We introduce PyVertical, a framework supporting vertical federated learning using split neural networks. The proposed framework allows a data scientist to train neural networks on data features vertically partitioned across multiple owners while keeping raw data on an owner's device. To link entities shared across different datasets' partitions, we use Private Set Intersection on IDs associated with data points. To demonstrate the validity of the proposed framework, we present the training of a simple dual-headed split neural network for a MNIST classification task, with data samples vertically distributed across two data owners and a data scientist.

We present a multi-language, cross-platform, open-source library for asymmetric private set intersection (PSI) and PSI-Cardinality (PSI-C). Our protocol combines traditional DDH-based PSI and PSI-C protocols with compression based on Bloom filters that helps reduce communication in the asymmetric setting. Currently, our library supports C++, C, Go, WebAssembly, JavaScript, Python, and Rust, and runs on both traditional hardware (x86) and browser targets. We further apply our library to two use cases: (i) a privacy-preserving contact tracing protocol that is compatible with existing approaches, but improves their privacy guarantees, and (ii) privacy-preserving machine learning on vertically partitioned data.

Further improvements in power density of polymer electrolyte fuel cells (PEFCs) are required to lower the cost. At high current densities, mass transport related resistances limit the performance due to, mainly, a poor water management within the cell. It is therefore paramount to develop advanced water management strategies to maximize power output. In this context, diffusion layers play a notable role. They must ensure efficient gas transport toward the catalyst layer while removing the electrochemically produced liquid water in the opposite flow direction. A substantial amount of work has been carried out to improve gas diffusion layers (GDLs) and microporous layer over the last years, mainly controlling microstructural features, such as pore sizes, fiber arrangements and porosities, and hydrophobic coating load and distribution [1]. Our group has recently developed an approach consisting of artificially creating hydrophilic patterns throughout the entire GDL thickness using the radiation grafting method. The hydrophilic areas act as low pressure liquid water removal pathways, therefore liberating the remaining areas for the gas to flow throughout less tortuous domains [2]. The performance of PEFCs was significantly increased thanks to the use of the modified GDLs [3]. However, there is still plenty of room for improvement. Capillary pressure is the driving force for liquid water transport and it is influenced by the microstructure and liquid-solid interaction [4]. It therefore stands to reason that, for a given application, further optimization can be achieved when selecting the appropriate diffusion layer microstructure and wettability ratio. While the first experimental data investigating the effect of GDL morphology, contact angle and coating load has been recently published, these experiments require notable infrastructure and time efforts [5]. For that reason, modelling capillary pressure experiments using pore network models can be a valid approach to theoretically assess optimal material parameters. The open-source OpenPNM framework, which has been extensively used in recent fuel cell literature [6], was the software of choice for this investigation. In this talk we will start by presenting the model physics in which a modification of the Washburn equation has been used as required in the intermediate wettability range where this model fails to predict accurately water imbibition and the validation strategy. The definition of coating load and contact angle will be discussed. A good agreement was obtained between model and experiments, specially investigating the influence of coating load, contact angle and pore size distributions. In the second part of the talk, we will provide some material design guidelines based on model output information to maximize the quality of water segregation in GDLs with patterned wettability. References: [1] S. Park, J.-W. Lee, B. N. Popov, Int. J. Hydrogen Energy 37(7), 5850 (2012). [2] A. Forner-Cuenca, J. Biesdorf, L. Gubler, P. M. Kristiansen, T. J. Schmidt, P. Boillat, Adv. Mater. 27, 6317 (2015). [3] A. Forner-Cuenca, J. Biesdorf, V. Manzi-Orezzoli, L. Gubler, T. J. Schmidt, P. Boillat, J. Electrochem. Soc. 163(13), F1389 (2016). [4] U. Pasaogullari, C. Y. Wang, J. Electrochem. Soc. 153(3) A399 (2004). [5] A. Forner-Cuenca, J. Biesdorf, A. Lamibrac, V. Manzi-Orezzoli, F. N. Büchi L. Gubler, T. J. Schmidt, P. Boillat, J. Electrochem. Soc. 163(9), F1038 (2016). [6] J. Gostick, M. Aghighi, J. Hinebaugh, T. Transter, M. A. Hoeh, H. Day, B. Spellacy, M. Eisharqawy, A. Bazylak, A. Burns, W. Lehnert, A. Putz, Comput. Sci. Eng. 18(4), 60 (2016).

Pore network modeling is a widely used technique for simulating multiphase transport in porous materials, but there are very few software options available. This work outlines the OpenPNM package that was jointly developed by several porous media research groups to help address this gap. OpenPNM is written in Python using NumPy and SciPy for most mathematical operations, thus combining Python's ease of use with the performance necessary to perform large simulations. The package assists the user with managing and interacting with all the topological, geometrical, and thermophysical data. It also includes a suite of commonly used algorithms for simulating percolation and performing transport calculations on pore networks. Most importantly, it was designed to be highly flexible to suit any application and be easily customized to include user-specified pore-scale physics models. The framework is fast, powerful, and concise. An illustrative example is included that determines the effective diffusivity through a partially water-saturated porous material with just 29 lines of code.

Polymer electrolyte membrane water electrolysis (PEL) cells are studied in-operando by synchrotron X-ray radiography. Two-phase flow phenomena associated with the evolution of oxygen and hydrogen in the surrounding water are investigated on a running electrolyzer cell. We examine the gas bubble discharge from the porous transport layer (PTL) into the flow channel and discuss the transport of bubbles in the flow channel. The transport of gas inside the PTL and the number of gas bubble discharge sites is examined and correlated with current density.

We present a fabrication method for flexible optically tunable terahertz metamaterial membranes based on thinning and embedding of commercially available silicon wafers in the metamaterial structure. The resulting membrane thickness of less than 25 em allows for quasi etalon-effect free devices which can be designed to show impedance matching to the surrounding air. We fabricated a thin film spectral bandpass filter with a maximal transmission of 85% and a modulation depth upon optical tuning of 98% at an operating frequency of 0.65THz. Further, we discussed the charge carrier dynamics and the requirements for optical tuning.

In the first part of this study, the hydrogen and oxygen permeabilities of Nafion were measured. The aim of the second part of this study presented here is to physically characterize the influence of the aqueous phase, the solid phase, and the intermediate phase in Nafion on the macroscopic hydrogen and oxygen permeabilities. Hereto, a resistor network model morphologically representative for Nafion based on structural investigations reported in the literature is presented in which the different phases are described by individual permeabilities. As a result of the simulations, an enlarged permeability of the solid phase in comparison to that of dry Nafion had to be assumed in order to reproduce the measured influence of temperature and relative humidity on the permeability. This increase of the permeability of the solid phase toward greater water uptake was explained by the effect of water as a plasticizer and the resulting softening of the polymeric matrix. On the basis of the identified mechanisms, approaches to reduce the gas permeability of polymer electrolyte membranes are identified and discussed.

A new fabrication technique for ultra-thin, flexible and optically tunable terahertz metamaterials with embedded silicon is presented. The implemented bandpass filter had a subwavelength thickness of 25 μm and provided a quasi-etalon free maximal amplitude transmission of 80 %. By optical tuning with a 500 mW strong modulation laser an amplitude modulation depth of 94 % was achieved at 0.65 THz.

(PEMFCs) are a major component of a sustainable energy economy. Their high energy and power density make them uniquely capable of replacing the internal combustion engine. The porous electrode in the PEMFC must be designed to withstand the presence of liquid water, which is a by-product of the electrochemical reaction. Ensuring high rates of gaseous reactant transport to the catalyst layer is essential to producing high power density, efficient and cost effective cells. Consequently, the study of liquid water behavior in the porous electrodes is of extreme interest. A very large number of numerical models have been published, attempting to use multiphase flow models in PEMFC and computational fluid dynamics packages based on continuum mechanics to optimize the electrode. There are several limitations to the continuum approach that will be discussed. The most obvious is that the transport properties must be measured experimentally, and then input into the computation as constitutive relationships. This is problematic for properties are difficult to measure, such as effective diffusivity in partially water saturated media [1, 2], gas-liquid surface area, etc. An alternative modeling approach that is receiving increased interest is pore network modeling (PNM). In PNMs, the media is mapped as a set of interconnected pores and throats, transport is modeled as a resistor network and capillary behavior is modeled using percolation theory concepts. In this paradigm there is no difficulty modeling the impact of multiphase flow, and importantly, PNMs do not require experimentally measured transport parameters are constitutive relationships. A pore network model produced using the OpenPNM package has been developed to simulate the impact of mass transfer through the GDL network on the electrochemical kinetics and fuel cell operation. OpenPNM is an open source framework implemented in Python. This framework is capable of building a porous structure, applying pore-scale physics, and simulating numerous algorithms on pore network models. In this work, constant voltage boundary conditions have been applied to the catalyst layer to predict overvoltage in proton exchange fuel cell systems, as shown by the polarization curve in Figure 1. As can be seen, the mass transfer losses place an upper limit on the maximum current that can be generated in a cell. Only mass transfer losses in the GDL are considered, but there is no theoretical reason why the catalyst layer can cannot be included in future work. This approach can be extended to include ionic losses in the membrane phase as well. The eventual goal is to provide a fully viable alternative to the continuum model approach. References 1. J. T. Gostick, M. A. Ioannidis, M. W. Fowler, M. D. Pritzker, J. Power Sources 194, 433 (2009). 2. J. T. Gostick, M. A. Ioannidis, M. W. Fowler, M. D. Pritzker, in Modern Aspects of Electrochemistry, C. Y. Wang, U. Pasaogullari, Eds. (Springer, Berlin, 2010), vol. 49. Acknowledgements This work was funded by AFCC and the NSERC CRD program.

We present a new fabrication technique for silicon-based optically tunable metamaterial membranes. The realized bandpass filter is only 25 μm thick, enabling quasi etalon-free performance. The maximal amplitude transmission of the filter was 80 %, and an amplitude modulation depth of 94 % was achieved at 0.65THz using a 500mW strong modulation laser.

We show that the spectrally wideband optical tunability of terahertz transmission through silicon can be strongly improved by deposition of graphene. We measured enhanced modulation up to ΔM=24% at a low modulation beam power of 40 mW and a maximal modulation depth of up to M=99%.