complexity
- Updating the distribution of the discrete network search space based on the reward
- Determining an updated differential architecture network based on the reward
Potential Applications
This technology could be applied in various fields such as computer vision, natural language processing, and speech recognition to optimize neural network architectures for specific tasks.
Problems Solved
This technology addresses the challenge of manually designing neural network architectures by automating the process through a neural architecture search, leading to improved performance and efficiency.
Benefits
The benefits of this technology include faster development of high-performing neural networks, reduced human effort in architecture design, and the potential for discovering novel network structures.
Potential Commercial Applications
One potential commercial application of this technology could be in the development of AI-powered products and services that require efficient and accurate neural networks, such as autonomous vehicles, medical diagnostics, and recommendation systems.
Possible Prior Art
One possible prior art in this field is the work on neural architecture search algorithms such as reinforcement learning-based methods and evolutionary algorithms that aim to automatically design neural network architectures.
Unanswered Questions
How does this technology compare to existing neural architecture search methods?
This article does not provide a direct comparison with other neural architecture search methods, leaving the reader to wonder about the specific advantages and limitations of this approach compared to existing techniques.
What are the computational requirements of implementing this neural architecture search method?
The article does not delve into the computational resources needed to perform the neural architecture search, leaving a gap in understanding the practical feasibility and scalability of the proposed method.
Original Abstract Submitted
A system and a method are disclosed for performing a neural architecture search. The method includes sampling a discrete network search space a first time, determining a differential architecture network sampled from a super-network using continuous relaxation of the discrete network search space over operators in the super-network, calculating a reward based on a proxy accuracy or a proxy complexity of the differential architecture network, updating a distribution of the discrete network search space based on the reward, and determining an updated differential architecture network based on the reward.
