Robust loss functions for boosting

Boosting is known as a gradient descent algorithm over loss functions. It is often pointed out that the typical boosting algorithm, Adaboost, is highly affected by outliers. In this letter, loss functions for robust boosting are studied. Based on the concept of robust statistics, we propose a transformation of loss functions that makes boosting algorithms robust against extreme outliers. Next, the truncation of loss functions is applied to contamination models that describe the occurrence of mislabels near decision boundaries. Numerical experiments illustrate that the proposed loss functions derived from the contamination models are useful for handling highly noisy data in comparison with other loss functions.     

Imitating others by composition of primitive actions: A neuro-dynamic model

This paper introduces a novel neuro-dynamical model that accounts for possible mechanisms of action imitation and learning. It is considered that imitation learning requires at least two classes of generalization. One is generalization over sensory–motor trajectory variances, and the other class is on cognitive level which concerns on more qualitative understanding of compositional actions by own and others which do not necessarily depend on exact trajectories. This paper describes a possible model dealing with these classes of generalization by focusing on the problem of action compositionality. The model was evaluated in the experiments using a small humanoid robot. The robot was trained with a set of different actions concerning object manipulations which can be decomposed into sequences of action primitives. Then the robot was asked to imitate a novel compositional action demonstrated by a human subject which are composed from prior-learned action primitives. The results showed that the novel action can be successfully imitated by decomposing and composing it with the primitives by means of organizing unified intentional representation hosted by mirror neurons even though the trajectory-level appearance is different between the ones of observed and those of self-generated.     

Precise Speed Estimation From a Low-Resolution Encoder by Dual-Sampling-Rate Observer

This paper describes an effective way to estimate state variables, such as motor speed and disturbance from a low-resolution encoder at low speed by using the dual-sampling-rate observer. The dual-sampling-rate observer estimates the state variables at every DSP control period and correct the estimation error at the instant that the measurement signal is detected. A novel pole assignment method, which considers the relation of the estimation and error correction periods, is proposed to maintain the stability for long error correcting period. Moreover, the dual-sampling-rate observer can be applied for higher order systems since it is generalized in state space. The effectiveness of the observer is verified through various simulations and experiments     

Evaluation of R-curve behavior of ceramic-metal functionally graded materials by stable crack growth

This paper deals with the fracture toughness and R-curve behavior of ceramic-metal functionally graded materials (FGMs). A possibility of stable crack growth in a three-point-bending specimen is examined based on the driving force and resistance for crack growth in FGMs, and the distribution of fracture toughness or R-curve behavior is evaluated on FGMs fabricated by powder metallurgy using partially stabilized zirconia (PSZ) and stainless steel (SUS 304). The materials have a functionally graded surface layer (FGM layer) with a thickness of 1 mm or 2 mm on a SUS 304 substrate. Three-point-bending tests are carried out on a rectangular specimen with a very short crack in the ceramics surface. On the three-point-bending test, a crack is initiated from a short pre-crack in unstable manner, and then it propagates in stable manner through the FGM layer with an increase in the applied load. From the relationship between applied load and crack length during the stable crack growth in the FGM layer, the fracture toughness is evaluated. The fracture toughness increases with an increase in a volume fraction of SUS 304 phase.     

Magnetorefractive Effect for Co/Ru Multilayer Film with Antiferromagnetic Interlayer Exchange Coupling

The spin‐dependent scattering of conduction electron in metallic magnetic multilayers was investigated to reveal the optical properties which corresponded to the magnetization state. From the measurement of magnetorefractive effect in antiferro‐magnetically exchange coupled Co/Ru multilayers, it was found that (1) the transmittance varied around 1% at wavelength of 1550 nm during the parallel/antiparallel spin alignment changes, (2) as a result of number of repetition and Co thickness for the multilayers, transmittance difference between the parallel and antiparallel spin alignment states had trade‐off relationship with the magnitude of transmittance, and (3) asymmetric Co/Ru/Co stacking structure brought different interlayer exchange coupling fields in each layer, which resulted in magnetic field sensitive transmittance.     

Impact of Serum Chemerin Levels on Liver Functional Reserves and Platelet Counts in Patients with Hepatocellular Carcinoma

Obesity-related metabolic abnormalities, including adipokine imbalance and chronic inflammation, are involved in liver carcinogenesis. Chemerin, a novel adipokine, plays a critical role in adipogenesis, energy metabolism, and inflammation. We evaluated the impact of serum chemerin levels on liver functional reserves in hepatocellular carcinoma (HCC) patients and on the recurrence and prognosis of HCC. This study included 44 patients with any stage of HCC who underwent curative treatment at Gifu Municipal Hospital (Gifu, Japan) between 2006 and 2007. Recurrence-free survival and overall survival were estimated using the Kaplan-Meier method. Serum albumin levels (Pearson’s correlation coefficient; r = 0.3110, p = 0.0399), platelet counts (r = 0.4159, p = 0.0050), and prothrombin times (r = 0.3775, p = 0.0115) were significantly correlated with serum chemerin levels in patients with HCC, and they were inversely correlated with Child-Pugh scores (r = −0.3732, p = 0.0126), serum alanine aminotransferase levels (r = −0.3864, p = 0.0105), and total bilirubin levels (r = −0.4023, p = 0.0068). Among these variables, a multiple comparison test identified that platelet counts and total bilirubin levels were associated with serum chemerin levels (p < 0.0083). No significant correlation was found between serum chemerin levels and recurrence-free survival (p = 0.3691) or overall survival (p = 0.7916). In HCC patients, serum chemerin concentrations were correlated with liver functional reserves and platelet counts, but not with recurrence or prognosis.     

Enhanced catalytic activity of nanoshell carbon co-doped with boron and nitrogen in the oxygen reduction reaction

The use of carbon cathode catalysts in polymer electrolyte fuel cells instead of the current platinum catalysts is attracting increasing attention. We claim that two factors are important for enhancing the activity of carbon cathode catalysts in the oxygen reduction reaction (ORR): the formation of a nanoshell structure and co-doping with boron and nitrogen. Herein, we investigate the preparation and characterization of active ORR carbon catalysts that combine the above factors. Boron and nitrogen (BN)-doped nanoshell-containing carbon (BN-NSCC) was prepared by carbonizing a mixture of poly(furfuryl alcohol), cobalt phthalocyanine, melamine, and a trifluoroborane–methanol complex at 1000 °C. Transmission electron microscopy and X-ray photoelectron spectroscopy revealed the formation of nanoshell structures with distorted graphitic layers and the introduction of boron and nitrogen atoms, respectively. The ORR activity was evaluated in oxygen-saturated 0.5 mol dm^?3 H₂SO₄ using Koutecky–Levich analysis. The BN-NSCC showed an eight to ten times higher ORR activity than undoped NSCC, with an increased number of electrons participating in the reaction. Tafel analysis revealed a change in the rate-determining step caused by BN-doping. Thus, the combination of a nanoshell structure and co-doping with boron and nitrogen was found to improve the ORR activity of carbon catalysts.