< Back to previous page

Project

AITIA: Embedded AI Techniques for Industrial applications. (VLATETRA1)

New achievements in artificial machine learning are reported almost daily by the big firms. While those achievements are mainly based on high-end fast processing and massive data techniques, the potential of embedded machine learning techniques is still not understood well by the majority of the industrial players and SMEs. Nevertheless, the potential of machine learning, directly embedded in a device or system which is trained in an online or an offline fashion is perceived as very high.
This has led to a broad demand by industry and SMEs for a practical and application oriented feasibility study which helps them to understand the potential benefits but also the limitations of embedded Artificial Intelligence. Currently, the question where specific algorithms are realized, at the embedded device or in the cloud, is under discussion in several fields of applications. E.g. Xilinx supports the so called edge computing paradigm where algorithms are realized on FPGA based SoCs. On the other hand, they support also the high performance FPGA based cloud system with Amazon. Both realization alternatives have their pros and cons and need to be analysed according the respective application domain. Furthermore, the realization in general on embedded systems is a crucial challenge for SMEs.

This project aims at developing and demonstrating ‘best practices for embedded AI’ by means of four relevant industrial case studies. In those case studies we will tackle several elements which are related to technology, safety and certifiability.
We will mainly focus on four application domains relevant to the European industry and SMEs:
- Embedded Security: Intrusion detection by AI in network routers
- Smart Sensing: Improving sensor accuracy by AI and detecting sensor-anomalies by AI
- Automotive and Mobile Robots: Machine Vision for Mobile Systems
- Industry 4.0: Machine Learning for Predictive Maintenance
The technological challenges which we will tackle are related to:
- The use of FPGAs as accelerators for embedded AI, bit-width optimizations, soft
processors versus dedicated processors versus hard embedded processors,
dynamic reconfiguration techniques, frameworks for FPGA AI code generation
- Certifiability of embedded AI techniques
- Safety aspects of AI in machines and smart objects
Date:1 Sep 2019 →  31 Aug 2021
Keywords:Artificial Intelligence, Embedded System Design
Disciplines:Artificial intelligence not elsewhere classified
Project type:Collaboration project
Results:This research resulted in four use cases:
Use case 1 - VUB
Title: Demonstrator of Embedded Classic AI Algorithms for Acoustic Sound Classification
Abstract: This use-case investigates methods aiming at the automatic recognition and classification of Urban sounds, for the purpose of subsequently applying these methods to the recognition of soundscapes. Research in audio recognition has traditionally focused on the domains of speech and music. Comparatively little research has been done towards recognizing non-speech urban sounds. For this reason, in this project, we apply and evaluate the existing machine learning techniques that have been proved efficient in the other two domains. These techniques are comprehensively compared, paying a special attention on the performance and computational time to determine the most appropriate one for addressing the problem of urban sound recognition. Furthermore with the vast improvements on the manufacture of ICs, the best machine learning technique is implemented on an embedded device.

Use Case 2 - BTU
Title: An AI Use-Case on the FGPU: An Open-Source FPGA-based General-Purpose GPU
Abstract: This use-case demonstrates the application of a flexible FPGA based General-Purpose GPU (FGPU) to implement machine learning algorithms. This use-case is worked out by our partners from Brandenburg University of Technology. The Open Source FGPU has been developed at their research group and is made available for the broad community through Github.

Use Case 3 - TUD
Title: RISC-V based architectures for AI applications
Abstract: In this use-case a RISC-V based architectures is presented for artificial intelligence applications. RISC-V is an open source instruction set architecture developed by University of California, Berkeley. RISC-V is a very modular ISA and allows custom extensions based on application requirements. In this use-case, different RISC-V based architectures are investigated; multi-core RISC-V, a vector extension (SIMD) of a RISC-V based processor, and a HLS based library (HiFlipVX) for hardware accelerators for RISC-V processor. All investigated hardware approaches are focusing on FPGA implementations. An application mapping tool will also be presented to map complex applications on a heterogeneous architecture.

Use Case 4 - KU Leuven
Title: Deep learning for hardware accelerated network intrusion detection
Abstract: This use-case applies deep learning for hardware accelerated network intrusion detection. In network intrusion detection, the goal is to detect malicious network traffic and report it. The use-case first provides some insights in how this can be done using deep learning, and then it demonstrates the implementation of deep learning architectures on FPGA using the Xilinx FINN toolchain.