Results of the summarizing process for the arXiv paper: 2006.13829v1

The complexity of deep learning (DL) models has been increasing, leading to higher compute requirements. The deployment of Convolutional Neural Networks (CNNs) involves two phases: training and inference. While the inference task is typically performed on resource-constrained devices, there has been significant research in low-power inference on custom hardware accelerators. On the other hand, training CNNs is more computationally and memory-intensive and is primarily done on power-hungry GPUs in large-scale data centers. However, CNN training on Field-Programmable Gate Arrays (FPGAs) is a relatively new area of research. This is mainly because there is a lack of tools that allow for easy prototyping and deployment of hardware and algorithmic techniques for power-efficient CNN training on FPGAs. To address this limitation, the authors present Barista - an automated toolflow that seamlessly integrates FPGAs into the training of CNNs within the popular deep learning framework Caffe. Barista provides versatile and rapid deployment of hardware and algorithms for FPGA-based CNN training while also serving as an infrastructure for further research and development in this field. By enabling researchers to prototype and deploy various hardware and algorithmic techniques easily, Barista opens up new possibilities for exploring power-efficient CNN training on FPGAs. Furthermore, it is currently the only tool available that offers comprehensive support for FPGA-based CNN training within Caffe. Thus, this work contributes to advancing the field of low-power DL by providing a much needed toolflow for FPGA based CNN training which allows efficient utilization of resources during model development - paving the way for improved performance and energy efficiency in DL applications.

• - Complexity of deep learning models increasing, leading to higher compute requirements
• - Deployment of Convolutional Neural Networks (CNNs) involves training and inference phases
• - Inference typically performed on resource-constrained devices
• - Research in low-power inference on custom hardware accelerators
• - CNN training done on power-hungry GPUs in large-scale data centers
• - CNN training on FPGAs a relatively new area of research due to lack of tools for prototyping and deployment
• - Barista is an automated toolflow that integrates FPGAs into CNN training within Caffe framework
• - Barista provides versatile and rapid deployment of hardware and algorithms for FPGA-based CNN training
• - Barista serves as infrastructure for further research and development in this field
• - Enables researchers to prototype and deploy various hardware and algorithmic techniques easily
• - Currently the only tool available with comprehensive support for FPGA-based CNN training within Caffe framework
• - Contributes to advancing low-power deep learning by providing a much-needed toolflow for FPGA-based CNN training
• - Allows efficient utilization of resources during model development
• - Paves the way for improved performance and energy efficiency in DL applications.

Deep learning models are becoming more complex, which means they need more computing power to work. Convolutional Neural Networks (CNNs) have two phases: training and inference. Inference is when the CNN makes predictions on devices that don't have a lot of resources. Researchers are studying how to make inference use less power by using special hardware. Training CNNs usually happens in big data centers with powerful GPUs, but some researchers are exploring using FPGAs instead. Barista is a tool that helps researchers use FPGAs for training CNNs in the Caffe framework. It's the only tool available right now for this kind of training and it helps improve the efficiency and performance of deep learning applications." Definitions- Complexity: how complicated something is - Compute requirements: the amount of computing power needed - Deployment: putting something into action or use - Inference: making predictions or decisions based on information - Resource-constrained devices: devices that have limited resources like memory or processing power - Low-power inference: finding ways to use less energy when making predictions - Custom hardware accelerators: specialized hardware that speeds up certain tasks - Power-hungry GPUs: graphics processing units that need a lot of electricity to work well - FPGAs: field-programmable gate arrays, which are chips that can be programmed after they're made - Prototyping and deployment: creating and using something in real-world situations - Versatile and rapid deployment: being able to

Exploring Power-Efficient CNN Training on FPGAs with Barista

The complexity of deep learning (DL) models has been increasing, leading to higher compute requirements. While the inference task is typically performed on resource-constrained devices, there has been significant research in low-power inference on custom hardware accelerators. On the other hand, training Convolutional Neural Networks (CNNs) is more computationally and memory-intensive and is primarily done on power-hungry GPUs in large-scale data centers. However, CNN training on Field-Programmable Gate Arrays (FPGAs) is a relatively new area of research that offers potential for improved performance and energy efficiency in DL applications. In order to enable researchers to explore this field further, the authors present Barista - an automated toolflow that seamlessly integrates FPGAs into the training of CNNs within the popular deep learning framework Caffe. This work contributes to advancing the field of low-power DL by providing a much needed toolflow for FPGA based CNN training which allows efficient utilization of resources during model development - paving the way for improved performance and energy efficiency in DL applications.

What are FPGAs?

Field Programmable Gate Arrays (FPGAs) are integrated circuits that can be programmed after manufacture to perform specific tasks such as digital signal processing or image recognition algorithms. The advantage of using FPGAs over traditional CPUs or GPUs lies in their flexibility; they can be reprogrammed quickly without having to redesign hardware components from scratch each time a new application needs support. Furthermore, they offer better performance than CPUs or GPUs when running certain types of algorithms due to their parallelism capabilities and lower latency compared with software implementations running on CPUs/GPUs.

Why Use Barista?

Barista provides versatile and rapid deployment of hardware and algorithms for FPGA-based CNN training while also serving as an infrastructure for further research and development in this field. By enabling researchers to prototype and deploy various hardware and algorithmic techniques easily, Barista opens up new possibilities for exploring power-efficient CNN training on FPGAs. Furthermore, it is currently the only tool available that offers comprehensive support for FPGA-based CNN training within Caffe - making it an invaluable asset for anyone looking into this area of research or development.

How Does It Work?

Barista consists of two main components: a compiler backend which translates high level descriptions written by users into optimized RTL code suitable for synthesis onto an FPGA device; and a runtime system which manages communication between host CPU/GPU systems used during model development with offloaded operations executed by dedicated accelerator cores implemented inside an FPGA device.

The compilation process begins with user input describing desired network topology along with parameters such as batch size etc., followed by generation of intermediate representations suitable both for software simulation as well as synthesis onto target device(s). After successful compilation step, generated RTL code can be synthesized onto target platform resulting in bitstream ready to be loaded onto target board.

At runtime phase communication between host CPU/GPU systems used during model development takes place via PCIe interface allowing offloaded operations executed by dedicated accelerator cores implemented inside an FPGA device.

Once deployed onto target platform trained models can then be used either directly from embedded system itself or transferred back into host environment where they can be deployed at scale across distributed clusters.

Conclusion

By providing researchers with easy access to powerful tools like Barista that allow them to rapidly prototype various hardware architectures tailored specifically towards their use case scenarios while also offering comprehensive support within popular deep learning frameworks like Caffe – developers now have unprecedented opportunities when it comes down exploring power efficient solutions when it comes down deploying complex neural networks at scale.