AI & math-modelling.

Projects that TOPIC realised within the domain of AI & math-modelling: 

- Measurement and analysis of microvascular health
- Preventative maintenance 
- Domestic monitoring system

Key technologies and skills: 

Neural network implementations using Tensor Flow or PyTorch/edge and cloud solutions using TPU, DPU and GPU/ implementation of research level algorithms/acceleration of algorithms on multi-core processor architectures. 

TOPIC contribution and know-how:

Our expertise in AI will bring added-value to our clients' development projects as we are able to:
- offer a solution where AI is an additional function to improve effectiveness
- bring improvement in the richness of sensor acquisition
- improve effectiveness of algorithms
- allow for deployment on edge devices or as part of a cloud solution
- understand what type of measurement through sensors needs to be done based on significant hardware and firmware experience

Our skills in Math Modelling facilitate the development, conceptualisation and implementation of algorithms on embedded systems (using Matlab/Simulink/Python) and the acceleration of algorithms on embedded devices. Our approach helps porting of academic and research concepts using abstract modelling into products that meet performance, cost and power requirements. The mathematical and physics expertise within TOPIC helps understanding the conceptual or simulation models. Using our vast programming skills an efficient translation from model into embedded software is reliably realized.

Domestic monitoring system

TOPIC designed and developed a compact autonomous observation system where, with multiple cameras, a full 360 degree view was created. This project was realised for one of our customers focused on the consumer market.

We implemented object recognition on the device by developing and implementing an AI algorithm. Using the Tensor Flow we mapped this on a neural network in the FPGA fabric of our Miami SOM using the DPU IP of AMD. 

The system is connected to the internet via a WiFi connection. Detected objects form a region-of-interest, can be compressed and uploaded to a cloud server. A large SSD is applied to facilitate fast storage and retrieval of recordings using the benefits of M2 PCIe based interfaces. Running embedded Linux, the system management is simple, robust and provides services like remote updates and diagnostics. Given the privacy-sensitive usage, data security is one of the key-features of the implementation.