Collaborating Authors

[Report] Instantaneous ion configurations in the K ion channel selectivity filter revealed by 2D IR spectroscopy


Potassium channels are responsible for the selective permeation of K ions across cell membranes. K ions permeate in single file through the selectivity filter, a narrow pore lined by backbone carbonyls that compose four K binding sites. Here, we report on the two-dimensional infrared (2D IR) spectra of a semisynthetic KcsA channel with site-specific heavy (13C18O) isotope labels in the selectivity filter. The ultrafast time resolution of 2D IR spectroscopy provides an instantaneous snapshot of the multi-ion configurations and structural distributions that occur spontaneously in the filter. Two elongated features are resolved, revealing the statistical weighting of two structural conformations.

Optimizing organic electrosynthesis through controlled voltage dosing and artificial intelligence


Organic electrosynthesis can transform the chemical industry by introducing electricity-driven processes that are more energy efficient and that can be easily integrated with renewable energy sources. However, their deployment is severely hindered by the difficulties of controlling selectivity and achieving a large energy conversion efficiency at high current density due to the low solubility of organic reactants in practical electrolytes. This control can be improved by carefully balancing the mass transport processes and electrocatalytic reaction rates at the electrode diffusion layer through pulsed electrochemical methods. In this study, we explore these methods in the context of the electrosynthesis of adiponitrile (ADN), the largest organic electrochemical process in industry. Systematically exploring voltage pulses in the timescale between 5 and 150 ms led to a 20% increase in production of ADN and a 250% increase in relative selectivity with respect to the state-of-the-art constant voltage process.

Selectivity considered harmful: evaluating the causal impact of class selectivity in DNNs Machine Learning

Class selectivity, typically defined as how different a neuron's responses are across different classes of stimuli or data samples, is a common metric used to interpret the function of individual neurons in biological and artificial neural networks. However, it remains an open question whether it is necessary and/or sufficient for deep neural networks (DNNs) to learn class selectivity in individual units. In order to investigate the causal impact of class selectivity on network function, we directly regularize for or against class selectivity. Using this regularizer, we were able to reduce mean class selectivity across units in convolutional neural networks by a factor of 2.5 with no impact on test accuracy, and reduce it nearly to zero with only a small ($\sim$2%) change in test accuracy. In contrast, increasing class selectivity beyond the levels naturally learned during training had rapid and disastrous effects on test accuracy. These results indicate that class selectivity in individual units is neither neither sufficient nor strictly necessary for DNN performance, and more generally encourage caution when focusing on the properties of single units as representative of the mechanisms by which DNNs function.

A Multi-Domain Evaluation of Scaling in a General Episodic Memory

AAAI Conferences

Episodic memory endows agents with numerous general cognitive capabilities, such as action modeling and virtual sensing. However, for long-lived agents, there are numerous unexplored computational challenges in supporting useful episodic-memory functions while maintaining real-time reactivity. In this paper, we review the implementation of episodic memory in Soar and present an expansive evaluation of that system. We demonstrate useful applications of episodic memory across a variety of domains, including games, mobile robotics, planning, and linguistics. In these domains, we characterize properties of environments, tasks, and episodic cues that affect performance, and evaluate the ability of Soar’s episodic memory to support hours to days of real-time operation.