-
QUBO-inspired Molecular Fingerprint for Chemical Property Prediction
Authors:
Koichiro Yawata,
Yoshihiro Osakabe,
Takuya Okuyama,
Akinori Asahara
Abstract:
Molecular fingerprints are widely used for predicting chemical properties, and selecting appropriate fingerprints is important. We generate new fingerprints based on the assumption that a performance of prediction using a more effective fingerprint is better. We generate effective interaction fingerprints that are the product of multiple base fingerprints. It is difficult to evaluate all combinati…
▽ More
Molecular fingerprints are widely used for predicting chemical properties, and selecting appropriate fingerprints is important. We generate new fingerprints based on the assumption that a performance of prediction using a more effective fingerprint is better. We generate effective interaction fingerprints that are the product of multiple base fingerprints. It is difficult to evaluate all combinations of interaction fingerprints because of computational limitations. Against this problem, we transform a problem of searching more effective interaction fingerprints into a quadratic unconstrained binary optimization problem. In this study, we found effective interaction fingerprints using QM9 dataset.
△ Less
Submitted 17 March, 2023;
originally announced March 2023.
-
QUBO Decision Tree: Annealing Machine Extends Decision Tree Splitting
Authors:
Koichiro Yawata,
Yoshihiro Osakabe,
Takuya Okuyama,
Akinori Asahara
Abstract:
This paper proposes an extension of regression trees by quadratic unconstrained binary optimization (QUBO). Regression trees are very popular prediction models that are trainable with tabular datasets, but their accuracy is insufficient because the decision rules are too simple. The proposed method extends the decision rules in decision trees to multi-dimensional boundaries. Such an extension is g…
▽ More
This paper proposes an extension of regression trees by quadratic unconstrained binary optimization (QUBO). Regression trees are very popular prediction models that are trainable with tabular datasets, but their accuracy is insufficient because the decision rules are too simple. The proposed method extends the decision rules in decision trees to multi-dimensional boundaries. Such an extension is generally unimplementable because of computational limitations, however, the proposed method transforms the training process to QUBO, which enables an annealing machine to solve this problem.
△ Less
Submitted 17 March, 2023;
originally announced March 2023.
-
A copula-based visualization technique for a neural network
Authors:
Yusuke Kubo,
Yuto Komori,
Toyonobu Okuyama,
Hiroshi Tokieda
Abstract:
Interpretability of machine learning is defined as the extent to which humans can comprehend the reason of a decision. However, a neural network is not considered interpretable due to the ambiguity in its decision-making process. Therefore, in this study, we propose a new algorithm that reveals which feature values the trained neural network considers important and which paths are mainly traced in…
▽ More
Interpretability of machine learning is defined as the extent to which humans can comprehend the reason of a decision. However, a neural network is not considered interpretable due to the ambiguity in its decision-making process. Therefore, in this study, we propose a new algorithm that reveals which feature values the trained neural network considers important and which paths are mainly traced in the process of decision-making. In the proposed algorithm, the score estimated by the correlation coefficients between the neural network layers that can be calculated by applying the concept of a pair copula was defined. We compared the estimated score with the feature importance values of Random Forest, which is sometimes regarded as a highly interpretable algorithm, in the experiment and confirmed that the results were consistent with each other. This algorithm suggests an approach for compressing a neural network and its parameter tuning because the algorithm identifies the paths that contribute to the classification or prediction results.
△ Less
Submitted 27 March, 2020;
originally announced March 2020.
-
Transposable character tables, dual groups
Authors:
Ivan Andrus,
Pál Hegedűs,
Tetsuro Okuyama
Abstract:
One way of expressing the self-duality $A\cong \Hom(A,\mathbb{C})$ of Abelian groups is that their character tables are self-transpose (in a suitable ordering). Noncommutative groups fail to satisfy this property. In this paper we extend the duality to some noncommutative groups considering when the character table of a finite group is close to being the transpose of the character table for some o…
▽ More
One way of expressing the self-duality $A\cong \Hom(A,\mathbb{C})$ of Abelian groups is that their character tables are self-transpose (in a suitable ordering). Noncommutative groups fail to satisfy this property. In this paper we extend the duality to some noncommutative groups considering when the character table of a finite group is close to being the transpose of the character table for some other group. We find that groups dual to each other have dual normal subgroup lattices. We show that our concept of duality cannot work for non-nilpotent groups and we describe $p$-group examples.
△ Less
Submitted 1 February, 2013; v1 submitted 27 December, 2012;
originally announced December 2012.
-
Role of the Gelatinous Layer (G-Layer) on the Origin of the Physical Properties of the Tension Wood of Acer sieboldianum
Authors:
Hiroyuki Yamamoto,
Kentaro Abe,
Yoshiharu Arakawa,
Takashi Okuyama,
Joseph Gril
Abstract:
The tension wood (TW) properties of a 70 year-old Acer sieboldianum Miq were analyzed by using the G-fiber model which was proposed in our previous paper. The roles of the G-layer on the origins of (1) a high large tensile growth stress, (2) a large longitudinal Young's modulus, and (3) a high longitudinal drying shrinkage in the tension wood xylem were discussed on the basis of the simulations…
▽ More
The tension wood (TW) properties of a 70 year-old Acer sieboldianum Miq were analyzed by using the G-fiber model which was proposed in our previous paper. The roles of the G-layer on the origins of (1) a high large tensile growth stress, (2) a large longitudinal Young's modulus, and (3) a high longitudinal drying shrinkage in the tension wood xylem were discussed on the basis of the simulations using the G-fiber model. The results suggest that the G-layer generates a high tensile stress in the longitudinal direction during the xylem maturation; the longitudinal Young's modulus of the green G-layer becomes significantly higher than that of the lignified layer; furthermore, the G-layer tends to shrink extraordinarily higher than that of the lignified layer during the moisture desorption.
△ Less
Submitted 26 March, 2005;
originally announced March 2005.
-
Sympatric Speciation in a Simple Food Web
Authors:
K. Luz-Burgoa,
Tony Dell,
Toshinori Okuyama
Abstract:
Observations of the evolution of species groups in nature, such as well recognized Galapagos finches, have motivated much theoretical research aimed at understanding the processes associated with such radiations. The Penna model is one such model and has been widely used to study aging. In this paper we use the basic Penna model to investigate the process of sympatric speciation in a simple food…
▽ More
Observations of the evolution of species groups in nature, such as well recognized Galapagos finches, have motivated much theoretical research aimed at understanding the processes associated with such radiations. The Penna model is one such model and has been widely used to study aging. In this paper we use the basic Penna model to investigate the process of sympatric speciation in a simple food web model. Initially our web consists of a primary food source and a single herbivore species that feeds on this resource. Subsequently we introduce a predator that feeds on the herbivore. In both instances we directly manipulate the food source, that is, its size distribution, and monitore the changes in the populations structures. Sympatric speciation is obtained for the consumer species in both webs, and our results confirm that the speciation velocity depends on how far up,in the food chain, the focus population is feeding. Simulations are done with three different sexual imprinting-like mechanisms, in order to discuss adaptation by natural selection.
△ Less
Submitted 11 February, 2005;
originally announced February 2005.
