A noise robust method for pattern discovery in quantized time series: the concept matrix approach

Räsänen, Okko; Laine, Unto K.; Altosaar, Toomas

doi:10.21437/interspeech.2009-562

Cited by 8 publications

(3 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In ACORNS we are experimenting with different struc ture discovery approaches: Non-negative M atrix Factorization (NMF) [2] [8], Concept Matrices (CM) [5] and DP-Ngrams [9]. All approaches are incremental and are able to discover recur rent structure in speech signals and to associate audio and visual information.…”

Section: Comparison Of Three Learning Methodsmentioning

confidence: 99%

“…The Concept M atrix (CM) approach [5] is a statistical method for weakly supervised pattern discovery from time-series input.…”

Section: CMmentioning

confidence: 99%

“…Since the algorithm does not make a M arkov assumption about the independence o f subsequent states, but rather inte grates information along the temporal dimension, it achieves high robustness against noise and variation in the input. For each concept, a separate co-occurrence matrix is created at each lag, and these concept-specific matrices are updated only in the presence o f the corresponding tag in the visual input [5]. W hen recognising novel input, activation values o f transitions occurring in the input at different lags are retrieved from co occurrence matrices and added together for each frame, leading to a temporal activation curve for each learned concept.…”

Section: CMmentioning

confidence: 99%

See 2 more Smart Citations

Do multiple caregivers speed up language acquisition?

Bosch¹,

Räsänen²,

Driesen³

et al. 2009

Interspeech 2009

Self Cite

View full text Add to dashboard Cite

Section: Comparison Of Three Learning Methodsmentioning

confidence: 99%

“…The Concept M atrix (CM) approach [5] is a statistical method for weakly supervised pattern discovery from time-series input.…”

Section: CMmentioning

confidence: 99%

Section: CMmentioning

confidence: 99%

See 1 more Smart Citation

Do multiple caregivers speed up language acquisition?

Bosch¹,

Räsänen²,

Driesen³

et al. 2009

Interspeech 2009

Self Cite

View full text Add to dashboard Cite

A computational model of word segmentation from continuous speech using transitional probabilities of atomic acoustic events

Räsänen¹

2011

Cognition

View full text Add to dashboard Cite

Signal dimensionality and the emergence of combinatorial structure

2017

View full text Add to dashboard Cite

In language, a small number of meaningless building blocks can be combined into an unlimited set of meaningful utterances. This is known as combinatorial structure. One hypothesis for the initial emergence of combinatorial structure in language is that recombining elements of signals solves the problem of overcrowding in a signal space. Another hypothesis is that iconicity may impede the emergence of combinatorial structure. However, how these two hypotheses relate to each other is not often discussed. In this paper, we explore how signal space dimensionality relates to both overcrowding in the signal space and iconicity. We use an artificial signalling experiment to test whether a signal space and a meaning space having similar topologies will generate an iconic system and whether, when the topologies differ, the emergence of combinatorially structured signals is facilitated. In our experiments, signals are created from participants' hand movements, which are measured using an infrared sensor. We found that participants take advantage of iconic signal-meaning mappings where possible. Further, we use trajectory predictability, measures of variance, and Hidden Markov Models to measure the use of structure within the signals produced and found that when topologies do not match, then there is more evidence of combinatorial structure. The results from these experiments are interpreted in the context of the differences between the emergence of combinatorial structure in different linguistic modalities (speech and sign).

show abstract

A noise robust method for pattern discovery in quantized time series: the concept matrix approach

Cited by 8 publications

References 10 publications

Do multiple caregivers speed up language acquisition?

Do multiple caregivers speed up language acquisition?

A computational model of word segmentation from continuous speech using transitional probabilities of atomic acoustic events

Signal dimensionality and the emergence of combinatorial structure

Contact Info

Product

Resources

About