An Approximation Workflow for Exploiting Data-Level Parallelism in FPGA Acceleration

Lotfi, Atieh; Rahimi, Abbas; Yazdanbakhsh, Amir; Esmaeilzadeh, Hadi; Gupta, Rajesh K.

doi:10.1007/978-3-319-53768-9_10

Cited by 4 publications

(2 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [16], a Bayesian network learns the cost and error models of an optimization problem offline and determines the quality control knob by solving this optimization problem. Rahimi et al [17] selectively reduce and dynamically tune the precision, subjecting to a statistical quality knob; consequently, an approximator within the fixed area budget can accommodate more parallel approximate kernels. These works strive to maximize the invocation of a single approximation scheme.…”

Section: A Model Based Quality Control For Approximate Computingmentioning

confidence: 99%

See 1 more Smart Citation

Invocation-driven neural approximate computing with a multiclass-classifier and multiple approximators

Song

et al. 2018

Proceedings of the International Conference on Computer-Aided Design

View full text Add to dashboard Cite

Neural approximate computing gains enormous energy-efficiency at the cost of tolerable quality-loss. A neural approximator can map the input data to output while a classifier determines whether the input data are safe to approximate with quality guarantee. However, existing works cannot maximize the invocation of the approximator, resulting in limited speedup and energy saving. By exploring the mapping space of those target functions, in this paper, we observe a nonuniform distribution of the approximation error incurred by the same approximator. We thus propose a novel approximate computing architecture with a Multiclass-Classifier and Multiple Approximators (MCMA). These approximators have identica network topologies, and thus can share the same hardware resource in an neural processing unit(NPU) clip. In the runtime, MCMA can swap in the invoked approximator by merely shipping the synapse weights from the on-chip memory to the buffers near MAC within a cycle. We also propose efficient co-training methods for such MCMA architecture. Experimental results show a more substantial invocation of MCMA as well as the gain of energy-efficiency.

show abstract

Section: A Model Based Quality Control For Approximate Computingmentioning

confidence: 99%

“…These works then advocate a rollback or online tuning mechanism to guarantee the computation reliability. On the other hand, a proactive strategy is to predict whether the load is safe-to-approximate or not using statistical and learning models [12], [14]- [17].…”

Section: Introductionmentioning

confidence: 99%