Object tracking traditionally depends on raw image data, which is often limited, encrypted, or unavailable in bandwidth-constrained environments. Prior work has demonstrated that compressed image domain features (such as frame sizes) can effectively support high-level vision tasks like action recognition, offering an alternative to full-frame analysis. Building on this premise, we extend the concept for position tracking: sequences of I-frame (intra-coded) and P-frame (predictive-coded) sizes (in bits) are modeled as a time series, functioning as inputs to a machine learning model. Our architecture utilizes a Transformer encoder with a fully connected neural network to estimate an object’s position over time by learning patterns in these compressed representations. This method offers a solution for object tracking in environments where conventional vision pipelines are infeasible. In experiments, the frame size only model attained a minimum mean L2 position error of 0.328 m, compared with 0.145 m for an image-only baseline model, demonstrating that it is viable to use compressed image domain information for tracking in resource-limited scenarios.

Classically-transmitted private messages require complex algorithms of encryption and decryption in order to decrease the chances of an interceptor understanding the message. The development of quantum computers poses a threat to current encryption protocols. The only way to guarantee that no unauthorized third-party observer has intercepted a message is to utilize quantum mechanics. By generating both frequency- and time- entangled signal and idler photons and distributing the photons into their respective channels via a Wavelength Division Multiplexer (WDM), information can be distributed to multiple nodes at the same time. A Joint Spectral Intensity (JSI) graph depicts the number of communicable photons between each channel connection. Only a few central channels can have a reasonably large key rate and photon information efficiency compared to the noise floor. By decreasing the frequency spacing of the WDM channels, a greater number of channels can have usably high key rates. Our goal is to recreate the JSI with lower WDM spacings for various pump frequencies and analyze the resulting photon information efficiency using deconvolution algorithms. We do this by simulating the expected number of coincidences using Python and Matlab scripts. The resulting graph is blurry, with many coincidences detected between non-communicating channels. To analyze the envelope of the main communicating channels, we use two deconvolution algorithms, Tikhonov and Total Variation, to sharpen the blurred JSI. The resulting envelope conveys the distribution of information across different frequency channels, and the number of channel connections with adequate key rates was increased. 

Recent advances in single-cell RNA-sequencing (scRNA-seq) and expanded protein-protein interaction databases have enabled more accurate inference of cell-cell interactions via cell-cell communication (CCC). However, current computational methods for analyzing CCC from scRNA-seq data measured across samples often aggregate signals at the cell-type level, obscuring single-cell heterogeneity.     To overcome these challenges and preserve single-cell resolution, we introduce CC-RISE, an unsupervised alignment-and-decomposition framework that learns shared cellular eigenstates across conditions using Reduction and Insight in Single-cell Exploration (RISE) and factors a cross-condition communication tensor into interpretable components with CANDECOMP/PARAFAC (CP) decomposition. RISE is first applied to the scRNA-seq data measured across samples (genes, cells, and conditions). RISE converts the cells of each condition to cellular eigenstates, allowing for the grouping of cells without cell-type assumptions. Then, the altered dataset (genes, eigenstates, and conditions) undergoes CCC scoring. This aligned 4D tensor describes CCC (ligand-receptor pairs, sender eigen-states, receiver eigen-states, and conditions) and is factored by CP, producing factor matrices that describe patterns across each axis.    We applied CC-RISE to bronchoalveolar lavage fluid scRNA-seq from COVID-19 patients and healthy controls (12 samples, 11,688 genes, 65,813 cells). CC-RISE recovered several ligand-receptor pairs that correlate with COVID-19 severity, previously reported by Tensor-cell2cell. However, whereas the original analysis implicated macrophages as the principal sender cell types, CC-RISE’s higher resolution localized the signal to two of three macrophage subpopulations. These results illustrate that CC-RISE enables granular characterization of cell-cell communication by preserving single-cell resolution, yielding a single-cell-level understanding of disease-linked communication patterns like COVID-19 severity.

The Terahertz (THz) frequency band offers unique advantages for high-resolution sensing and imaging due to its non-ionizing nature and sensitivity to molecular composition. This research presents a versatile free-space THz system built from a core set of components, demonstrating its capability in two distinct configurations: high-precision non-contact displacement measurement and material characterization of aqueous solutions.  Based on a 400 GHz transmitter and receiver, the system was used in two configurations. For vibrometry, it was arranged as a Michelson interferometer, where a split THz beam's interference pattern measures target displacement. To determine the system's performance, a simulation model was calibrated using initial experimental data. For material characterization, the setup was reconfigured into a focused-beam reflectometer by removing the reference arm. The beam was focused directly onto a sample, and the reflected power was measured across a range of frequencies to analyze its properties.  The interferometric configuration proved highly effective for signal reconstruction. The calibrated simulation showed a displacement resolution of 1 µm or better, limited primarily by the receiver's noise floor. This model also successfully simulated the reconstruction of periodic waveforms, such as pure tones and human heartbeat patterns.  In the reflectometer configuration, the system successfully differentiated between several samples, including water, oil, saline, and glucose solutions. These materials each exhibited unique, frequency-dependent reflective signatures across the 370-430 GHz band, enabling their identification. Furthermore, the system demonstrated sensitivity to dissolved solutes, clearly distinguishing a 1 M glucose solution from pure water based on the reflected power, confirming its potential for concentration sensing.

The RNA World Hypothesis proposes RNA as the foundation of early life due to its ability to store genetic information and catalyze bioreactions. Encapsulation into primitive cells (protocells) enhances RNA thermostability, promotes folding and prevents chemical degradation. This supports the possibility of self-replication and function emerging from RNA-based systems through natural selection. To evaluate the evolutionary significance of encapsulation, a high degree of RNA encapsulation is required; however, current experimental methods yield only limited encapsulation. Although many factors can influence RNA encapsulation (temperature, pH, encapsulation method, vesicle size and composition), this study investigates the effect of ionic strength as an encapsulation optimization parameter. The malachite green RNA aptamer (a short RNA sequence) was encapsulated inside model protocells composed of prebiotically plausible oleic acid vesicles with potassium chloride (KCl) concentrations of 0, 10, 50 and 100 mM in the buffer solution. The encapsulation efficiency was calculated by measuring the ratio of malachite green dye fluorescence in the encapsulated fraction to the total pool’s fluorescence using samples collected after size extrusion and size-exclusion chromatographic separation. It was found that increasing the concentration of KCl decreases encapsulation efficiency, with 21% at 0 mM and 3% at 100 mM. This is a strongly-correlated linear relationship with a R-squared value of 0.99. These findings suggest that low ionic strength conditions may favor the development of functionality in RNA aptamers on early earth. These insights can refine future experimental methods to gain more meaningful insights on the effects of protocellular confinement on RNA activity.