Three approaches to facilitate invariant neurons and generalization to out-of-distribution orientations and illuminations.

The training data distribution is often biased towards objects in certain orientations and illumination conditions. While humans have a remarkable capability of recognizing objects in out-of-distribution (OoD) orientations and illuminations, Deep Neural Networks (DNNs) severely suffer in this case, even when large amounts of training examples are available. Neurons that are invariant to orientations and illuminations have been proposed as a neural mechanism that could facilitate OoD generalization, but it is unclear how to encourage the emergence of such invariant neurons. In this paper, we investigate three different approaches that lead to the emergence of invariant neurons and substantially improve DNNs in recognizing objects in OoD orientations and illuminations. Namely, these approaches are (i) training much longer after convergence of the in-distribution (InD) validation accuracy, i.e., late-stopping, (ii) tuning the momentum parameter of the batch normalization layers, and (iii) enforcing invariance of the neural activity in an intermediate layer to orientation and illumination conditions. Each of these approaches substantially improves the DNN's OoD accuracy (more than 20% in some cases). We report results in four datasets: two datasets are modified from the MNIST and iLab datasets, and the other two are novel (one of 3D rendered cars and another of objects taken from various controlled orientations and illumination conditions). These datasets allow to study the effects of different amounts of bias and are challenging as DNNs perform poorly in OoD conditions. Finally, we demonstrate that even though the three approaches focus on different aspects of DNNs, they all tend to lead to the same underlying neural mechanism to enable OoD accuracy gains - individual neurons in the intermediate layers become invariant to OoD orientations and illuminations. We anticipate this study to be a basis for further improvement of deep neural networks' OoD generalization performance, which is highly demanded to achieve safe and fair AI applications.

Expression of Tissue Factor in Epithelial Ovarian Carcinoma Is Involved in the Development of Venous Thromboembolism.

OBJECTIVES: Our 2007 study of 32 patients with ovarian cancer reported the possible involvement of tissue factor (TF) in the development of venous thromboembolism (VTE) before treatment, especially in clear cell carcinoma (CCC). This follow-up study further investigated this possibility in a larger cohort. METHODS: We investigated the intensity of TF expression (ITFE) and other variables for associations with VTE using univariate and multivariate analyses in 128 patients with epithelial ovarian cancer initially treated between November 2004 and December 2010, none of whom had received neoadjuvant chemotherapy. Before starting treatment, all patients were ultrasonographically screened for VTE. The ITFE was graded based on immunostaining of surgical specimens. RESULTS: Histological types were serous carcinoma (n = 42), CCC (n = 12), endometrioid carcinoma (n = 15), mucinous carcinoma (n = 53), and undifferentiated carcinoma (n = 6). The prevalence of VTE was significantly higher in CCC (34%) than in non-CCC (17%, P = 0.03). As ITFE increased, the frequencies of CCC and VTE increased significantly (P < 0.001 and P = 0.014, respectively). Multivariate analysis identified TF expression and pretreatment dimerized plasmin fragment D level as significant independent risk factors for VTE development. These factors showed particularly strong impacts on advanced-stage disease (P = 0.021). CONCLUSIONS: The 2007 cohort was small, preventing multivariate analysis. This study of a larger cohort yielded stronger evidence that the development of VTE in epithelial ovarian cancer may involve TF expression in cancer tissues.

MNBA-mediated ß-lactone formation: mechanistic studies and application for the asymmetric total synthesis of tetrahydrolipstatin.

Various ß-lactones were prepared from ß-hydroxycarboxylic acids by intramolecular dehydration condensation using MNBA, an effective coupling reagent, along with a nucleophilic catalyst. The transition state that provides the desired 4-membered ring model system is disclosed by density functional theory (DFT) calculations, and the structural features of the transition form are discussed. This method was successfully applied to the asymmetric total synthesis of tetrahydrolipstatin (THL), an antiobestic drug used in clinical treatment to inhibit the activity of pancreatic lipase.

Evaluation of the efficiency of the macrolactonization using MNBA in the synthesis of erythromycin A aglycon.

Various intermediates for the synthesis of erythronolide A, an aglycon of erythromycin A, are prepared from the corresponding seco-acids using 2-methyl-6-nitrobenzoic anhydride (MNBA) in the presence of 4-(dimethylamino)pyridine (DMAP) with or without triethylamine. The efficiency of the MNBA lactonization is assessed by studying this method and comparing the results with those of the other established macrocyclization protocols. It has been finally concluded that (i) the conformationally appropriate substrate for the monomeric cyclization gave the desired lactone in excellent yield under mild reaction conditions in the presence of MNBA and DMAP, (ii) the highly-strained substrate for the cyclization also afforded the monomeric lactone in relatively good yield at 100 degrees C in toluene, and (iii) the seco-acid having stable linear conformation, which preferred dimerizing more than forming the monomeric lactone, provided the corresponding diolide in high yield with the constant ratio of the monomer to dimeric lactone (approximately 1/5).

Synthesis of sterically hindered phthalocyanines and their applications to dye-sensitized solar cells.

Phthalocyanines with high peripheral substitutions and free from potential contamination by regioisomers have been synthesized and evaluated as photosensitizers for dye-sensitized solar cell applications. Each of the sterically hindered precursor compounds was accomplished by Suzuki-Miyaura cross-coupling reactions with the arylchloride and corresponding boronic acids. Metal free phthalocyanine-sensitized solar cells showed no photocurrent generation due to its low excited singlet state (LUMO) compared with the conduction band of TiO(2). Upon zinc metalation, the LUMO level of the phthalocyanine was pushed up, and this variation afforded an exergonic free energy change for electron injection. The zinc phthalocyanine-sensitized solar cell displayed 0.57% power conversion efficiency (eta) and 4.9% maximal IPCE in the near infrared region. More importantly, the cell prepared with and without the presence of chenodeoxycholic acid revealed no difference in the power conversion efficiency. This implies that the well-known aggregation tendency of phthalocyanines that is considered to enhance the self-quenching of the phthalocyanine excited singlet state is effectively suppressed by the high degree of substitutions. The significance of the driving force for electron injection and the distance between the dye core and the TiO(2) surface is also highlighted for devising high performance phthalocyanine photosensitizers.

Metastasis of the gastrointestinal tract: FDG-PET imaging.

We assess the usefulness of F-18-fluoro-deoxyglucose (FDG) positron emission tomography (PET) in the evaluation of gastrointestinal metastases. Four cases (five lesions) in which metastases from three lung cancers and one malignant fibrous histiocytoma (MFH) of the femur were found in the gastrointestinal tract were reviewed (men/women 3 : 1, age 63-78 years, mean 72 years). The five lesions were duodenal, jejunal metastasis, and two stomach metastases from lung carcinoma, and rectal metastasis from MFH of the femur. FDG-PET was unable to detect small masses, but it was able to detect unforeseen lesions such as gastrointestinal metastases because FDG-PET is a whole-body scan in a single-operation examination. FDG-PET imaging provided valuable information for the diagnosis of gastrointestinal metastasis.

Reconstruction of Romberg disease defects by omental flap.

Romberg disease is a progressive hemifacial atrophy of unknown etiology. Several methods have been described for its reconstruction. Microsurgical reconstruction, focusing on the correction of facial asymmetry and restoration of contour, has become the standard. We have succeeded a reconstructive technique for Romberg disease using by omental flap. All patients achieved good healing of the transferred omentum without recurrences and complications with a good cosmetic result on long-term follow-up. Donor-site morbidity resulting from omental free flap harvest was minimal. Based on the results of our patients in this series, the surgical results were favorable. This report describes the successful treatment of 8 patients using this approach.