Graph Community Infomax

Graph representation learning aims at learning low-dimension representations for nodes in graphs, and has been proven very useful in several downstream tasks. In this article, we propose a new model, Graph Community Infomax (GCI), that can adversarial learn representations for nodes in attributed networks. Different from other adversarial network embedding models, which would assume that the data follow some prior distributions and generate fake examples, GCI utilizes the community information of networks, using nodes as positive(or real) examples and negative(or fake) examples at the same time. An autoencoder is applied to learn the embedding vectors for nodes and reconstruct the adjacency matrix, and a discriminator is used to maximize the mutual information between nodes and communities. Experiments on several real-world and synthetic networks have shown that GCI outperforms various network embedding methods on community detection tasks.

Auto IV: Counterfactual Prediction via Automatic Instrumental Variable Decomposition

Instrumental variables (IVs), sources of treatment randomization that are conditionally independent of the outcome, play an important role in causal inference with unobserved confounders. However, the existing IV-based counterfactual prediction methods need well-predefined IVs, while it’s an art rather than science to find valid IVs in many real-world scenes. Moreover, the predefined hand-made IVs could be weak or erroneous by violating the conditions of valid IVs. These thorny facts hinder the application of the IV-based counterfactual prediction methods. In this article, we propose a novel Automatic Instrumental Variable decomposition (AutoIV) algorithm to automatically generate representations serving the role of IVs from observed variables (IV candidates). Specifically, we let the learned IV representations satisfy the relevance condition with the treatment and exclusion condition with the outcome via mutual information maximization and minimization constraints, respectively. We also learn confounder representations by encouraging them to be relevant to both the treatment and the outcome. The IV and confounder representations compete for the information with their constraints in an adversarial game, which allows us to get valid IV representations for IV-based counterfactual prediction. Extensive experiments demonstrate that our method generates valid IV representations for accurate IV-based counterfactual prediction.

Bilateral Personalized Dialogue Generation with Contrastive Learning

Generating personalized responses is one of the major challenges in natural human-robot interaction. Current researches in this field mainly focus on generating responses consistent with the robot’s pre-assigned persona, while ignoring the user’s persona. Such responses may be inappropriate or even offensive, which may lead to the bad user experience. Therefore, we propose a Bilateral Personalized Dialogue Generation (BPDG) method for dyadic conversation, which integrates user and robot personas into dialogue generation via designing a dynamic persona-aware fusion method. To bridge the gap between the learning objective function and evaluation metrics, the Conditional Mutual Information Maximum (CMIM) criterion is adopted with contrastive learning to select the proper response from the generated candidates. Moreover, a bilateral persona accuracy metric is designed to measure the degree of bilateral personalization. Experimental results demonstrate that, compared with several state-of-the-art methods, the final results of the proposed method are more personalized and consistent with bilateral personas in terms of both automatic and manual evaluations.

The Image Annotation Refinement in Embedding Feature Space based on Mutual Information

The ever-increasing size of images has made automatic image annotation one of the most important tasks in the fields of machine learning and computer vision. Despite continuous efforts in inventing new annotation algorithms and new models, results of the state-of-the-art image annotation methods are often unsatisfactory. In this paper, to further improve annotation refinement performance, a novel approach based on weighted mutual information to automatically refine the original annotations of images is proposed. Unlike the traditional refinement model using only visual feature, the proposed model use semantic embedding to properly map labels and visual features to a meaningful semantic space. To accurately measure the relevance between the particular image and its original annotations, the proposed model utilize all available information including image-to-image, label-to-label and image-to-label. Experimental results conducted on three typical datasets show not only the validity of the refinement, but also the superiority of the proposed algorithm over existing ones. The improvement largely benefits from our proposed mutual information method and utilizing all available information.

Correcting B0 inhomogeneity-induced distortions in whole-body diffusion MRI of bone

Diffusion-weighted magnetic resonance imaging (DWI) of the musculoskeletal system has various applications, including visualization of bone tumors. However, DWI acquired with echo-planar imaging is susceptible to distortions due to static magnetic field inhomogeneities. This study aimed to estimate spatial displacements of bone and to examine whether distortion corrected DWI images more accurately reflect underlying anatomy. Whole-body MRI data from 127 prostate cancer patients were analyzed. The reverse polarity gradient (RPG) technique was applied to DWI data to estimate voxel-level distortions and to produce a distortion corrected DWI dataset. First, an anatomic landmark analysis was conducted, in which corresponding vertebral landmarks on DWI and anatomic T2-weighted images were annotated. Changes in distance between DWI- and T2-defined landmarks (i.e., changes in error) after distortion correction were calculated. In secondary analyses, distortion estimates from RPG were used to assess spatial displacements of bone metastases. Lastly, changes in mutual information between DWI and T2-weighted images of bone metastases after distortion correction were calculated. Distortion correction reduced anatomic error of vertebral DWI up to 29 mm. Error reductions were consistent across subjects (Wilcoxon signed-rank p < 10–20). On average (± SD), participants’ largest error reduction was 11.8 mm (± 3.6). Mean (95% CI) displacement of bone lesions was 6.0 mm (95% CI 5.0–7.2); maximum displacement was 17.1 mm. Corrected diffusion images were more similar to structural MRI, as evidenced by consistent increases in mutual information (Wilcoxon signed-rank p < 10–12). These findings support the use of distortion correction techniques to improve localization of bone on DWI.

Prosodic Feature-Based Discriminatively Trained Low Resource Speech Recognition System

Speech recognition has been an active field of research in the last few decades since it facilitates better human–computer interaction. Native language automatic speech recognition (ASR) systems are still underdeveloped. Punjabi ASR systems are in their infancy stage because most research has been conducted only on adult speech systems; however, less work has been performed on Punjabi children’s ASR systems. This research aimed to build a prosodic feature-based automatic children speech recognition system using discriminative modeling techniques. The corpus of Punjabi children’s speech has various runtime challenges, such as acoustic variations with varying speakers’ ages. Efforts were made to implement out-domain data augmentation to overcome such issues using Tacotron-based text to a speech synthesizer. The prosodic features were extracted from Punjabi children’s speech corpus, then particular prosodic features were coupled with Mel Frequency Cepstral Coefficient (MFCC) features before being submitted to an ASR framework. The system modeling process investigated various approaches, which included Maximum Mutual Information (MMI), Boosted Maximum Mutual Information (bMMI), and feature-based Maximum Mutual Information (fMMI). The out-domain data augmentation was performed to enhance the corpus. After that, prosodic features were also extracted from the extended corpus, and experiments were conducted on both individual and integrated prosodic-based acoustic features. It was observed that the fMMI technique exhibited 20% to 25% relative improvement in word error rate compared with MMI and bMMI techniques. Further, it was enhanced using an augmented dataset and hybrid front-end features (MFCC + POV + Fo + Voice quality) with a relative improvement of 13% compared with the earlier baseline system.

Hydrodynamics of quantum entropies in Ising chains with linear dissipation

We study the dynamics of quantum information and of quantum correlations after a quantum quench, in transverse field Ising chains subject to generic linear dissipation. As we show, in the hydrodynamic limit of long times, large system sizes, and weak dissipation, entropy-related quantities —such as the von Neumann entropy, the Rényi entropies, and the associated mutual information— admit a simple description within the so-called quasiparticle picture. Specifically, we analytically derive a hydrodynamic formula, recently conjectured for generic noninteracting systems, which allows us to demonstrate a universal feature of the dynamics of correlations in such dissipative noninteracting system. For any possible dissipation, the mutual information grows up to a time scale that is proportional to the inverse dissipation rate, and then decreases, always vanishing in the long time limit. In passing, we provide analytic formulas describing the time-dependence of arbitrary functions of the fermionic covariance matrix, in the hydrodynamic limit.

Calculation of centralities in protein kinase A

Topological analysis of amino acid networks is a common method that can help to understand the roles of individual residues. The most popular approach for network construction is to create a connection between residues if they interact. These interactions are usually weighted by absolute values of correlation coefficients or mutual information. Here we argue that connections in such networks have to reflect levels of cohesion within the protein instead of a simple fact of interaction between residues. If this is correct, an indiscriminate combination of correlation and anti-correlation, as well as the all-inclusive nature of the mutual information metrics, should be detrimental for the analysis. To test our hypothesis, we studied amino acid networks of the protein kinase A created by Local Spatial Pattern alignment, a method that can detect conserved patterns formed by Cα-Cβ vectors. Our results showed that, in comparison with the traditional methods, this approach is more efficient in detecting functionally important residues. Out of four studied centrality metrics, Closeness centrality was the least efficient measure of residue importance. Eigenvector centrality proved to be ineffective as the spectral gap values of the networks were very low due to the bilobal structure of the kinase. We recommend using joint graphs of Betweenness centrality and Degree centrality to visualize different aspects of amino acid roles.

Atrial fibrillation driver identification through regional mutual information networks: a modeling perspective

Purpose Effective identification of electrical drivers within remodeled tissue is a key for improving ablation treatment for atrial fibrillation. We have developed a mutual information, graph-based approach to identify and propose fault tolerance metric of local efficiency as a distinguishing feature of rotational activation and remodeled atrial tissue. Methods Voltage data were extracted from atrial tissue simulations (2D Karma, 3D physiological, and the Multiscale Cardiac Simulation Framework (MSCSF)) using multi-spline open and parallel regional mapping catheter geometries. Graphs were generated based on varied mutual information thresholds between electrode pairs and the local efficiency for each graph was calculated. Results High-resolution mapping catheter geometries can distinguish between rotational and irregular activation patterns using the derivative of local efficiency as a function of increasing mutual information threshold. The derivative is decreased for rotational activation patterns comparing to irregular activations in both a simplified 2D model (0.0017 ± 1 × 10−4 vs. 0.0032 ± 1 × 10−4, p < 0.01) and a more realistic 3D model (0.00092 ± 5 × 10−5 vs. 0.0014 ± 4 × 10−5, p < 0.01). Average local efficiency derivative can also distinguish between degrees of remodeling. Simulations using the MSCSF model, with 10 vs. 90% remodeling, display distinct derivatives in the grid design parallel spline catheter configuration (0.0015 ± 5 × 10−5 vs. 0.0019 ± 6 × 10−5, p < 0.01) and the flower shaped open spline configuration (0.0011 ± 5 × 10−5 vs. 0.0016 ± 4 × 10−5, p < 0.01). Conclusion A decreased derivative of local efficiency characterizes rotational activation and varies with atrial remodeling. This suggests a distinct communication pattern in cardiac rotational activation detectable via high-resolution regional mapping and could enable identification of electrical drivers for targeted ablation.