Two-step interpretable modeling of ICU-AIs.

We present a novel methodology for integrating high resolution longitudinal data with the dynamic prediction capabilities of survival models. The aim is two-fold: to improve the predictive power while maintaining the interpretability of the models. To go beyond the black box paradigm of artificial neural networks, we propose a parsimonious and robust semi-parametric approach (i.e., a landmarking competing risks model) that combines routinely collected low-resolution data with predictive features extracted from a convolutional neural network, that was trained on high resolution time-dependent information. We then use saliency maps to analyze and explain the extra predictive power of this model. To illustrate our methodology, we focus on healthcare-associated infections in patients admitted to an intensive care unit.

Iontronic Neuromorphic Signaling with Conical Microfluidic Memristors.

Experiments have shown that the conductance of conical channels, filled with an aqueous electrolyte, can strongly depend on the history of the applied voltage. These channels hence have a memory and are promising elements in brain-inspired (iontronic) circuits. We show here that the memory of such channels stems from transient concentration polarization over the ionic diffusion time. We derive an analytic approximation for these dynamics which shows good agreement with full finite-element calculations. Using our analytic approximation, we propose an experimentally realizable Hodgkin-Huxley iontronic circuit where micrometer cones take on the role of sodium and potassium channels. Our proposed circuit exhibits key features of neuronal communication such as all-or-none action potentials upon a pulse stimulus and a spike train upon a sustained stimulus.

Unveiling the capabilities of bipolar conical channels in neuromorphic iontronics.

Conical channels filled with an aqueous electrolyte have been proposed as promising candidates for iontronic neuromorphic circuits. This is facilitated by a novel analytical model for the internal channel dynamics [T. M. Kamsma, W. Q. Boon, T. ter Rele, C. Spitoni and R. van Roij, Phys. Rev. Lett., 2023, 130(26), 268401], the relative ease of fabrication of conical channels, and the wide range of achievable memory retention times by varying the channel lengths. In this work, we demonstrate that the analytical model for conical channels can be generalized to channels with an inhomogeneous surface charge distribution, which we predict to exhibit significantly stronger current rectification and more pronounced memristive properties in the case of bipolar channels, i.e. channels where the tip and base carry a surface charge of opposite sign. Additionally, we show that the use of bipolar conical channels in a previously proposed iontronic circuit features hallmarks of neuronal communication, such as all-or-none action potentials and spike train generation. Bipolar channels allow, however, for circuit parameters in the range of their biological analogues, and exhibit membrane potentials that match well with biological mammalian action potentials, further supporting their potential biocompatibility.

Physics captured by data-based methods in El Niño prediction.

On average once every four years, the Tropical Pacific warms considerably during events called El Niño, leading to weather disruptions over many regions on Earth. Recent machine-learning approaches to El Niño prediction, in particular, Convolutional Neural Networks (CNNs), have shown a surprisingly high skill at relatively long lead times. In an attempt to understand this high skill, we here use data from distorted physics simulations with the intermediate-complexity Zebiak-Cane model to determine what aspects of El Niño physics are represented in a specific CNN-based classification method. We find that the CNN can adequately correct for distortions in the ocean adjustment processes, but that the machine-learning method has far more trouble in dealing with distortions in upwelling feedback strength.