Embedded Artificial Intelligence:Systems And Applications
Sensors and related technologies have enjoyed considerable success, as demonstrated by the increasing number of “smart” devices in use. From bracelets to glasses, from smartphones to smart shoes, from smart devices in home automation to professional healthcare systems, the commercial success of these smart objects is the result of vast improvements in electronics that has led to an increase in performance, while reducing the operational power as well as by the availability of ubiquitous wireless connectivity. The clear trend is toward billions of smart sensors, in the “Internet of Things” (IoT) vision. In fact, IoT creates formidable challenges for academic and industrial research. In particular, these billions of smart devices with sensors will produce a mind-boggling quantity of data that need to be processed to provide useful information. In fact, the data by themselves are not able to provide value unless they are not processed. Big-data mining techniques allow us to gain new insights by batch-processing and off-line analysis. However, this is not sufficient: in-situ real-time feature extraction, analysis, classification, and local decision-making are essential for a truly scalable and robust IoT infrastructure.
Machine learning technologies are extensively used with great success in many application domains to solve real-world problems in entertainment e-health, automatic surveillance, assisted living, and assistant driving among many others. More and more researchers are tackling classification and decision-making problems with the help of brain-inspired algorithms (i.e. neural network or convolutional neural network), featuring many stages of feature extractors and classifiers with lots of parameters that are optimized using the unprecedented wealth of training data that is currently available. These brain-inspired techniques are also known as multi-layer neural networks. In recent years, neural network has achieved results that exceed those achieved by humans on very challenging problems and datasets, and routinely surpass more mature ad hoc approaches. Neural network algorithms are extremely flexible and applicable to various data sources. They perform at best when information is spatially or temporally well localized, but still has to be seen in a more global context. However, neural networks approaches are still not suitable for wearable, ultra-miniaturized IoT devices, because, in their current embodiments, they require massive amounts of computational power.
Many of them are in collaboration with Swiss's companies that are interested in IoT and especially autonomous sensing systems. The following projects focus on processing data with machine learning on low power single-core microcontrollers (i.e. Arm-Cortex-M family) or MultiCore (i.e. PULP Processor designed in IIS)
Depending on the applicant's profile and project type, his tasks may involve some of the following:
- lab. testing/characterization of the existing prototype: verification of the prototype's characteristics w.r. design specification (simulations), measuring power-consumption, and assessing detection performance in lab. Conditions
- High-level software programming, machine learning, wireless communication
- Programming the circuit for specific applications, field testing, data acquisition
- Designing PCB with micro-controllers, wireless interfaces, energy harvesting and sensors and test on the field.
- Firmware implementation for low power and application.
- Analog/digital design using discrete components.
(not all need to be met by the single candidate)
- Experience using the laboratory instrumentation - signal generators, oscilloscopes, DAQ cards, Matlab etc.
- analog electronics and signal conditioning with operational amplifiers: amplifiers, filters, integrators etc.
- knowledge of microcontroller programming and PC programming (C/C++, preferably microcontroller with Bluetooth Low Energy but it is not mandatory)
- basic knowledge on signal processing is a plus.
- plus is knowledge on printed circuit board (PCB) using Altium.
For a complete list of many cool projects please go to the following links
- Hyper-Dimensional Computing Based Predictive Maintenance
- Real-Time Motor-Imagery Classification Using Neuromorphic Processor
- Pressure and acoustic Smart Sensors Network for Wind Turbines Monitoring
- Neural Network Algorithms and Interfaces with Accelerators for Embedded Platforms with Real World Applications
- Design Of A Biomarker Assay Based On Responsive Magnetic Nanoparticles
- High-throughput Embedded System For Neurotechnology in collaboration with INI
- Neuromorphic Intelligence In An Embedded System in Collaboration with AiCTX
- Adversarial Attacks Against Deep Neural Networks In Wearable Cameras
- Wearable Smart Camera With Deep Learning Algorithms For Automatic Detecion
- Energy Efficient Smart Devices For Construction Building Maintenance Hilti Collaboration
- Improving Cold-Start in Batteryless And Energy Harvesting Systems
- Edge Computing for Long-Term Wearable Biomedical Systems
- Optogenetics And Game Theory Applied To Small Side Bird Using Smart Sensing
- Wireless Sensing With Long Range Comminication (LoRa)
- Indoor Smart Tracking of Hospital instrumentation
- A Wearable Wireless Kidney Function Monitoring System For BioMedical Applications
- Embedded Gesture Recognition Using Novel Mini Radar Sensors
- Contextual Intelligence on Resource-constraint Bluetooth LE IoT Devices