An online guided tuning approach to run CNN pipelines on edge devices

Soomro, Pirah Noor; Abduljabbar, Mustafa; Castrillón, Jerónimo; Pericàs, Miquel

doi:10.1145/3457388.3458662

Cited by 6 publications

(4 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Pipe-Search [30] is an online approach that uses a database of pipeline configurations sorted w.r.t the distribution of workload among pipeline stages. It tests pipeline configurations of various depth and converges to a solution when no better solution is found by a time limit set by the user.…”

Section: Baseline and Test Applicationsmentioning

confidence: 99%

“…We demonstrate that Shisha is able to converge faster than existing algorithms (Simulated Annealing, Hill Climbing and Pipe-Search) and that it is able to find a solution within practical time limits. -We show that Shisha scales better with deeper CNNs and with larger amount of EPs per processing unit which is one of the limitations of prior online tuning approaches such as Pipe-Search [30].…”

Section: Introductionmentioning

confidence: 95%

“…In this context, trying and testing hundreds of schedules is acceptable, including schedules that are too expensive to test in an online setting [1,2]. Pipe-Search [30] adopts an online exploration approach for finding a pipeline configuration. Pipe-Search generates a database of pipeline configurations which is space-intensive and prohibitively slow for larger systems and deeper CNNs.…”

Section: Introductionmentioning

confidence: 99%

“…The configuration consists of the number of pipeline stages, CNN layers per pipeline stage and a mapping of pipeline stages to EPs. In the literature, various stochastic optimization and machine learning algorithms have been used such as Simulated Annealing [38], evolutionary algorithms [1,30], reinforcement learning [2,23] and deep neural network techniques [3]. The design space under consideration is large and complex, requiring tens of thousands of trials in order to reach a near optimum with current search schemes.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Shisha: Online scheduling of CNN pipelines on heterogeneous architectures

Soomro¹,

Abduljabbar²,

Castrillón³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Chiplets have become a common methodology in modern chip design. Chiplets improve yield and enable heterogeneity at the level of cores, memory subsystem and the interconnect. Convolutional Neural Networks (CNNs) have high computational, bandwidth and memory capacity requirements owing to the increasingly large amount of weights. Thus to exploit chiplet-based architectures, CNNs must be optimized in terms of scheduling and workload distribution among computing resources. We propose Shisha, an online approach to generate and schedule parallel CNN pipelines on chiplet architectures. Shisha targets heterogeneity in compute performance and memory bandwidth and tunes the pipeline schedule through a fast online exploration technique. We compare Shisha with Simulated Annealing, Hill Climbing and Pipe-Search. On average, the convergence time is improved by ∼ 35× in Shisha compared to other exploration algorithms. Despite the quick exploration, Shisha's solution is often better than that of other heuristic exploration algorithms.

show abstract

Section: Baseline and Test Applicationsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 95%