PRIN 2022 - RETICULATE
RETICULATE: REal-TIme and seCUre acceLeration framework for Artificial inTElligence
Next-generation cyber-physical systems (CPS), such as autonomous vehicles and advanced robots, must rely on Artificial Intelligence (AI) to increase their level of autonomy. In particular, Deep Neural Networks (DNNs) are the de-facto approach for implementing perception tasks in CPSs. Although the support for DNN execution (i.e., the inference phase) is today quite mature for workstations and servers, DNN execution frameworks for resource-constrained embedded computing platforms are still lacking some peculiar features that are crucial to enable the “AI at the edge” paradigm for CPS.
RETICULATE aims at filling this gap by tackling the following technical challenges:
- TC1. Enable time-predictable acceleration of DNN with real-time performance.
- TC2. Support design space exploration for DNN accelerators while accounting for timing requirements.
- TC3. Contain energy consumption during DNN acceleration.
- TC4. Ensure integrity and confidentiality of DNN models on embedded computing platforms.
- TC5. Protect DNN from adversarial attacks when accelerated on embedded computing platforms.
- TC6. Equip the underlying technologies (IoT- and edge-class devices and FPGA platforms) with consistent support for establishing a trusted execution environment.
ENTE PROMOTORE: MUR
NOME PROGETTO: RETICULATE
PERIODO E DURATA: 24 mesi, 28/09/2023 – 27/09/2025
FINANZIAMENTO: € 99.659
COORDINATORI: Scuola Superiore Sant’Anna di Pisa
REFERENTI SSSA: Alessandro Biondi
