Learning domain invariant representations by joint Wasserstein distance minimization.

Domain shifts in the training data are common in practical applications of machine learning; they occur for instance when the data is coming from different sources. Ideally, a ML model should work well independently of these shifts, for example, by learning a domain-invariant representation. However, common ML losses do not give strong guarantees on how consistently the ML model performs for different domains, in particular, whether the model performs well on a domain at the expense of its performance on another domain. In this paper, we build new theoretical foundations for this problem, by contributing a set of mathematical relations between classical losses for supervised ML and the Wasserstein distance in joint space (i.e. representation and output space). We show that classification or regression losses, when combined with a GAN-type discriminator between domains, form an upper-bound to the true Wasserstein distance between domains. This implies a more invariant representation and also more stable prediction performance across domains. Theoretical results are corroborated empirically on several image datasets. Our proposed approach systematically produces the highest minimum classification accuracy across domains, and the most invariant representation.

Deep Clustering With a Constraint for Topological Invariance Based on Symmetric InfoNCE.

We consider the scenario of deep clustering, in which the available prior knowledge is limited. In this scenario, few existing state-of-the-art deep clustering methods can perform well for both noncomplex topology and complex topology data sets. To address the problem, we propose a constraint utilizing symmetric InfoNCE, which helps an objective of the deep clustering method in the scenario of training the model so as to be efficient for not only noncomplex topology but also complex topology data sets. Additionally, we provide several theoretical explanations of the reason that the constraint can enhances the performance of deep clustering methods. To confirm the effectiveness of the proposed constraint, we introduce a deep clustering method named MIST, which is a combination of an existing deep clustering method and our constraint. Our numerical experiments via MIST demonstrate that the constraint is effective. In addition, MIST outperforms other state-of-the-art deep clustering methods for most of the commonly used 10 benchmark data sets.

Operando Spectroscopic Study of the Dynamics of Ru Catalyst during Preferential Oxidation of CO and the Prevention of Ammonia Poisoning by Pt.

Hydrogen is a promising clean energy source. In domestic polymer electrolyte fuel cell systems, hydrogen is produced by reforming of natural gas; however, the reformate contains carbon monoxide (CO) as a major impurity. This CO is removed from the reformate by a combination of the water-gas shift reaction and preferential oxidation of CO (PROX). Currently, Ru-based catalysts are the most common type of PROX catalyst; however, their durability against ammonia (NH3) as an impurity produced in situ from trace amounts of nitrogen also contained in the reformate is an important issue. Previously, we found that addition of Pt to an Ru catalyst inhibited deactivation by NH3. Here, we conducted operando XAFS and FT-IR spectroscopic analyses with simultaneous gas analysis to investigate the cause of the deactivation of an Ru-based PROX catalyst (Ru/α-Al2O3) by NH3 and the mechanism of suppression of the deactivation by adding Pt (Pt/Ru/α-Al2O3). We found that nitric oxide (NO) produced by oxidation of NH3 induces oxidation of the Ru nanoparticle surface, which deactivates the catalyst via a three-step process: First, NO directly adsorbs on Ru0 to form NO-Ruδ+, which then induces the formation of O-Ru n+ by oxidation of the surrounding Ru0. Then, O-Ru m+ is formed by oxidation of Ru0 starting from the O-Ru n+ nuclei and spreading across the surface of the nanoparticle. Pt inhibits this process by alloying with Ru and inducing the decomposition of adsorbed NO, which keeps the Ru in a metallic state.

Co Nanoparticle Catalysts Encapsulated by BaO-La2O3 Nanofractions for Efficient Ammonia Synthesis Under Mild Reaction Conditions.

Ruthenium catalysts may allow for realization of renewable energy-based ammonia synthesis processes using mild reaction conditions (<400 °C, <10 MPa). However, ruthenium is relatively rare and therefore expensive. Here, we report a Co nanoparticle catalyst loaded on a basic Ba/La2O3 support and prereduced at 700 °C (Co/Ba/La2O3_700red) that showed higher ammonia synthesis activity at 350 °C and 1.0-3.0 MPa than two benchmark Ru catalysts, Cs+/Ru/MgO and Ru/CeO2. The synthesis rate of the catalyst at 350 °C and 1.0 MPa (19.3 mmol h-1 g-1) was 8.0 times that of Co/Ba/La2O3_500red and 6.9 times that of Co/La2O3_700red. The catalyst showed ammonia synthesis activity at temperatures down to 200 °C. Reduction at the high temperature induced the formation of BaO-La2O3 nanofractions around the Co nanoparticles by decomposition of BaCO3, which increased turnover frequency, inhibited the sintering of Co nanoparticles, and suppressed ammonia poisoning. These strategies may also be applicable to other non-noble metal catalysts, such as nickel.

Spectral Embedded Deep Clustering.

We propose a new clustering method based on a deep neural network. Given an unlabeled dataset and the number of clusters, our method directly groups the dataset into the given number of clusters in the original space. We use a conditional discrete probability distribution defined by a deep neural network as a statistical model. Our strategy is first to estimate the cluster labels of unlabeled data points selected from a high-density region, and then to conduct semi-supervised learning to train the model by using the estimated cluster labels and the remaining unlabeled data points. Lastly, by using the trained model, we obtain the estimated cluster labels of all given unlabeled data points. The advantage of our method is that it does not require key conditions. Existing clustering methods with deep neural networks assume that the cluster balance of a given dataset is uniform. Moreover, it also can be applied to various data domains as long as the data is expressed by a feature vector. In addition, it is observed that our method is robust against outliers. Therefore, the proposed method is expected to perform, on average, better than previous methods. We conducted numerical experiments on five commonly used datasets to confirm the effectiveness of the proposed method.

Pt-Co Alloy Nanoparticles on a γ-Al2 O3 Support: The Synergistic Effect between Isolated Electron-Rich Pt and Co for Automotive Exhaust Purification.

Invited for this month's cover is the group of Dr. Katsutoshi Sato and Prof. Dr. Katsutoshi Nagaoka (Kyoto University) and collaborators at Oita and Kyushu Universities. The cover picture shows the proposed mechanism for automotive exhaust purification over a Pt-Co alloy nanoparticle catalyst with an extremely low Pt/Co molar ratio. In the catalyst, the isolated electron-rich Pt atoms are present on the surface of the nanoparticles and play an important role in NOx capture and activation, which are important elementary steps in exhaust purification. Read the full text of the article at 10.1002/cplu.201800542.

Pt-Co Alloy Nanoparticles on a γ-Al2 O3 Support: Synergistic Effect between Isolated Electron-Rich Pt and Co for Automotive Exhaust Purification.

There is interest in minimizing or eliminating the use of Pt in catalysts by replacing it with more widely abundant and cost-effective elements. The alloying of Pt with non-noble metals is a potential strategy for reducing Pt use because interactions between Pt and non-noble metals can modify the catalyst structure and electronic properties. Here, a γ-Al2 O3 -supported bimetallic catalyst [Pt(0.1)Co(1)/Al2 O3 ] was prepared which contained 0.1 wt % Pt and 1 wt % Co and thus featured an extremely low Pt : Co ratio (<1 : 30 mol/mol). The Pt and Co in this catalyst formed alloy nanoparticles in which isolated electron-rich Pt atoms were present on the nanoparticle surface. The activity of this Pt(0.1)Co(1)/Al2 O3 catalyst for the purification of automotive exhaust was comparable to the activities of 0.3 and 0.5 wt % Pt/γ-Al2 O3 catalysts. Electron-rich Pt and metallic Co promoted activation of NOx and oxidization of CO and hydrocarbons, respectively. This strategy of tuning the surrounding structure and electronic state of a noble metal by alloying it with an excess of a non-noble metal will enable reduced noble metal use in catalysts for exhaust purification and other environmentally important reactions.

Photocatalytic chemoselective cleavage of C-O bonds under hydrogen gas- and acid-free conditions.

In the presence of a palladium-loaded TiO2 photocatalyst, the cleavage of benzyl phenyl ether in low-molecular-weight alcohol solvents under de-aerated conditions afforded toluene and phenol simultaneously in a 1 : 1 molar ratio.

Detection of retinoic acid receptor agonistic activity and identification of causative compounds in municipal wastewater treatment plants in Japan.

Retinoic acid (RA) receptor (RAR) agonists are potential toxicants that can cause teratogenesis in vertebrates. To determine the occurrence of RAR agonists in municipal wastewater treatment plants (WWTPs), we examined the RARα agonistic activities of influent and effluent samples from several municipal WWTPs in Osaka, Japan, using a yeast two-hybrid assay. Significant RARα agonistic activity was detected in all the influent samples investigated, suggesting that municipal wastewater consistently contains RAR agonists. Fractionations using high-performance liquid chromatography, directed by the bioassay, found several bioactive peaks from influent samples. The RAR agonists, all-trans RA (atRA), 13-cis RA (13cRA), 4-oxo-atRA, and 4-oxo-13cRA, possibly arising from human urine, were identified by liquid chromatography ion trap time-of-flight mass spectrometry. Quantification of the identified compounds in municipal WWTPs confirmed that they were responsible for the majority of RARα agonistic activity in WWTP influents, and also revealed they were readily removed from wastewater by activated sludge treatment. Simultaneous measurement of the RARα agonistic activity revealed that although total activity typically declined concomitant with the reduction of the four identified compounds, it remained high after the decline of RAs and 4-oxo-RAs in one WWTP, suggesting the occurrence of unidentified RAR agonists during the activated sludge treatment.