ESP for Machine Learning
Video
Team Information
Team Members
Kuan-lin Chiu, PhD Candidate, Computer Science, Graduate School of Arts and Sciences, Columbia University
Davide Giri, PhD Candidate, Computer Science, Graduate School of Arts and Sciences, Columbia University
Giuseppe Di Guglielmo, Associate Research Scientist, Department of Computer Science, School of Engineering and Applied Sciences (SEAS), Columbia Engineering
Paolo Mantovani, Associate Research Scientist, Department of Computer Science, School of Engineering and Applied Sciences (SEAS), Columbia Engineering
Faculty Advisor: Luca Carloni, Professor, Department of Computer Science, School of Engineering and Applied Sciences (SEAS), Columbia Engineering
Abstract
Recent advances in machine learning (ML) have depended on the continued progress of hardware computing platforms. Future advances will depend even more on the synergistic progress of hardware and software. The emerging open-source hardware community can play a unique role in supporting embedded ML research. We present ESP4ML, an open-source system-level design flow to build and program system-on-chip (SoC) for embedded applications that require the hardware acceleration of ML algorithms. We realized ESP4ML by combining two established open-source projects (ESP and HLS4ML) into a new, fully-automated design flow. ESP is our open-source research platform for agile heterogeneous SoC design that allows the rapid FPGA prototyping of complex SoCs. By combining a modular tile-based architecture with a flexible design methodology, ESP simplifies the development and integration of accelerators generated with the ESP design flows (including HLS4ML) as well as the reuse of third-party components, like the NVIDIA Deep Learning Accelerator (NVDLA).
Contact this Team
Team Contact: Davide Giri (use form to send email)
