Deep Learning for Computer Architects

This is a primer written for computer architects in the new and rapidly evolving field of deep learning. It reviews how machine learning has evolved since its inception in the 1960s and tracks the key developments leading up to the emergence of the powerful deep learning techniques that emerged in the last decade.Machine learning, and specifically deep learning, has been hugely disruptive in many fields of computer science. The success of deep learning techniques in solving notoriously difficult classification and regression problems has resulted in their rapid adoption in solving real-world problems. The emergence of deep learning is widely attributed to a virtuous cycle whereby fundamental advancements in training deeper models were enabled by the availability of massive datasets and high-performance computer hardware.It also reviews representative workloads, including the most commonly used datasets and seminal networks across a variety of domains. In addition to discussing the workloads themselves, it also details the most popular deep learning tools and show how aspiring practitioners can use the tools with the workloads to characterize and optimize DNNs.The remainder of the book is dedicated to the design and optimization of hardware and architectures for machine learning. As high-performance hardware was so instrumental in the success of machine learning becoming a practical solution, this chapter recounts a variety of optimizations proposed recently to further improve future designs. Finally, it presents a review of recent research published in the area as well as a taxonomy to help readers understand how various contributions fall in context.Review: The book is mainly for hardware architects who are interested in exploiting properties of neural networks to optimize performance, power and area for an inference accelerator. This book assumes the readers have certain level of background in deep neural networks....

This is a primer written for computer architects in the new and rapidly evolving field of deep learning. It reviews how machine learning has evolved since its inception in the 1960s and tracks the key developments leading up to the emergence of the powerful deep learning techniques that emerged in the last decade.

Machine learning, and specifically deep learning, has been hugely disruptive in many fields of computer science. The success of deep learning techniques in solving notoriously difficult classification and regression problems has resulted in their rapid adoption in solving real-world problems. The emergence of deep learning is widely attributed to a virtuous cycle whereby fundamental advancements in training deeper models were enabled by the availability of massive datasets and high-performance computer hardware.

It also reviews representative workloads, including the most commonly used datasets and seminal networks across a variety of domains. In addition to discussing the workloads themselves, it also details the most popular deep learning tools and show how aspiring practitioners can use the tools with the workloads to characterize and optimize DNNs.

The remainder of the book is dedicated to the design and optimization of hardware and architectures for machine learning. As high-performance hardware was so instrumental in the success of machine learning becoming a practical solution, this chapter recounts a variety of optimizations proposed recently to further improve future designs. Finally, it presents a review of recent research published in the area as well as a taxonomy to help readers understand how various contributions fall in context.

Machine learning, and specifically deep learning, has been hugely disruptive in many fields of computer science. The success of deep learning techniques in solving notoriously difficult classification and regression problems has resulted in their rapid adoption in solving real-world problems. The emergence of deep learning is widely attributed to a virtuous cycle whereby fundamental advancements in training deeper models were enabled by the availability of massive datasets and high-performance computer hardware. This text serves as a primer for computer architects in a new and rapidly evolving field. We review how machine learning has evolved since its inception in the 1960s and track the key developments leading up to the emergence of the powerful deep learning techniques that emerged in the last decade. Next we review representative workloads, including the most commonly used datasets and seminal networks across a variety of domains. In addition to discussing the workloadsthemselves, we also detail the most popular deep learning tools and show how aspiring practitioners can use the tools with the workloads to characterize and optimize DNNs. The remainder of the book is dedicated to the design and optimization of hardware and architectures for machine learning. As high-performance hardware was so instrumental in the success of machine learning becoming a practical solution, this chapter recounts a variety of optimizations proposed recently to further improve future designs. Finally, we present a review of recent research published in the area as well as a taxonomy to help readers understand how various contributions fall in context. Machine learning, and specifically deep learning, has been hugely disruptive in many fields of computer science. The success of deep learning techniques in solving notoriously difficult classification and regression problems has resulted in their rapid adoption in solving real-world problems. The emergence of deep learning is widely attributed to a virtuous cycle whereby fundamental advancements in training deeper models were enabled by the availability of massive datasets and high-performance computer hardware. This text serves as a primer for computer architects in a new and rapidly evolving field. We review how machine learning has evolved since its inception in the 1960s and track the key developments leading up to the emergence of the powerful deep learning techniques that emerged in the last decade. Next we review representative workloads, including the most commonly used datasets and seminal networks across a variety of domains. In addition to discussing the workloads themselves, we also detail the most popular deep learning tools and show how aspiring practitioners can use the tools with the workloads to characterize and optimize DNNs. The remainder of the book is dedicated to the design and optimization of hardware and architectures for machine learning. As high-performance hardware was so instrumental in the success of machine learning becoming a practical solution, this chapter recounts a variety of optimizations proposed recently to further improve future designs. Finally, we present a review of recent research published in the area as well as a taxonomy to help readers understand how various contributions fall in context 4. Neural network accelerator optimization: a case study -- 4.1 Neural networks and the simplicity wall -- 4.1.1 Beyond the wall: bounding unsafe optimizations -- 4.2 Minerva: a three-pronged approach -- 4.3 Establishing a baseline: safe optimizations -- 4.3.1 Training space exploration -- 4.3.2 Accelerator design space -- 4.4 Low-power neural network accelerators: unsafe optimizations -- 4.4.1 Data type quantization -- 4.4.2 Selective operation pruning -- 4.4.3 SRAM fault mitigation -- 4.5 Discussion -- 4.6 Looking forward 3. Methods and models -- 3.1 An overview of advanced neural network methods -- 3.1.1 Model architectures -- 3.1.2 Specialized layers -- 3.2 Reference workloads for modern deep learning -- 3.2.1 Criteria for a deep learning workload suite -- 3.2.2 The fathom workloads -- 3.3 Computational intuition behind deep learning -- 3.3.1 Measurement and analysis in a deep learning framework -- 3.3.2 Operation type profiling -- 3.3.3 Performance similarity -- 3.3.4 Training and inference -- 3.3.5 Parallelism and operation balance A primer for computer architects in a new and rapidly evolving field. The authors review how machine learning has evolved since its inception in the 1960s and track the key developments leading up to the emergence of the powerful deep learning techniques that have emerged in the last decade. 5. A literature survey and review -- 5.1 Introduction -- 5.2 Taxonomy -- 5.3 Algorithms -- 5.3.1 Data types -- 5.3.2 Model sparsity -- 5.4 Architecture -- 5.4.1 Model sparsity -- 5.4.2 Model support -- 5.4.3 Data movement -- 5.5 Circuits -- 5.5.1 Data movement -- 5.5.2 Fault tolerance 2. Foundations of deep learning -- 2.1 Neural networks -- 2.1.1 Biological neural networks -- 2.1.2 Artificial neural networks -- 2.1.3 Deep neural networks -- 2.2 Learning -- 2.2.1 Types of learning -- 2.2.2 How deep neural networks learn