Complex energies of the coherent longitudinal optical phonon-plasmon coupled mode according to dynamic mode decomposition analysis.

In a dissipative quantum system, we report the dynamic mode decomposition (DMD) analysis of damped oscillation signals. We used a reflection-type pump-probe method to observe time-domain signals, including the coupled modes of long-lived longitudinal optical phonons and quickly damped plasmons (LOPC) at various pump powers. The Fourier transformed spectra of the observed damped oscillation signals show broad and asymmetric modes, making it difficult to evaluate their frequencies and damping rates. We then used DMD to analyze the damped oscillation signals by precisely determining their frequencies and damping rates. We successfully identified the LOPC modes. The obtained frequencies and damping rates were shown to depend on the pump power, which implies photoexcited carrier density. We compared the pump-power dependence of the frequencies and damping rates of the LOPC modes with the carrier density dependence of the complex eigen-energies of the coupled modes by using the non-Hermitian phenomenological effective Hamiltonian. Good agreement was obtained between the observed and calculated dependences, demonstrating that DMD is an effective alternative to Fourier analysis which often fails to estimate effective damping rates.

Yield-prediction models for efficient exfoliation of soft layered materials into nanosheets.

Yield-prediction models were studied for efficient exfoliation of soft layered materials stacked via van der Waals interactions with the assistance of machine learning on small experimental data. High-yield exfoliation of graphite and layered organic polymer was achieved under the conditions guided by the models in a limited number of experiments.

Lateral-size control of exfoliated transition-metal-oxide 2D materials by machine learning on small data.

A wide variety of nanosheets including monolayers and few-layers have attracted much interest as two-dimensional (2D) materials for the unique anisotropic structures and properties. On the other hand, one of the significant remaining and challenging issues is the lateral-size control. Since 2D materials are generally synthesized by the exfoliation of layered materials, the lateral size is not easily controlled in the breaking-down processes. The experimental factors have not been found for the control and prediction. In the present work, lateral sizes of the exfoliated transition-metal-oxide nanosheets were predicted and controlled by the assistance of machine learning. Layered composites of host inorganic layers and guest organic molecules were exfoliated into nanosheets in organic dispersion media. The lateral size of the nanosheets was estimated by dynamic light scattering (DLS), instead of microscopy methods, to achieve high-throughput analyses. Factors governing the lateral size are explored on the small experimental data by the assistance of sparse modeling, a method of machine learning. The eight physicochemical parameters of the organic guests and dispersion media were extracted by sparse modeling for the construction of the size-prediction model. The size-prediction model accelerated the selective syntheses of the nanosheets with large and small lateral sizes in a limited number of experiments. The results indicate that the prediction model is a guideline to find suitable exfoliation conditions for size control. Size-selective syntheses of a variety of 2D materials can be achieved by similar methods using sparse modeling on small experimental data. Moreover, sparse modeling is applicable to control the design and exploration of other materials and their processing based on small data.

Normal mode analysis of a relaxation process with Bayesian inference.

Measurements of relaxation processes are essential in many fields, including nonlinear optics. Relaxation processes provide many insights into atomic/molecular structures and the kinetics and mechanisms of chemical reactions. For the analysis of these processes, the extraction of modes that are specific to the phenomenon of interest (normal modes) is unavoidable. In this study we propose a framework to systematically extract normal modes from the viewpoint of model selection with Bayesian inference. Our approach consists of a well-known method called sparsity-promoting dynamic mode decomposition, which decomposes a mixture of damped oscillations, and the Bayesian model selection framework. We numerically verify the performance of our proposed method by using coherent phonon signals of a bismuth polycrystal and virtual data as typical examples of relaxation processes. Our method succeeds in extracting the normal modes even from experimental data with strong backgrounds. Moreover, the selected set of modes is robust to observation noise, and our method can estimate the level of observation noise. From these observations, our method is applicable to normal mode analysis, especially for data with strong backgrounds.

Machine learning prediction of coordination energies for alkali group elements in battery electrolyte solvents.

We combined a data science-driven method with quantum chemistry calculations, and applied it to the battery electrolyte problem. We performed quantum chemistry calculations on the coordination energy (Ecoord) of five alkali metal ions (Li, Na, K, Rb, and Cs) to electrolyte solvent, which is intimately related to ion transfer at the electrolyte/electrode interface. Three regression methods, namely, multiple linear regression (MLR), least absolute shrinkage and selection operator (LASSO), and exhaustive search with linear regression (ES-LiR), were employed to find the relationship between Ecoord and descriptors. Descriptors include both ion and solvent properties, such as the radius of metal ions or the atomic charge of solvent molecules. Our results clearly indicate that the ionic radius and atomic charge of the oxygen atom that is connected to the metal ion are the most important descriptors. Good prediction accuracy for Ecoord of 0.127 eV was obtained using ES-LiR, meaning that we can predict Ecoord for any alkali ion without performing quantum chemistry calculations for ion-solvent pairs. Further improvement in the prediction accuracy was made by applying the exhaustive search with Gaussian process, which yields 0.016 eV for the prediction accuracy of Ecoord.

A longitudinal study of infant view-invariant face processing during the first 3-8 months of life.

View-invariant face processing emerges early in life. A previous study (Nakato et al., 2009) measured infant hemodynamic responses to faces from the frontal and profile views in the bilateral temporal areas, which have been reported to be involved in face processing using near-infrared spectroscopy. It was reported that 5-month-old infants showed increased oxyhemoglobin (oxy-Hb) responses to frontal faces, but not to profile faces. In contrast, 8-month-old infants displayed increased oxy-Hb responses to profile faces as well as to frontal faces. In this study, we used the experimental method developed in the previous study to investigate the development of view-invariant face processing, every month for 5 months (from the first 3-8 months of life). We longitudinally measured hemodynamic responses to faces from the frontal and profile views in 14 infants. The longitudinal measurements allowed us to investigate individual differences in each participant. We modeled each infant's hemodynamic oxy-Hb responses to frontal and profile faces using linear regression analysis. Processing of profile faces emerged later and underwent larger improvements than that of frontal faces. We also found an anticorrelation between the speed of improvement in face processing and the hemodynamic response to faces at the age of 3- months. Group analysis of the averaged hemodynamic data from the 14 infants using linear regression revealed that the processing of profile faces emerged between 5 and 6 months of age. Infant view-invariant face processing developed first for frontal faces. This was followed by the emergence of processing of profile faces.

Liquid electrolyte informatics using an exhaustive search with linear regression.

Exploring new liquid electrolyte materials is a fundamental target for developing new high-performance lithium-ion batteries. In contrast to solid materials, disordered liquid solution properties have been less studied by data-driven information techniques. Here, we examined the estimation accuracy and efficiency of three information techniques, multiple linear regression (MLR), least absolute shrinkage and selection operator (LASSO), and exhaustive search with linear regression (ES-LiR), by using coordination energy and melting point as test liquid properties. We then confirmed that ES-LiR gives the most accurate estimation among the techniques. We also found that ES-LiR can provide the relationship between the "prediction accuracy" and "calculation cost" of the properties via a weight diagram of descriptors. This technique makes it possible to choose the balance of the "accuracy" and "cost" when the search of a huge amount of new materials was carried out.

Estimation of the reaction times in tasks of varying difficulty from the phase coherence of the auditory steady-state response using the least absolute shrinkage and selection operator analysis.

Quantitative estimation of the workload in the brain is an important factor for helping to predict the behavior of humans. The reaction time when performing a difficult task is longer than that when performing an easy task. Thus, the reaction time reflects the workload in the brain. In this study, we employed an N-back task in order to regulate the degree of difficulty of the tasks, and then estimated the reaction times from the brain activity. The brain activity that we used to estimate the reaction time was the auditory steady-state response (ASSR) evoked by a 40-Hz click sound. Fifteen healthy participants participated in the present study and magnetoencephalogram (MEG) responses were recorded using a 148-channel magnetometer system. The least absolute shrinkage and selection operator (LASSO), which is a type of sparse modeling, was employed to estimate the reaction times from the ASSR recorded by MEG. The LASSO showed higher estimation accuracy than the least squares method. This result indicates that LASSO overcame the over-fitting to the learning data. Furthermore, the LASSO selected channels in not only the parietal region, but also in the frontal and occipital regions. Since the ASSR is evoked by auditory stimuli, it is usually large in the parietal region. However, since LASSO also selected channels in regions outside the parietal region, this suggests that workload-related neural activity occurs in many brain regions. In the real world, it is more practical to use a wearable electroencephalography device with a limited number of channels than to use MEG. Therefore, determining which brain areas should be measured is essential. The channels selected by the sparse modeling method are informative for determining which brain areas to measure.

Recurrent network for multisensory integration-identification of common sources of audiovisual stimuli.

We perceive our surrounding environment by using different sense organs. However, it is not clear how the brain estimates information from our surroundings from the multisensory stimuli it receives. While Bayesian inference provides a normative account of the computational principle at work in the brain, it does not provide information on how the nervous system actually implements the computation. To provide an insight into how the neural dynamics are related to multisensory integration, we constructed a recurrent network model that can implement computations related to multisensory integration. Our model not only extracts information from noisy neural activity patterns, it also estimates a causal structure; i.e., it can infer whether the different stimuli came from the same source or different sources. We show that our model can reproduce the results of psychophysical experiments on spatial unity and localization bias which indicate that a shift occurs in the perceived position of a stimulus through the effect of another simultaneous stimulus. The experimental data have been reproduced in previous studies using Bayesian models. By comparing the Bayesian model and our neural network model, we investigated how the Bayesian prior is represented in neural circuits.

Theory of correlation in a network with synaptic depression.

Synaptic depression affects not only the mean responses of neurons but also the correlation of response variability in neural populations. Although previous studies have constructed a theory of correlation in a spiking neuron model by using the mean-field theory framework, synaptic depression has not been taken into consideration. We expanded the previous theoretical framework in this study to spiking neuron models with short-term synaptic depression. On the basis of this theory we analytically calculated neural correlations in a ring attractor network with Mexican-hat-type connectivity, which was used as a model of the primary visual cortex. The results revealed that synaptic depression reduces neural correlation, which could be beneficial for sensory coding. Furthermore, our study opens the way for theoretical studies on the effect of interaction change on the linear response function in large stochastic networks.