Publications
“PERTINENCE: Input-based Opportunistic Neural Network Dynamic Execution”,
Omkar Shende, Gayathri Ananthanarayanan, and Marcello Traiola
arXiv, jul, 2025.
DOI: 10.48550/arXiv.2507.01695
Events and scientific dissemination
06/06/2025 – Scientific seminar at Inria Center at Rennes University by Gayathri Ananthanarayanan, ass. professor, Department of Computer Science and Engineering of the Indian Institute of Technology (IIT) Dharwad.
“Towards Efficient, Adaptive, and Dynamic Neural Inference on Heterogeneous Platforms”
Abstract: Deploying deep neural networks (DNNs) on edge devices presents unique challenges due to limited computational resources, diverse hardware architectures, and dynamic runtime conditions. This talk presents a comprehensive overview of our research efforts aimed at making DNN inference on heterogeneous edge platforms more efficient, adaptive, and responsive to real-world constraints.
The talk will begin with the introduction of a throughput-oriented scheduling framework tailored for ARM big.LITTLE processors, which exploits pipeline parallelism to maximize inference performance. Next, I will discuss a co-execution-aware framework that intelligently maps multiple DNNs to various heterogeneous accelerators based on user-defined trade-offs between power, throughput, and accuracy. The talk will also introduce PERTINENCE, an input-aware inference strategy that dynamically selects the most suitable model from a pool of pre-trained networks based on input complexity, achieving significant computational savings without compromising accuracy.
26/06/2025 – Keynote talk at The 13th Prague Embedded Systems Workshop (PESW) 2025, Marcello Traiola, Ph.D. (Inria centre at Rennes University, France)
“Toward Adaptive Embedded Systems: from Multi-Objective Design to Runtime Adaptation”,
Abstract: Embedded systems operate under tight and often conflicting constraints, such as energy, performance, accuracy, and reliability. Optimizing across these dimensions is complex, especially as systems must now operate under dynamic and unpredictable conditions.
This talk presents a vision for a two-phase approach to embedded system design that brings together design-time optimization and runtime adaptation. We begin with concrete examples of design space exploration, where multi-objective optimization techniques are used to identify Pareto-optimal configurations that span energy, reliability, and accuracy trade-offs. These configurations serve as a foundation for system flexibility.
Next, we shift focus to runtime adaptation, showcasing examples where systems can dynamically adapt to real-time conditions such as workload variations or energy constraints, enabling a new generation of adaptive, context-aware embedded architectures.