A Probabilistic Re-Intepretation of Confidence Scores in Multi-Exit Models

In this paper, we propose a new approach to train a deep neural network with multiple intermediate auxiliary classifiers, branching from it. These ‘multi-exits’ models can be used to reduce the inference time by performing early exit on the intermediate branches, if the confidence of the prediction is higher than a threshold. They rely on the assumption that not all the samples require the same amount of processing to yield a good prediction. In this paper, we propose a way to train jointly all the branches of a multi-exit model without hyper-parameters, by weighting the predictions from each branch with a trained confidence score. Each confidence score is an approximation of the real one produced by the branch, and it is calculated and regularized while training the rest of the model. We evaluate our proposal on a set of image classification benchmarks, using different neural models and early-exit stopping criteria.     

Effects of Regioisomerism on the Antiproliferative Activity of Hydroxystearic Acids on Human Cancer Cell Lines

A series of regioisomers of the hydroxystearic acid (HSA) was prepared, and the effect of the position of the hydroxyl group along the chain on a panel of human cancer cell lines was investigated. Among the various regioisomers, those carrying the hydroxyl at positions 5, 7, and 9 had growth inhibitor activity against various human tumor cell lines, including CaCo-2, HT29, HeLa, MCF7, PC3, and NLF cells. 10-HSA and 11-HSA showed a very weak effect. 8-HSA did not show inhibitory activity in all cell lines. The biological role of 7-HSA and 9-HSA is widely recognized, while little is known about the effects of 5-HSA. Therefore, the biological effects of 5-HSA in HeLa, HT29, MCF7, and NLF cell lines were investigated using the Livecyte’s ptychography technology, which allows correlating changes in proliferation, motility, and morphology as a function of treatment at the same time. 5-HSA not only reduces cell proliferation but also induces changes in cell displacement, directionality, and speed. It is important to characterize the biological effects of 5-HSA, this molecule being an important component of fatty acyl esters of hydroxy fatty acids (FAHFA), a class of endogenous mammalian lipids with noticeable anti-diabetic and anti-inflammatory effects.     

One-pot synthesis of stable NIR tetracene diimides via double cross-coupling

Tetracene tetracarboxylic diimides have been synthesized based on direct double ring extension of electron-deficient naphthalene diimides involving metallacyclopentadienes. Atomic structure and electronic transitions responsible for their NIR absorption spectra are investigated with quantum-chemical calculations. In light of their unique structure and admirable photophysical and electronic properties, this new molecular skeleton is promising candidate for n-type semiconductors.     

Electrochemiluminescence Bioassays with a Water‐Soluble Luminol Derivative Can Outperform Fluorescence Assays

The most efficient and commonly used electrochemiluminescence (ECL) emitters are luminol, [Ru(bpy)₃]²⁺, and derivatives thereof. Luminol stands out due to its low excitation potential, but applications are limited by its insolubility under physiological conditions. The water‐soluble m‐carboxy luminol was synthesized in 15 % yield and exhibited high solubility under physiological conditions and afforded a four‐fold ECL signal increase (vs. luminol). Entrapment in DNA‐tagged liposomes enabled a DNA assay with a detection limit of 3.2 pmol L^?1, which is 150 times lower than the corresponding fluorescence approach. This remarkable sensitivity gain and the low excitation potential establish m‐carboxy luminol as a superior ECL probe with direct relevance to chemiluminescence and enzymatic bioanalytical approaches.     

L‐lysine and EDTA polymer mimics as resins for the quantitative and reversible removal of heavy metal ion water pollutants

Traditional precipitation methods for inorganic micropollutant removal from waters are increasingly being replaced by sorption methods based on both natural and synthetic materials. In this context, two novel effective heavy metal ions absorbers are presented. These resins, LYMA and LMT85, were crosslinked poly(amidoamine)s carrying amine and carboxyl groups in their repeating units. In particular, the LYMA‐repeating unit contains one carboxyl and two amine groups and is a mimic of L ‐lysine, whereas LMT85 contains two amine and five carboxyl groups and is a mimic of EDTA. Both resins were prepared at moderate cost by simple eco‐friendly procedures. The heavy metal ion set adopted as benchmark was Cu²⁺, Cd²⁺, Pb²⁺, Zn²⁺, Ni²⁺, and Co²⁺. LYMA proved selective for Cu²⁺ and Ni²⁺, the other ions tested being negligibly absorbed, whereas LMT85 proved capable of rapidly and quantitatively absorbing all the ions tested either singly or in mixed solution. The absorption process was reversible, and the resins were easily regenerated by acidification. The absorption of several metal ions imparted intense coloring to the resins, a feature possibly exploitable for analytical purposes. © 2012 Wiley Periodicals, Inc. J Polym Sci Part A: Polym Chem, 2012     

A random telegraph signal of Mittag-Leffler type

A general method is presented to explicitly compute autocovariance functions for non-Poisson dichotomous noise based on renewal theory. The method is specialized to a random telegraph signal of Mittag-Leffler type. Analytical predictions are compared to Monte Carlo simulations. Non-Poisson dichotomous noise is non-stationary and standard spectral methods fail to describe it properly as they assume stationarity.     

Incorporation of smooth spherical bodies in the Lattice Boltzmann method

A method to simulate bodies suspended in a Lattice Boltzmann solvent is proposed. It is based on a generalized reaction force that enforces no-slip boundary conditions at the fluid–body interface as the limiting case of an iterative procedure. A smooth version of the Heaviside function allows to treat spherical particles of arbitrary size and produces smooth hydrodynamic forces as particles move in the continuum. Numerical tests demonstrate the accuracy of the method in reproducing the hydrodynamic field around a single particle and the fluid-mediated forces between pairs of particles. The drag force experienced by a particle moving in a straight channel and at various Reynolds numbers is studied as a non-trivial testcase.     

Experimental and theoretical analysis of bias ionization by?α-particles in a nitrogen laser

Nitrogen laser performance with TE configuration and wedge electrodes is analyzed with background ionization in the laser discharge channel by α particles at a low exposition rate. With the bias ionization, the laser power presents two peaks as a function of gas pressure, with one at the normal low pressure, without bias ionization, and the other at high pressure generated by bias ionization. A simple theoretical model has been developed in a trial to understand this behavior. This model was first tested in later results for a TE configuration nitrogen laser, with flat electrodes, without and with bias ionization. It has been observed that due to the competition between electrode shielding by positively charged α particles and bulk ionization by impact, the laser energy is suppressed with pressure below 50 Torr and enhanced above it.