Open Source Summit Japan & Automotive Linux Summit 2020

Smart speakers and personal assistants have taken a strong presence in our daily life: nearly every modern smart phone comes with a personal voice assistant, and smart speakers a la Google Home and Amazon's Alexa/Echo can be found in many households. But these commodities come at a price: The price is loss of privacy, the price of ever more transparent customer, not to speak of unwanted transcription of private communication due to the permanent listening for trigger words. A team of software and hardware developers at FOSSASIA is working on an alternative: a open source personal voice assistant and smart speaker that respects the privacy of the user. This system is called SUSI.AI, and targets not only the dedicated smart speaker, but also the general Linux desktop with integration into the major desktop environments. Norbert Preining will present the state of SUSI.AI, both on the desktop as well as the smart speaker. After an introduction to the SUSI.AI ecosystem, he will focus on the integration of recent offline speech recognition systems into the device, and the state of open source speech recognition in general.

Edge computing is becoming the way data is being handled to reduce unnecessary data transfer, better latency and data security considerations. To bring AI to the Edge and provide AI as a Service (AI-aaS), AI edge stack should provide Edge ML platforms, libraries, computing, and edge data platforms. This talk will describe how to build Cloud Native Edge Stack service framework based on Kubedge (a CNCF Sandbox project), providing an edge-cloud collaborative AI service. It shares the details of edge AI stack architecture and design, edge-cloud application deployment, collaborative model management, dataset management etc. The work is part of Akraino KubeEdge Service Blueprint open source project. A demo will be also detailed to show case mobile device offloading, collaborative training (aka federated learning) can be easily achieved using the stack. The future roadmap will also be shared.

Around 2006, Apache Hadoop realized the open source based “Data Lake” architecture for enterprises to utilize large amounts of data, "Big Data". However, there are also growing expectations against "real-time analysis" that delivers analyzed results to end-users in seconds to minutes by immediately processing a large amount of “stream data”. In this talk, we present the history of open source software related to Data Lake, the overview of current software, and the potential tradeoffs.We also talk about how recent storage technologies, such as Apache Iceberg, Apache Hudi, Delta Lake, try to provide features to leverage both of historical and stream data on Data Lake in a different way from Lambda Architecture. Finally, we summarize these products based on the comparison of internal architectures. Attendances will learn about the overview of current storage software, and similarities and differencesof architectures. This helps you to design the system architecture build on Data Lake technologies to realize both batch and real-time based analysis. This post reflects some software upgrades from previous domestic presentation.

NEC has continually contributed to the Open Source ecosystem for over 20 years, and we are working together in many Open Source communities like Kubernetes, Linux, Hyperledger and OpenStack, etc.

In this panel discussion, we will gather Open Source contributors from different communities and discuss what are the hard challenges to contribute to Open Source and how to overcome those difficulties.

In addition, there are benefits from companies' viewpoints and fun from the developers' viewpoints. These factors motivate contributors well, so we will enjoy discussing them also in this panel discussion.

Giant Hogweed is a highly toxic plant originating in the Western Caucasus. It has spread across Central and Western Europe and there are sightings of Giant Hogweed reported from North America, too. Landowners are obliged to eradicate it, due to its toxicity and invasive nature in Europe. However, it is difficult for landowners to find and remove Giant Hogweed across large areas of land since it is a very cumbersome manual process. To automate the process of detecting the Giant Hogweed by exploiting technologies like drones and image recognition/detection using Machine Learning is an effective way to address this problem. In this presentation, we show you how we designed the architecture towards the Petabyte scale, how we took advantage of both of Big Data and Machine / Deep Learning technologies and lessons learned through this project. For example, we integrated a drone, Apache Hadoop, Apache Spark and TensorFlow to achieve the usability, flexibility and scalability for both of data engineers and data analysts. We talk about why this integration was needed for us, technical challenges from the viewpoint of enterprises and tips to leverage the above open source software.

Deep learning technologies for images are adopted in various kinds of business applications such as autonomous driving and cancer prediction. Preprocessing massive amounts of images in a short time is necessary for building models that are applied to business applications. Popular deep learning frameworks such as PyTorch offer their preprocessing libraries such as torchvision, which are compatible with those deep learning frameworks.

We evaluated the performance of preprocessing libraries and the effect of the preprocessing tuning in deep learning frameworks.

We adopted MLPerf, which is the popular benchmark for machine learning applications, and confirmed that parallelization and queueing could reduce the preprocessing time dramatically in both torchvision and OpenCV case.

However, in PyTorch framework, preprocessing and training are conducted in parallel, the effect of the preprocessing tuning can be small when the training workload is larger than the preprocessing workload.

We concluded that whether we should tune preprocessing or training depends on the purpose of the machine learning system such as parameter search and transfer learning.

To achieve a goal of accident-free roads with increased road safety, 5G C-V2X and autonomous driving ecosystem players are developing AI/ML enabled solutions leveraging the data from the in-vehicle sensors and traffic infrastructure. While conventional ML approaches rely on centralized servers to process the data from distributed entities, this is not always feasible in this case because of the inaccessibility of private data and large communication overhead required to transmit raw data to central ML processors. In such cases, decentralized approaches of federated learning, will be a good fit. In this presentation, we will discuss how technologies like edge computing and AI on edge can help develop context-aware 5G C-V2X use cases and provide cooperative awareness to each C-V2X enabled devices in vicinity. We will also talk about the open source frameworks to implement distributed V2X use cases and focus on key technical challenges of federated learning in wireless communications