Collaboration

We explore efficient perception using bio-inspired sensors and neuromorphic computing paradigms, focusing on multimodal event-based visual attention and efficient hardware processing of asynchronous, sparse data.

We collaborate on compiler optimizations for FPGA dataflow accelerators of quantized neural networks built with the open-source FINN framework. Together we developed scaled-integer range analysis (SIRA), which tracks the range, scale and bias of every tensor to generate smaller and more efficient streaming hardware.

We collaborate on the efficient deployment of quantized neural networks on FPGAs using the open-source FINN dataflow compiler. Our joint work on scaled-integer range analysis (SIRA) reduces accumulator bitwidths and folds scales, biases and elementwise operations into thresholds, cutting the resource usage of FPGA dataflow accelerators.

We collaborate on energy-efficient hardware for AI at the edge, focusing on FPGA dataflow accelerators for quantized neural networks. Our joint work on scaled-integer range analysis (SIRA) uses interval arithmetic to derive the range, scale and bias of tensors, enabling narrower accumulators and cheaper streaming hardware.

We collaborated on implementation of graph neural networks for event-based data, mainly event audio sensors. We developed the first embedded solutions for audio classification and keyword spotting using SoC FPGA devices.

We collaborated on implementation of graph neural networks for event-based data, mainly event audio sensors. We developed the first embedded solutions for audio classification and keyword spotting using SoC FPGA devices.

We investigate the design of hardware-aware algorithms for event-based data processing on FPGAs, utilising audio data from Neuromorphic Auditory Sensors and visual data from Dynamic Vision Sensors. We combine our expertise in reconfigurable logic design with event-based perception.

We collaborated on applying Tangled Program Graphs (TPGs) to autonomous UAV control. We have compared their performance with deep reinforcement learning (DRL) in a task involving autonomous flight through an unknown environment.

We collaborated on an embedded vision system for automated drone landing site detection (eGPU and SoC FPGA). We published two papers: a demo at the DASIP 2018 conference and a paper at the ICCVG 2018 conference. Andre also served as a visiting professor in our group.

We have recently started a collaboration on unmanned aerial vehicles (UAVs), combining our expertise in embedded vision and hardware acceleration with SDU's experience in drone technology and field operations. Our team members spent a week at the SDU Drone Center in Odense, and we also met there as part of EuroDroS.

Tomasz Kryjak served as a visiting professor at Polytechnique Montréal for two weeks in April 2024, hosted by Pierre Langlois and Jean-Pierre David.

We collaborated on applying graph neural networks to event-based data classification. This work resulted in three conference papers (SPA 2023, DASIP 2024 and FPT 2024) and a paper in the Journal of Systems Architecture (JSA, 2026). Additionally, Tomasz Kryjak was a visiting professor at the LIP6 lab in November 2022.