Integrating Design Space Exploration in Modern Compilation Toolchains for Deep Learning
keywords ARTIFICIAL INTELLIGENCE, C, COMPILERS, CONVOLUTIONAL NEURAL NETWORKS, DEEP LEARNING, DEEP NEURAL NETWORKS, DESIGN SPACE EXPLORATION, EMBEDDED SYSTEMS, ENERGY EFFICIENCY, FIRMWARE, HARDWARE ACCELERATORS, LOW POWER, MICROCONTROLLERS, SOFTWARE, SOFTWARE ACCELERATION
Reference persons DANIELE JAHIER PAGLIARI
External reference persons Marian Verhelst (KU Leuven)
Alessio Burrello (Politecnico di Torino)
Research Groups DAUIN - GR-06 - ELECTRONIC DESIGN AUTOMATION - EDA, ELECTRONIC DESIGN AUTOMATION - EDA, GR-06 - ELECTRONIC DESIGN AUTOMATION - EDA
Thesis type EXPERIMENTAL, SOFTWARE DEVELOPMENT
Description Modern domain-specific compilers for Deep Learning have shown great advancements in optimizing code deployment, minimizing data transfers and optimizing the latency, throughput and energy consumption associated with the execution of Deep Neural Networks (DNN) on the target hardware platform.
In the same direction, new design space exploration (DSE) tools have been created to explore the vast space of DNN execution flows for a given hardware platform, while possibly also tuning the parameters of the hardware itself (HW-SW Co-design).
What is missing in modern solutions is a link between what DSE tools produce and what is deployable on hardware.
In this thesis, the candidate will integrate two tools:
- ZigZag, a powerful DSE tool used to optimize neural network loop ordering and tiling and to co-design a hardware accelerator and a NN schedule
- HTVM, a DNN compiler which exploits the open-source TVM tool, integrated with custom plug-ins to produce optimized code for heterogeneous platforms.
The goal of the thesis will be to connect the two tools and allow the automatic deployment of ZigZag-generated solutions, to enhance the performance of NN execution on edge devices.
Interested candidates must send an email to email@example.com attaching their CV and exams' transcript with scores.
Required skills Required skills include C and python programming. Further, a basic knowledge of computer architectures and embedded systems is necessary. Desired (but not required) skills include some familiarity with basic machine/deep learning concepts and corresponding models.
Notes Thesis in collaboration with Prof. Marian Verhelst’s research group at KU Leuven. The thesis can be carried out either in Torino or in Leuven, Belgium.
Deadline 14/12/2023 PROPONI LA TUA CANDIDATURA