Search

Link prediction, the computational problem of determining whether there is a link between two objects, is important in machine learning and data mining. Feature-based link prediction, in which the feature vectors of the two objects are given, is of particular interest because it can also be used for various identification-related problems. Although the factorization machine and the higher-order factorization machine (HOFM) are widely used for feature-based link prediction, they use feature combinations not only across the two objects but also from the same object. Feature combinations from the same object are irrelevant to major link prediction problems such as predicting identity because using them increases computational cost and degrades accuracy. In this paper, we present novel models that use higher-order feature combinations only across the two objects. Since there were no algorithms for efficiently computing higher-order feature combinations only across two objects, we derive one by leveraging reported and newly obtained results of calculating the ANOVA kernel. We present an efficient coordinate descent algorithm for proposed models. We also improve the effectiveness of the existing one for the HOFM. Furthermore, we extend proposed models to a deep neural network. Experimental results demonstrated the effectiveness of our proposed models.

Gradient descent is a widely used optimization method. The adjustment of the learning rate is an important factor in improving its performance, and many researchers have investigated methods for automatically adjusting the learning rate. One such method, hypergradient descent, automatically adjusts the learning rate by using gradient descent. However, it introduces the “learning rate of the learning rate,” and an appropriate value for the learning rate of the learning rate must be chosen in order to effectively adjust the learning rate. We investigated the use of two datasets and two optimization methods for doing this and achieved an effective adjustment of the learning rate when the objective function was convex and L -smooth.

Deep generative models are being actively studied, particularly variational autoencoders (VAEs) because they can generate high-quality images. The M2 model supports semi-supervised learning from both labeled and unlabeled data, which enables the generated images to be easily controlled by changing the class label values. However, generative models must be learned from only unlabeled data when class labels are not available. A model is presented that incorporates a deep clustering method into the M2 model, which enables clusters to be identified among unlabeled data so that each data point can be assigned to one of the clusters. The generated images in unlabeled datasets can easily be controlled by changing the cluster assignment of each data point.

Open data are becoming increasingly available in various domains, and many organizations rely on making decisions according to data. Such decision making requires care to distinguish between correlations and causal relationships. Among data analysis tasks, causal relationship analysis is especially complex because of unobserved confounders. For example, to correctly analyze the causal relationship between two variables, the possible confounding effect of a third variable should be considered. In the open-data environment, however, it is difficult to consider all possible confounders in advance. In this paper, we propose a framework for exploratory causal analysis of open data, in which possible confounding variables are collected and incrementally tested from a large volume of open data. To the extent of the authors’ knowledge, no framework has been proposed to incorporate data for possible confounders in causal analysis process. This paper shows an original way to expand causal structures and generate reasonable causal relationships. The proposed framework accounts for the effect of possible confounding in causal analysis by first using a crowdsourcing platform to collect explanations of the correlation between variables. Keywords are then extracted using natural language processing methods. The framework searches the related open data according to the extracted keywords. Finally, the collected explanations are tested using several automated causal analysis methods. We conducted experiments using open data from the World Bank and the Japanese government. The experimental results confirmed that the proposed framework enables causal analysis while considering the effects of possible confounders.

The effect of confounding factors cannot be ignored in real world causal discovery tasks. A common cause is a general confounder between two variables. In this paper, we propose using intrinsic dimension estimation as a necessary condition to determine a possible common cause for two variables. Simulated application showed that the proposed method worked well for both linear and non-linear functions. Testing using different types of noise showed that it generally worked well for different types of added noise. In particular, it worked better than a kernel-based conditional independence test for Poisson noise. Testing of how the estimated intrinsic dimension is affected by different types of distributions showed that the estimated dimension is nearly not affected by the type of distribution. Simulation of mixed pattern showed that the proposed method can still tell a possible common cause when it is mixed with causal relationship. Finally, experiments using variables from the CauseEffectPairs dataset showed that the proposed method can give correct inferred results for real world data., 2019 IEEE International Conference will be held 27 Feb.-2 March 2019 at Kyoto, Japan

Molecular graphs are one of the established representations for small molecules, and even steric or electronic information can be encoded as node and edge features. Naturally, graph neural networks have been intensively investigated to solve various chemical problems at molecular levels. However, it remains unclear how to encode relevant chemical information into graphs. We investigate this problem by proposing three models of graph neural networks with self-attention mechanisms at different levels to adaptively select relevant chemical information for each input. Using neural graph fingerprint (NFP) as a baseline, we introduce three types of attention mechanisms on the top of NFPs. Our experimental evaluations suggest that introducing these self-attention mechanisms contributes to not only improving the prediction accuracy but also providing quantitative interpretation using obtained attention coefficients., 2019 IEEE International Conference will be held 9-12 Dec. 2019 at Los Angeles, CA, USA