Page
%P
-
Chapter
Conclusion
The use of deep neural networks (DNNs) has recently seen explosive growth. They are currently widely used for many artificial intelligence (AI) applications including computer vision, speech recognition, and r...
-
Chapter
Kernel Computation
The fundamental computation of both CONV and FC layers described in Chapter 2 are multiply-and-accumulate (MAC) operations. Because there are negligible dependencies between these operations and the accumulati...
-
Chapter
Introduction
Deep neural networks (DNNs) are currently the foundation for many modern artificial intelligence (AI) applications [5]. Since the breakthrough application of DNNs to speech recognition [6] and image recognition1 ...
-
Book
-
Chapter
Overview of Deep Neural Networks
Deep Neural Networks (DNNs) come in a wide variety of shapes and sizes depending on the application.1 The popular shapes and sizes are also evolving rapidly to improve accuracy and efficiency. In all cases, the i...
-
Chapter
Operation Map** on Specialized Hardware
In Chapter 5, we discussed various key design considerations and techniques for the implementation of specialized DNN hardware. Also introduced was the notion of the map** of the computation for a particular wo...
-
Chapter
Exploiting Sparsity
A salient characteristic of the data used in DNN computations is that it is (or can be made to be) sparse. By saying that the data is sparse, we are referring to the fact that there are many repeated values in...
-
Chapter
Advanced Technologies
As highlighted throughout the previous chapters, data movement dominates energy consumption. The energy is consumed both in the access to the memory as well as the transfer of the data. The associated physical...
-
Chapter
Key Metrics and Design Objectives
Over the past few years, there has been a significant amount of research on efficient processing of DNNs. Accordingly, it is important to discuss the key metrics that one should consider when comparing and eva...
-
Chapter
Designing DNN Accelerators
In Chapter 4, we discussed how DNN processing can undergo transforms to leverage optimized libraries or reduce the number of operations, specifically multiplications, in order to achieve higher performance (i....
-
Chapter
Reducing Precision
As highlighted in the previous chapters, data movement dominates energy consumption and can affect the throughput for memory-bound systems. One way to address this issue to reduce the number of bits (bit width...
-
Chapter
Designing Efficient DNN Models
The previous two chapters discussed the use of DNN model and hardware co-design approaches, such as reducing precision (Chapter 7) and exploiting sparsity (Chapter 8), to reduce storage requirements, data move...
-
Chapter and Conference Paper
NetAdapt: Platform-Aware Neural Network Adaptation for Mobile Applications
This work proposes an algorithm, called NetAdapt, that automatically adapts a pre-trained deep neural network to a mobile platform given a resource budget. While many existing algorithms simplify networks based o...
