Denoising with Early Exit

This repository contains the point demo for the GNA use-case of speech enhancement using early-exiting neural networks.

Try the demo

---

Additional information

This work was carried out collaboratively by two partners UIR and GNA.
The demo was produced by Alaa Zniber (UIR).

Efficient Hyperspectral Imaging

This repository contains the point demo for the TASE use-case of efficient hyperspectral imaging, esp. band selection and image segmentation through early exiting.

Try the demo

---

Additional information

This work was carried out collaboratively by two partners UIR and TASE.
The demo was produced by Ikram El Hajri and Saad Mokssit (UIR).

Approximate CGRA (AxCGRA)

This demo presents the Approximate CGRA (AxCGRA), which implements a heterogeneous approach to classical CGRA architectures by combining them with state-of-the-art approximate components to balance error introduced at the operation level with exergy and area considerations.

Design space exploration is used to enable the selection of appropriate approximate components, supporting balanced error–energy–area trade-offs. A dynamic mapping scheme guides DNN compilation, allowing the system to be configured for either error resilience or energy efficiency based on user-defined Quality of Result (QoR) constraints.

Watch the demo

---

Additional information

This work was carried out by ICCS.