Neural Network Architecture Search
Define connotation
Neural network architecture search is the task of automatically finding one or more architectures for a given data set, which will generate models with good results for the given data set, and its essence is the optimal parameter search problem in high-dimensional space.
technical background
The use of deep learning models is becoming more and more popular and indispensable in many industries. However, the implementation of efficient neural networks
usually requires knowledge of the architecture and a lot of time to use the knowledge to explore reasonable solutions in an iterative process
. The form and structure of the neural network will vary according to specific needs, so
specific structures need to be set for different tasks and requirements. But designing these networks by trial and error is a time-consuming and tedious task that requires not only
architectural skills but also domain expertise. Typically, experts use their past experience or technical knowledge
to create and design neural networks.
In most industries, attention will be paid to model efficiency (efficiency here refers not only to execution efficiency, but also to development efficiency).
In order for a neural network to generalize without overfitting the training dataset, it is important to find the optimal structure. But in an era when productivity is more important than quality, some industries neglect the efficiency of their models, and are satisfied with the first model
that achieves their goals (as long as it works) without further optimizing the performance and efficiency of the model. Finding a suitable network architecture not only requires a lot of time and professional architecture design skills, but also easily falls into a suboptimal solution. Many industries do not have “enough” models to fully leverage the potential of their data due to lack of time or architectural expertise . Neural Architecture Search (Neural Architecture Search, NAS) [19] is an optimization-based algorithm that aims to find the best structure of a neural network for a specific task on a specific data set and solve the problem of imperfect predefined architecture. NAS transforms the design of the neural network architecture from a complicated manual method to an automated method. The effect of NAS on tasks such as image classification, target detection, and semantic segmentation can reach or even outperform manually designed network architectures.
NAS implements a specific search strategy in a defined exponential search space, looking for
neural network architectures that achieve high predictive performance on unseen data. Search strategies for exploring the space of network structures include random search, Bayesian
optimization, evolutionary methods, reinforcement learning, and gradient-based search methods. 
NAS provides more flexibility for model selection in MLOps by automatically exploring a variety of potential network architectures. Flexibility
and a more efficient solution
The idea solution
NNI implements the algorithms of ENAS, DARTS, and P-DARTS, and provides the interface of one-shot algorithm. In addition, it also
supports classic search methods such as Network Morphism [20].
NAS methods explore many potential solutions with variable complexity and thus are computationally expensive. The larger their
search space, the more architectures to test, train, and evaluate. These methods require a lot of resources and time to find
a good enough model, so when creating a neural network architecture search task, SecXOps will be more inclined to
assign the Pod where the project is located to a Node node with more idle computing resources. In , the program will output the model results in real time. SecXOps also establishes port mapping for such tasks,
allowing users to access and view their own task completion progress, as shown in the figure below. 
At the same time, click on the corresponding task to allow users to view the searched nerve Network Architecture
References
NSFOCUS SecXOps Security Intelligent Analysis Technology White Paper
Links
Hunan Province Network Security and Informatization Regulations 2021