Enhancing ensemble learning and transfer learning in multimodal data analysis by adaptive dimensionality reduction

Abstract

Modern data analytics take advantage of ensemble learning and transfer learning approaches to tackle some of the most relevant issues in data analysis, such as lack of labeled data to use to train the analysis models, sparsity of the information, and unbalanced distributions of the records. Nonetheless, when applied to multimodal datasets (i.e., datasets acquired by means of multiple sensing techniques or strategies), the state-of-theart methods for ensemble learning and transfer learning might show some limitations. In fact, in multimodal data analysis, not all observations would show the same level of reliability or information quality, nor an homogeneous distribution of errors and uncertainties. This condition might undermine the classic assumptions ensemble learning and transfer learning methods rely on. In this work, we propose an adaptive approach for dimensionality reduction to overcome this issue. By means of a graph theory-based approach, the most relevant features across variable size subsets of the considered datasets are identified. This information is then used to set-up ensemble learning and transfer learning architectures. We test our approach on multimodal datasets acquired in diverse research fields (remote sensing, brain-computer interfaces, photovoltaic energy). Experimental results show the validity and the robustness of our approach, able to outperform state-of-the-art techniques.

A graph representation based on fluid diffusion model for data analysis: theoretical aspects and enhanced community detection

“…In this work, we propose to address this task by exploring the relevance of the features at global (i.e., across all modalities) and local (i.e., across samples for each feature) scale. Following the successful approach proposed in [64], we quantify the multiscale significance of the features by using information theory-based metrics. Specifically, we consider to measure the degree of redundancy and intercorrelation between features across the whole dataset by employing mutual information [65], [66].…”

“…11 by means of state-of-the-art supervised and semisupervised methods based on the classic graph representation introduced in Section II-A [90]- [93]. Moreover, for comparison, we used a method based on ensemble learning approach [64]. It is worth noting that deeply investigating the capacity and limitations of all these methods is out of the scope of this paper.…”

“…In particular, focusing once again on the "apple trees" class, it is possible to compute the average misclassification error between the "apple trees" and "wood" classes obtained over all the experiments we ran: Table IV reports these results. It is therefore possible to appreciate that the algorithms based on the analysis of the graph structure defined according to the guidelines in Section II-A show a substantial increase of the error with respect to the ensemble learning-based architecture in [64]. Hence, the classic graph representation based on heat diffusion model apparently leads to a substantial degradation of the ability of the architectures to characterize and interpret the samples in the dataset.…”

“…At this point, we computed the Q matrix and the weight matrix W according to the classic graph representation based on heat diffusion mechanism. For the Q matrix, we considered the definition as in (13), we used the method in [64] to determine the values of the K tensor, and assumed that the distribution of the B matrix would be uniform. For the classic definition of the weight matrix W based on heat diffusion mechanism, we used a Gaussian kernel to define the function η mentioned in Section II-A [6], [31].…”