scholarly journals Composition of Weighted Finite Transducers in MapReduce

2020 ◽  
Author(s):  
Bilal Elghadyry ◽  
Faissal Ouardi ◽  
Sébastien Verel
Keyword(s):  
Speech Processing ◽  
Large Scale ◽  
Large Scale Data ◽  
Finite State Transducers ◽  
Wide Range ◽  
Finite State ◽  
Common Operation ◽  
Efficient Representation ◽  
Weighted Finite State Transducers ◽  
Np Hardness

Abstract Weighted finite-state transducers have been shown to be a general and efficient representation in many applications such as text and speech processing, computational biology, and machine learning. The composition of weighted finite-state transducers constitutes a fundamental and common operation between these applications. The NP-hardness of the composition computation problem presents a challenge that leads us to devise efficient algorithms on a large scale when considering more than two transducers. This paper describes a parallel computation of weighted finite transducers composition in MapReduce framework. To the best of our knowledge, this paper is the first to tackle this task using MapReduce methods. First, we analyze the communication cost of this problem using Afrati et al. model. Then, we propose three MapReduce methods based respectively on input alphabet mapping, state mapping, and hybrid mapping. Finally, intensive experiments on a wide range of weighted finite-state transducers are conducted to compare the proposed methods and show their efficiency for large-scale data.

scholarly journals Composition of weighted finite transducers in MapReduce

2021 ◽  
Vol 8 (1) ◽  
Author(s):  
Bilal Elghadyry ◽  
Faissal Ouardi ◽  
Sébastien Verel
Keyword(s):  
Speech Processing ◽  
Large Scale ◽  
Large Scale Data ◽  
Finite State Transducers ◽  
Wide Range ◽  
Finite State ◽  
Common Operation ◽  
Efficient Representation ◽  
Weighted Finite State Transducers ◽  
Np Hardness

AbstractWeighted finite-state transducers have been shown to be a general and efficient representation in many applications such as text and speech processing, computational biology, and machine learning. The composition of weighted finite-state transducers constitutes a fundamental and common operation between these applications. The NP-hardness of the composition computation problem presents a challenge that leads us to devise efficient algorithms on a large scale when considering more than two transducers. This paper describes a parallel computation of weighted finite transducers composition in MapReduce framework. To the best of our knowledge, this paper is the first to tackle this task using MapReduce methods. First, we analyze the communication cost of this problem using Afrati et al. model. Then, we propose three MapReduce methods based respectively on input alphabet mapping, state mapping, and hybrid mapping. Finally, intensive experiments on a wide range of weighted finite-state transducers are conducted to compare the proposed methods and show their efficiency for large-scale data.

Abstract 37: ASPECTS and Stratified Outcomes After Endovascular Therapy in the Trevo Retriever Registry: Benefit in Low ASPECTS

Stroke ◽  
2017 ◽  
Vol 48 (suppl_1) ◽  
Author(s):  
David S Liebeskind ◽  
Graham W Woolf ◽  
Bin Xiang ◽  
Ryan Shields ◽  
Erol Veznedaroglu ◽  
...  
Keyword(s):  
Endovascular Therapy ◽  
Large Scale ◽  
Arterial Occlusion ◽  
Routine Practice ◽  
Anterior Circulation ◽  
Endovascular Thrombectomy ◽  
Large Scale Data ◽  
Wide Range ◽  
Therapy Studies ◽  
Occlusion Effect

Background: Most endovascular stroke therapy studies and subsequent guidelines restrict intervention based on ASPECTS. A wide range of ASPECTS scores may be encountered in practice and individual patient benefit may be realized even at low ASPECTS. We examined large-scale data on outcomes after endovascular therapy, stratified by baseline ASPECTS in the Trevo Retriever Registry. Methods: The independent Imaging Core Lab of the Trevo Retriever Registry prospectively determines ASPECTS on baseline imaging acquired immediately prior to endovascular thrombectomy. ASPECTS scores and regional involvement were analyzed with respect to site of arterial occlusion, effect of time from symptom onset, co-morbidities and clinical outcomes, based on ASPECTS strata. Results: Baseline ASPECTS data was reviewed by the Imaging Core Lab in 426 subjects with anterior circulation stroke enrolled in the Trevo Retriever Registry, as of July 2016. Mean age was 68.8 ± 13.7 yrs, with 20.9% > 80 years old. Baseline NIHSS was median 15.0 (10.0, 19.0). Onset to CT was median 3.8 (1.5, 9.0) hrs, with median ASPECTS of 8.0 (7.0, 9.0), ranging from 3-10. Baseline ASPECTS 0-7 occurred in 118/426 (27.7%) subjects, including 39.0% of ICA, 27.1% M1 and 16.9% M2/3 arterial occlusions at angiography. Baseline clinical variables predicting ASPECTS included age and NIHSS, whereas the ASPECTS score was mildly associated with final TICI2C reperfusion (r=0.24, p<0.001). Subsequent symptomatic ICH was 1.7% with baseline ASPECTS 0-7 versus 2.0% with ASPECTS 8-10. The distribution of mRS at 90 days based on individual ASPECTS strata from 10 to 3 revealed a trend to worse outcomes with lower ASPECTS, yet good outcomes (mRS 0-2) were 60.7% (ASPECTS 10), 55.3% (9), 60.2% (8), 54.9% (7), 55.1% (3-6). Conclusions: Discrete ASPECTS strata may influence outcomes of endovascular therapy conducted in routine practice around the world, yet individuals with low ASPECTS may still achieve reasonable outcomes.

Sign in / Sign up

Export Citation Format

Share Document