In this HiPEAC 2019 workshop we aim to bring together researchers working in Machine Learning and System Architectures to discuss requirements, opportunities, challenges and next steps in developing novel approaches for accelerating deep neural networks.
Find Out MoreDeep Learning is receiving much attention these days due to remarkable performance achieved in several fields (e.g. Computer Vision, Speech, Translations, etc), although this brings some challenges to hardware architects and computation optimization researchers. Deep Learning models are generally very large in memory and require many computation instructions to train and perform inferences. Accelerating these operations has obvious advantages, first by reducing the energy consumption (e.g. in data centers) and secondly, making these models usable on smaller devices at the edge of the Internet. This workshop aims to enable discussions on emerging acceleration techniques and computation paradigms for deep learning algorithms. The timing of this workshop is ideal, with European regulations tightening data privacy, thus forcing more computations/inferences to be performed at the Edge.
Topics of interest include (but are not limited to):
Novel parallel computing architectures: GPUs, FPGAs, and heterogeneous multi/many-core designs;
Crazy architectural ideas: focused on accelerating deep learning workloads/algorithms;
Cloud and edge computing: hardware and software methods focused on accelerating both training (cloud) and inference (edge);
Software accelerators: primitives, libraries, compilers and frameworks;
Submission deadline: November 9th November 23th, 2018 (11:59 PM PDT)
Notification to authors: December 18th, 2018
Camera-ready deadline: January 10th, 2019
Regular (up to 9 pages) or short (up to 5 pages) paper using SIGCHI Extended Abstract format. Papers should be in PDF format and not anonymized.
Submissions can be made at easychair.org/conferences/?conf=edla2019.
Papers will be reviewed by the workshop's technical program committee according to criteria regarding a submission's quality, relevance to the workshop's topics, and, foremost, its potential to spark discussions about directions, insights, and solutions in the context of deep learning accelerators. Research papers, case studies, and position papers are all welcome.
In particular, we encourage authors to keep the following options in mind when preparing submissions:
Tentative Research Ideas: Presenting your research idea early one to get feedback and enable collaborations.
Works-In-Progress: To facilitate sharing of thought-provoking ideas and high-potential though preliminary research, authors are welcome to make submissions describing early-stage, in-progress, and/or exploratory work in order to elicit feedback, discover collaboration opportunities, and generally spark discussion.
Director, ARCO Research Group
Professor, Computer Architecture Department
Universitat Politècnica de Catalunya, Barcelona
Title: Milliwatt Human-Quality Speech Recognition
Abstract
Automatic speech recognition (ASR) will be a key feature for many computing systems, and in particular for mobile devices such as smartphones, tablets, home devices and wearables. For instance, ASR technology is at the heart of popular applications with voice-based user interfaces for mobile devices such as Google Now, Apple Siri, Microsoft Cortana or Amazon Alexa. These systems require support for real-time, large-vocabulary, speaker-independent, highly-accurate, continuous speech recognition. Unfortunately, supporting fast and accurate speech recognition requires a huge computational power, which is especially challenging to attain in devices with very tight constraints in energy consumption. Improvements in energy-efficiency is a key requirement for high-fidelity ASR.
The main driving forces in the past for improving energy-efficiency were based on process technology and microarchitecture innovations in general-purpose processors. However, both of them are reaching a point of diminishing returns. On the other hand, domain-specific architectures offer great potential to keep delivering dramatic improvements in energy-efficiency, and we believe they will become a key ingredient of future computing systems.
In this talk, we will first review the main trends in computing and the state-of-the-art approaches for ASR and then, we will present a novel domain-specific architecture that provides dramatic improvements in terms of energy-efficiency for ASR.
Time | Event (in room 5) 21st January 2019 |
---|---|
14:00–14:05 | Welcome (José Cano, Valentin Radu) |
14:05–15:00 | Keynote: Milliwatt Human-Quality Speech Recognition (Antonio González) |
15:00–15:30 | Paper presentations
|
15:30–16:00 | Coffee break |
16:00–17:30 | Paper presentations
|
17:30–17:35 | Closing remarks (José Cano, Valentin Radu) |