The International Conference on Machine Learning (ICML) is one of the fastest growing artificial intelligence conferences in the world. Gathering the professionals, dedicated to the advancement of the artificial intelligence, the ICML is presenting and publishing the cutting-edge research on all aspects of machine learning used in data science, statistics, speech recognition and robotics. This year the conference hosted over 6,000 participants and began on Sunday June 9th with a full exposition consisting of talks, panels, demonstrations and workshops.
Co-organised by JOANNEUM RESEARCH, the joint workshop on "On-Device Machine Learning & Compact Deep Neural Network Representations" took place on June 14th, was a part of ICML 2019 and brought together more than 150 researchers and practitioners working on the compression of neural network and optimised architectures for hardware acceleration of neural networks. The aims of the workshop were to establish close connection between researchers and engineers in industry and to evaluate and compare the resource-efficient machine learning methods, compact and efficient network representations and their relation to particular target platforms.
With the rapidly growing number of applications of neural networks there is the increased need for scalability, and for enabling the use of these technologies on resource constrained devices such as mobile phones, smart cameras or embedded processors in vehicles. In content analysis applications, for example in the Horizon 2020 project MARCONI, local processing can solve privacy issues, as the input data does not need to be sent to a central server.
The workshop hosted invited talks with speakers from MIT, Google, IBM and NVIDIA, five oral presentations, a poster session and a panel discussion on the research challenges ahead. The participants noted the growing number of target hardware platforms coming with their specific toolkits for optimization, creating interoperability challenges. Thus standardization activities on exchange formats and representation of compressed networks are required. One example is the standardisation activity in MPEG, to which JOANNEUM RESEARCH is contributing.
More information: Connected Computing Group @ DIGITAL