White Paper

DeciNets: Automatic Discovery of Fast Neural Networks on Specific AI Chips

Resource Featured Image

How Deci’s AutoNAC Automatically Extended the Accuracy-Latency Efficient Frontier for NVIDIA Cloud and Edge GPUs with DeciNets

Fast and accurate deep neural networks (DNNs) are key for successfully solving and deploying commercial AI applications. However, the computational costs of operating DNNs can also be very high. Another related but distinct obstacle is the need to deploy strong DNNs on edge devices that have limited computing power.

Neural architecture search (NAS) is a potentially viable approach to creating such models but existing NAS algorithms fall short of this objective. Automated Neural Architecture Construction (AutoNAC), a new technology developed by Deci AI, can optimize any given neural architecture. It can also discover fast and accurate neural architectures for any task on any specific AI chip, while taking both compilation and quantization into consideration.

This white paper introduces the AutoNAC technology and describes how it can be used to create proprietary DeciNet architectures that extend the efficient frontiers of latency/accuracy tradeoffs on NVIDIA cloud and edge GPUs.

Moreover, you will also learn about:

  • The opportunities and challenges in deep neural networks
  • The importance of hardware dependence and the deep inference stack
  • An overview of neural architecture design, search, and AutoNAC

Ran El-Yaniv Ph.D. is the author of this white paper. He is Deci’s Chief Scientist, a professor at the Technion – Israel Institute of Technology, and a former research scientist at Google.

Complete the form to get immediate access to the technical white paper.

Download Technical White Paper

Share
Add Your Heading Text Here
				
					from transformers import AutoFeatureExtractor, AutoModelForImageClassification

extractor = AutoFeatureExtractor.from_pretrained("microsoft/resnet-50")

model = AutoModelForImageClassification.from_pretrained("microsoft/resnet-50")