Search Results

Documents authored by Stechele, Walter


Document
HW-Flow: A Multi-Abstraction Level HW-CNN Codesign Pruning Methodology

Authors: Manoj-Rohit Vemparala, Nael Fasfous, Alexander Frickenstein, Emanuele Valpreda, Manfredi Camalleri, Qi Zhao, Christian Unger, Naveen-Shankar Nagaraja, Maurizio Martina, and Walter Stechele

Published in: LITES, Volume 8, Issue 1 (2022): Special Issue on Embedded Systems for Computer Vision. Leibniz Transactions on Embedded Systems, Volume 8, Issue 1


Abstract
Convolutional neural networks (CNNs) have produced unprecedented accuracy for many computer vision problems in the recent past. In power and compute-constrained embedded platforms, deploying modern CNNs can present many challenges. Most CNN architectures do not run in real-time due to the high number of computational operations involved during the inference phase. This emphasizes the role of CNN optimization techniques in early design space exploration. To estimate their efficacy in satisfying the target constraints, existing techniques are either hardware (HW) agnostic, pseudo-HW-aware by considering parameter and operation counts, or HW-aware through inflexible hardware-in-the-loop (HIL) setups. In this work, we introduce HW-Flow, a framework for optimizing and exploring CNN models based on three levels of hardware abstraction: Coarse, Mid and Fine. Through these levels, CNN design and optimization can be iteratively refined towards efficient execution on the target hardware platform. We present HW-Flow in the context of CNN pruning by augmenting a reinforcement learning agent with key metrics to understand the influence of its pruning actions on the inference hardware. With 2× reduction in energy and latency, we prune ResNet56, ResNet50, and DeepLabv3 with minimal accuracy degradation on the CIFAR-10, ImageNet, and CityScapes datasets, respectively.

Cite as

Manoj-Rohit Vemparala, Nael Fasfous, Alexander Frickenstein, Emanuele Valpreda, Manfredi Camalleri, Qi Zhao, Christian Unger, Naveen-Shankar Nagaraja, Maurizio Martina, and Walter Stechele. HW-Flow: A Multi-Abstraction Level HW-CNN Codesign Pruning Methodology. In LITES, Volume 8, Issue 1 (2022): Special Issue on Embedded Systems for Computer Vision. Leibniz Transactions on Embedded Systems, Volume 8, Issue 1, pp. 03:1-03:30, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2022)


Copy BibTex To Clipboard

@Article{vemparala_et_al:LITES.8.1.3,
  author =	{Vemparala, Manoj-Rohit and Fasfous, Nael and Frickenstein, Alexander and Valpreda, Emanuele and Camalleri, Manfredi and Zhao, Qi and Unger, Christian and Nagaraja, Naveen-Shankar and Martina, Maurizio and Stechele, Walter},
  title =	{{HW-Flow: A Multi-Abstraction Level HW-CNN Codesign Pruning Methodology}},
  journal =	{Leibniz Transactions on Embedded Systems},
  pages =	{03:1--03:30},
  ISSN =	{2199-2002},
  year =	{2022},
  volume =	{8},
  number =	{1},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/LITES.8.1.3},
  doi =		{10.4230/LITES.8.1.3},
  annote =	{Keywords: Convolutional Neural Networks, Optimization, Hardware Modeling, Pruning}
}
Document
Lessons Learned from last 4 Years of Reconfigurable Computing

Authors: Walter Stechele, Christopher Claus, and Andreas Laika

Published in: Dagstuhl Seminar Proceedings, Volume 10281, Dynamically Reconfigurable Architectures (2010)


Abstract
Partial dynamic reconfiguration of FPGAs was investigated for video-based driver assistance applications during the last 4 years. High-level application software was combined with dynamically reconfigurable hardware accelerators in selected scenarios, e.g. vehicle lights detection, optical flow motion detection. From the beginning of the project, various research challenges have been targeted, including hardware/software partitioning between embedded RISC and accelerators, granularity of reconfigurable regions, as well as the impact of the reconfiguration process on system performance. This article will review the status of these research challenges and present an outlook on future challenges, including reconfiguration look ahead. Challenges will be illustrated on robotic vision scenarios with dynamically changing computational load from soft real-time and hard real-time applications.

Cite as

Walter Stechele, Christopher Claus, and Andreas Laika. Lessons Learned from last 4 Years of Reconfigurable Computing. In Dynamically Reconfigurable Architectures. Dagstuhl Seminar Proceedings, Volume 10281, pp. 1-7, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2010)


Copy BibTex To Clipboard

@InProceedings{stechele_et_al:DagSemProc.10281.8,
  author =	{Stechele, Walter and Claus, Christopher and Laika, Andreas},
  title =	{{Lessons Learned from last 4 Years of Reconfigurable Computing}},
  booktitle =	{Dynamically Reconfigurable Architectures},
  pages =	{1--7},
  series =	{Dagstuhl Seminar Proceedings (DagSemProc)},
  ISSN =	{1862-4405},
  year =	{2010},
  volume =	{10281},
  editor =	{Peter M. Athanas and J\"{u}rgen Becker and J\"{u}rgen Teich and Ingrid Verbauwhede},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/DagSemProc.10281.8},
  URN =		{urn:nbn:de:0030-drops-28352},
  doi =		{10.4230/DagSemProc.10281.8},
  annote =	{Keywords: Reconfigurable computing, vision-based driver assistance}
}
Document
Dynamically Reconfigurable Systems-on-Chip

Authors: Walter Stechele

Published in: Dagstuhl Seminar Proceedings, Volume 6141, Dynamically Reconfigurable Architectures (2006)


Abstract
The design space for dynamically reconfigurable SoCs can be seen in three dimensions: 1) the system architecture for computation and communication, ranging from dataflow-oriented dedicated logic blocks to instruction flow-oriented microprocessor cores, from dedicated point-to-point connections to Networks-on-Chip. 2) the granularity of reconfigurable elements, ranging from simple logic Look-Up-Tables to complex hardware accelerator engines and reconfigurable interconnect structures. 3) the configuration life cycle, ranging from application changes (in the order of seconds) to instruction-based reconfiguration (in the order of nanoseconds). We propose to use dynamically reconfigurable computing for video processing in driver assistance applications. In future automotive systems, video-based driver assistance will improve security. Video processing for driver assistance requires real time implementation of complex algorithms. A pure software implementation, based on low cost embedded CPUs in automotive environments, does not offer the required real time processing. Therefore hardware acceleration is necessary. Dedicated hardware circuits (ASICs) can offer the required real time processing, but they do not offer the necessary flexibility. Specific driving conditions, e.g. highway, country side, urban traffic, tunnel, require specific optimized algorithms. Reconfigurable hardware offers high potential for real time video processing and adaptability to various driving conditions. Our system architecture consists of embedded CPU cores for high-level application code, dedicated hardware accelerator engines for low level pixel processing, and an application-specific memory system. The hardware accelerators and the memory system are dynamically reconfigurable, i.e. hardware accelerator engines can be exchanged during runtime, controlled by the application code on the CPU. The life cycle of a configuration depends on the change of driving conditions. A requirement on the reconfiguration time is given by the frame rate of the video signal, e.g. 40 msec for the exchange and relocation of new engines.

Cite as

Walter Stechele. Dynamically Reconfigurable Systems-on-Chip. In Dynamically Reconfigurable Architectures. Dagstuhl Seminar Proceedings, Volume 6141, p. 1, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2006)


Copy BibTex To Clipboard

@InProceedings{stechele:DagSemProc.06141.6,
  author =	{Stechele, Walter},
  title =	{{Dynamically Reconfigurable Systems-on-Chip}},
  booktitle =	{Dynamically Reconfigurable Architectures},
  pages =	{1--1},
  series =	{Dagstuhl Seminar Proceedings (DagSemProc)},
  ISSN =	{1862-4405},
  year =	{2006},
  volume =	{6141},
  editor =	{Peter M. Athanas and J\"{u}rgen Becker and Gordon Brebner and J\"{u}rgen Teich},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/DagSemProc.06141.6},
  URN =		{urn:nbn:de:0030-drops-7446},
  doi =		{10.4230/DagSemProc.06141.6},
  annote =	{Keywords: Dynamic reconfiguration, design space, video processing}
}
Questions / Remarks / Feedback
X

Feedback for Dagstuhl Publishing


Thanks for your feedback!

Feedback submitted

Could not send message

Please try again later or send an E-mail