Ultra-Fast Data-Mining Hardware Architecture Based on Stochastic Computing by Antoni Morro, et al.
Abstract:
Minimal hardware implementations able to cope with the processing of large amounts of data in reasonable times are highly desired in our information-driven society. In this work we review the application of stochastic computing to probabilistic-based pattern-recognition analysis of huge database sets. The proposed technique consists in the hardware implementation of a parallel architecture implementing a similarity search of data with respect to different pre-stored categories. We design pulse-based stochastic-logic blocks to obtain an efficient pattern recognition system. The proposed architecture speeds up the screening process of huge databases by a factor of 7 when compared to a conventional digital implementation using the same hardware area.
I haven’t included the hyperlinks, but:
…
In this work we present a highly efficient methodology for data mining based on probabilistic processing. High dimensional data is inherently complex in clustering, classification and similarity search [15]. The proposed approach is evaluated showing its application to a similarity search over a huge database. Most data mining algorithms use similarity search as a subroutine core [16–18], and thus the time taken for this task is the bottleneck of virtually all data mining algorithms [19]. Similarity search plays a fundamental role in many data mining and machine learning problems, e.g. text categorization [20], collaborative filtering [21], time-series analysis [22,23], protein sequencing [24] or any application-specific task as petroglyphs comparison [25]. At the same time, the mining of huge datasets implies the use of large computer clusters [26,27]. The proposed approach based on the use of probabilistic processing shows large improvements in terms of hardware resources when compared with conventional solutions.
…
Sorry they omitted topic maps but what is a merging criteria if it isn’t a type of “similarity?”
From the conclusion:
…
This implementation uses less hardware resources than conventional digital methodologies (based on binary and not probabilistic logic) and is able to process the order of 13GBytes of information per second (in contrast to the estimated 2GBytes/s of speed that could be achieved by the conventional implementation using the same hardware area). With the 12-dimensional space used to allocate each vector in the example shown in this paper we obtain the order of 1 billion of comparisons per second. A patent application has been done for this new mining methodology [32].
…
The patent was filed in Spanish but English and French auto-translations are available.
Hopefully the patent will be used in such a way as to promote widespread implementation of this technique.
I could stand 1 billion comparisons a second, quite easily. Interactive development of merging algorithms anyone?
I first saw this in a tweet by Stefano Bertolo.