Structured Prediction of Music Mood with Twin Gaussian Processes

Chapaneri, Santosh; Jayaswal, Deepak

doi:10.1007/978-3-319-69900-4_82

Cited by 6 publications

(3 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Since not all annotators annotated every audio clip and they may not be equally reliable, truth discovery analysis [34] of this crowdsourced data is required to determine the estimated consensus. Instead of taking the average VA value for each audio clip from the crowdsourced annotated data, the EM algorithm proposed in [35] is used to consider the reliability of each annotator and compute the estimated consensus ground truth y ∈ R 2 . This estimated consensus algorithm also tackles the long-tail phenomenon commonly observed in crowdsourced data by using the upper bound of the confidence interval of χ 2 distribution.…”

Section: B Covariate Shift In Music Mood Estimationmentioning

confidence: 99%

“…For each audio clip, standard acoustic features are extracted using MIRToolbox [36] across four categories (dynamics, spectral, timbral and tonal) resulting in a 70-dimensional feature vector per frame of 50 ms duration with 50% overlap. For an effective prototypical feature representation, the variational Bayesian inference algorithm is used to compute the Bayesian Acoustic Gaussian Mixture Model (BAGMM) posterior probability feature vector x ∈ R K opt for each audio clip, where K opt = 117 [35]. The advantage of Bayesian inference is that the number of latent audio topics K can be determined from the data automatically, thus avoiding the problems of singularity and over/underfitting with ad-hoc values of K [9].…”

Section: B Covariate Shift In Music Mood Estimationmentioning

confidence: 99%

See 1 more Smart Citation

Covariate Shift Adaptation for Structured Regression With Frank–Wolfe Algorithms

Chapaneri

Jayaswal

2019

IEEE Access

Self Cite

View full text Add to dashboard Cite

This paper concerns structured regression problems wherein the issue of covariate shift is addressed, which aims at reducing the discrepancy in training and test data distributions, using computationally efficient and sparse optimization principles. In particular, the projection-free Frank-Wolfe optimization algorithms are used to learn the importance weights and re-weight the training data in the context of covariate shift. To determine the unbiased estimates of the weights, Kullback-Leibler importance estimation procedure is used but its computational cost can be high since it is based on projected gradient optimization. Instead of using the standard Frank-Wolfe algorithm, we adapt its variants and propose away-steps Frank-Wolfe and pairwise Frank-Wolfe covariate shift algorithms to correct the covariate shift. The results highlight the improved computational efficiency and sparsity achieved while learning the importance weights on synthetic as well as benchmark datasets. Furthermore, importance weighted Sharma-Mittal twin Gaussian process structured regression framework is proposed to incorporate the learned weights from covariate shift algorithms, and its equations are derived for importance weighted derivatives and uncertainties. The performance of proposed algorithms is evaluated on two applications of structured regression, namely, human pose estimation and music mood estimation, where the benefit of handling covariate shift is demonstrated with improved performance relative to the state-of-the-art techniques.

show abstract

Section: B Covariate Shift In Music Mood Estimationmentioning

confidence: 99%

Section: B Covariate Shift In Music Mood Estimationmentioning

confidence: 99%

Covariate Shift Adaptation for Structured Regression With Frank–Wolfe Algorithms

Chapaneri

Jayaswal

2019

IEEE Access

Self Cite

View full text Add to dashboard Cite

show abstract

“…As for speech, we found publications related to the prediction of speaker likability (Hantke et al, ) and the evaluation of dysarthric speech (Tu et al, ). In the case of music, we found applications related to chord estimation (Ni et al, ) and music mood prediction (Chapaneri & Jayaswal, ). Lastly, we found applications dealing with emotion recognition from utterances (Hantke et al, ) and acoustic classification of animal species (Zhang et al, ).…”

Section: Publication Areasmentioning

confidence: 99%

Machine learning from crowds: A systematic review of its applications

Rodrigo

Aledo

Gámez

2018

WIREs Data Mining Knowl Discov

View full text Add to dashboard Cite

Crowdsourcing opens the door to solving a wide variety of problems that previously were unfeasible in the field of machine learning, allowing us to obtain relatively low cost labeled data in a small amount of time. However, due to the uncertain quality of labelers, the data to deal with are sometimes unreliable, forcing practitioners to collect information redundantly, which poses new challenges in the field. Despite these difficulties, many applications of machine learning using crowdsourced data have recently been published that achieved state of the art results in relevant problems. We have analyzed these applications following a systematic methodology, classifying them into different fields of study, highlighting several of their characteristics and showing the recent interest in the use of crowdsourcing for machine learning. We also identify several exciting research lines based on the problems that remain unsolved to foster future research in this field.

show abstract

Structured Prediction of Music Mood with Twin Gaussian Processes

Cited by 6 publications

References 11 publications

Covariate Shift Adaptation for Structured Regression With Frank–Wolfe Algorithms

Covariate Shift Adaptation for Structured Regression With Frank–Wolfe Algorithms

Machine learning from crowds: A systematic review of its applications

Contact Info

Product

Resources

About