Articles written in Sadhana

    • Deep Gaussian processes for music mood estimation and retrieval with locally aggregated acoustic Fisher vector


      More Details Abstract Fulltext PDF

      Due to the subjective nature of music mood, it is challenging to computationally model the affective content of the music. In this work, we propose novel features known as locally aggregated acoustic Fisher vectors based on the Fisher kernel paradigm. To preserve the temporal context, onset-detected variable-lengthsegments of the audio songs are obtained, for which a variational Bayesian approach is used to learn the universal background Gaussian mixture model (GMM) representation of the standard acoustic features. The local Fisher vectors obtained with the soft assignment of GMM are aggregated to obtain a better performance relative to the global Fisher vector. A deep Gaussian process (DGP) regression model inspired by the deep learning architectures is proposed to learn the mapping between the proposed Fisher vector features and the mood dimensions of valence and arousal. Since the exact inference on DGP is intractable, the pseudo-data approximation is used to reduce the training complexity and the Monte Carlo sampling technique is used to solve the intractability problem during training. A detailed derivation of a 3-layer DGP is presented that can be easily generalized to an L-layer DGP. The proposed work is evaluated on the PMEmo dataset containing valence and arousal annotations of Western popular music and achieves an improvement in R² of 25% for arousal and 52% for valence for music mood estimation and an improvement in the Gamma statistic of 68% for music mood retrieval relative to the baseline single-layer Gaussian process.

    • Enhancing the aggregate diversity with mutual trust computations for context-aware recommendations


      More Details Abstract Fulltext PDF

      Context-aware Recommender Systems (CARS) deal with modeling and prediction of user interests and preferences according to contextual information while generating a recommendation. In contextual modeling- based CARS, the context information is used straight into the recommendation function as a predictor explicitly. Thus, this approach formulates a multidimensional recommendation model and is best realized through Tensor Factorization (TF) based techniques. It efficiently handles the data sparsity problem faced bymost of the traditional RS. However, the recent TF-based CARS face issues such as differentiating amongst relevant and irrelevant context variables, biased recommendations, and long-tail problem. In this paper, we propose a fusion-based approach for determining the list of most relevant and optimum contexts for two datasets, namely the LDos Comoda and Travel dataset. The mutual trust model that combines user level and item level trust is proposed further which utilizes the concept of trust propagation to calculate the inferred trust betweenusers/items. Finally, a hybrid reranking technique combining the item popularity and item absolute likeability reranking approaches with the standard ranking technique of generating recommendations is proposed to generate diversified recommendations. Comparative experiments on the LDos Comoda and the Travel datasets are conducted and the experimental results show an improvement of the proposed work with respect to RMSE of 50%, 55%, and 59% compared to MF-based RS, trust-based RS, and context-aware RS respectively. Also, the proposed reranking technique shows approximately three times more diversified recommendations than the standard ranking approach without a significant loss in precision.

    • Feature sensitive geometrically faithful highly regular direct triangular isotropic surface remeshing


      More Details Abstract Fulltext PDF

      Achieving a geometrically faithful, feature preserved, highly regular remesh with lower mesh complexity and high element quality is an ill-posed problem of surface remeshing in research. Individual surface remeshing techniques differ based on their end goals while ignoring the other enhancements in the remesh. In this research work, we present a surface remeshing framework that aim to balance the various crucial remeshing goals to enhance the remesh quality. In surface remeshing, the mesh quality comprises—(i) mesh complexity, (ii) mesh element quality, (iii) vertex regularity, (iv) geometric fidelity, and (v) feature preservation. Our remeshing approach uses the local edge operators to achieve mesh decimation, enhance element quality, and regularize the vertex valence of the remesh while conserving the features in the remesh. During mesh decimation, we preserve features using an updated quadric error metric. The mesh element quality is enhanced bysplitting maximal angles and uplifting the minimal angles using edge split and edge collapse respectively. We maintain dynamic priority queues to maintain the maximal and minimal angles that require attention and improve them using local edge operators. High vertex regularity is achieved by valence optimization. Thegeometric faithfulness of the remesh with the original input mesh is maintained by constraining the bounds on the approximation error computed by the two-sided Hausdorff distance. In succession with other local edge operators, our algorithm can remesh low-quality mesh surfaces efficiently. The remeshes generated using our remeshing framework were compared with recent remeshing approaches using various performance metrics.

  • Sadhana | News

    • Editorial Note on Continuous Article Publication

      Posted on July 25, 2019

      Click here for Editorial Note on CAP Mode

© 2022-2023 Indian Academy of Sciences, Bengaluru.