In the field of organic synthesis, there is significant interest in hydride reductants capable of functioning within acidic environments, though such reagents are scarce. Diborane is one such compound that can generate “electrophilic” hydride anions for reduction and can exist in acidic media. However, diborane is difficult to utilize as it is a gas and reagents derived from it are air and water sensitive as well as toxic. Sharing a similar molecular geometry with diborane, δ6-borophane offers a solid-state, scalable, and ambient-stable material. In this work we investigate the reactivity of δ6-borophane and potential applications as a substitute for conventional laboratory reductants. We also explore the interaction between organic substrates and the δ6-borophane surface, assessing the prospects of surface functionalization while preserving structural integrity, thus broadening its usage as a nanomaterial. Solid and liquid nuclear magnetic resonance (NMR) was used to monitor the reactions between δ6-borophane and olefins and epoxides, to determine reactivity and regioselectivity. As a solid, δ6-borophane demonstrates prominent acidic properties. However, when in solution, it shows promising reactivity as a hydride.

MetaDrive (MD) is an autonomous driving simulator. With its lightweight framework and rendering, it can train an agent via an Artificial Intelligence (AI) model, made by an Intelligent Driver Model (IDM) policy, to control an ego vehicle (vehicle in control by a human or AI model). The overall goal is a town simulation where MetaDrive and the agent help the model learn autonomous driving via reinforcement learning. The following objectives are critical to achieving such a goal: importing town maps, running vehicle simulations from imported maps, and having humans take over AI-controlled vehicles. The first objective is the import of the town maps, such as from the Carla simulator. We will then document the happenings when training the model on such imported maps using the IDM policy, which is the second objective. Running the IDM policy will help the agent understand how the map will have the agent control the ego vehicle under a set of traffic rules learned through reinforcement learning. The final objective is collecting data from human control in such an environment, collecting similar data to that of the AI. This data will then feed back to the model, allowing for autonomous driving. We expect MetaDrive to be an efficient way to train AI to drive vehicles, leading to autonomous vehicles for real-life applications, such as driving people in a town using what the AI knows of the roads and directions.

Norm-conserving pseudopotentials (NCPPs) are one of the most popular frameworks for electronic structure theory methods in computational chemistry. Among many other benefits, they allow for convergence to be checked only with the kinetic energy cutoff parameter and retain the orthogonality of wave functions. However, they prove to be computationally demanding for first-row and transition metal elements due to the localized nature of 2p and 3d orbitals. The projector augmented-wave (PAW) method resolves this issue by making smooth orbitals that allow for a lower kinetic energy cutoff and larger grid spacings. However, PAW is unable to retain the orthogonality of wave functions present in NCPPs. We resolved the issues of both frameworks with the recently-developed orthogonal projector-augmented-wave (OPAW) method, a framework that combines the orthogonality of wave functions of NCPPs with the lower kinetic energy cutoff and larger grid spacings of PAW. To test the applicability of OPAW, we combined this framework with the time-dependent (TD) density functional theory (DFT) approach and plan to calculate the exciton peaks of various systems. We aim to show how OPAW is able to both reproduce the results of NCPPs and outperform them by maintaining its accuracy across a larger range of grid spacings. This development will provide a new computational chemistry framework that can make quantum mechanical-level calculations of large systems containing first-row and transition metal elements not only possible but also efficient.

Event cameras, also known as neuromorphic cameras, are a new paradigm of digital vision based on the all-or-nothing response of human retina cells. The “pixels” of the sensor only transmit ternary data based on log-fold changes in intensity, thus only the intensity-transitions are recorded as data. As a consequence, the data is sparse; the camera has high temporal resolution; and the camera has high dynamic range. To leverage the high temporal resolution in machine vision applications, the processing of data must be quick and efficient. Existing image processing techniques, such as Gabor filters, require large amounts of MAC(multiply-and-accumulate) operations, thus a high time-complexity; this conclusion was corroborated with a runtime study of the entire object-tracking pipeline. With optimizations on the convolutions, the entire pipeline would see a significant reduction in run time. As a solution, we looked at stochastic computing(SC). SC stores “data” in the density of “1’s” in a stochastic bitstream. This allows for multiplication and addition to be approximated with a logical AND or logical OR of the bitstreams. Provided by SC approximations of multiplication and addition, we want to leverage the high-potential for parallelism in our designed hardware, to perform low-precision, sparse convolutions on mass amounts of data. To test the implementation of SCIMITAR, we decided to implement the exact hardware-algorithm in Python to provide a point-of-comparison.