Вештачка интелигенција

New Tool for Planning Deep Neural Network Accelerators

Summary

With the increasing demand for resource-intensive machine learning applications, device manufacturers often integrate specialized hardware components to quickly move and process large amounts of data required by these systems. Choosing the best design for these components, known as deep neural […]

New Tool for Planning Deep Neural Network Accelerators

With the increasing demand for resource-intensive machine learning applications, device manufacturers often integrate specialized hardware components to quickly move and process large amounts of data required by these systems.

Choosing the best design for these components, known as deep neural network accelerators, can be challenging due to a wide range of design possibilities. This problem becomes even more complex when designers want to add cryptographic operations to protect data from attackers.

Researchers at MIT have now developed a search tool that can effectively identify optimal designs for deep neural network accelerators, maintaining data security while increasing performance.

Their search tool, known as SecureLoop, is designed to consider how the addition of encryption and data authentication affects the performance and energy consumption of accelerators. Engineers can use this tool to obtain an optimal accelerator design tailored to their neural network and machine learning task.

Compared to conventional planning techniques that do not consider security, SecureLoop can improve the performance of accelerator designs while keeping data protected against attacks.

Using SecureLoop can help users enhance the speed and performance of demanding artificial intelligence applications, such as autonomous driving or medical image classification, while ensuring sensitive user data remains protected from certain types of attacks.

SecureLoop is the result of work by a team of scientists at MIT, led by Professor Joel Emer and researcher Kjungmi Li. Their research will be presented at the international IEEE/ACM Symposium on Microarchitecture.

With the help of the SecureLoop tool, designers and engineers can efficiently design deep neural network accelerators that provide high performance while maintaining reliability and data security.

Frequently Asked Questions

1. How does SecureLoop help improve the performance of machine learning accelerators?
SecureLoop identifies optimal accelerator designs that enable high performance while keeping data secure from certain types of attacks.

2. What are the benefits of using SecureLoop in demanding artificial intelligence applications?
SecureLoop can improve the speed and performance of applications such as autonomous driving or medical image classification, while maintaining data security and protection.

3. How does SecureLoop work in identifying the optimal accelerator design?
SecureLoop utilizes a search algorithm that considers performance, energy efficiency, and data security when identifying the optimal accelerator design for a neural network.

Original article: [source]