Search for: All records

ABSTRACT This study centers the idea that it is not just what science teacher educators (STEs) teach, but how they teach it, that matters. To prepare future teachers who can enact more equitable and transformative reform‐oriented science instruction with multilingual learners, research must explore what STEs are doing, and how, to develop preservice teachers' expansive views of language and understandings around the nuanced ways students might use their diverse language repertoires for sensemaking. Wanting to explore whether our instructional practices as STEs aligned to the translanguaging pedagogy we espouse within our bilingual elementary science methods course, we employed self‐study methodology to critically examine our own instruction across a semester, specifically in terms of how we engaged in the core practice of “eliciting student ideas.” Findings revealed particularities around the evolution of our translanguaging pedagogy with respect to this core practice, the extensive and intentional effort that went into designing learning activities strongly suited to facilitate our elicitation of PSTs' ideas in language‐expansive ways, and the vulnerable space that we had to hold, as individuals and as a collective, in order to (un)learn and carry out this work. These findings highlight the importance of STEs addressing their own continued professional growth, and of the power of collaboration in supporting this growth through self‐examination and loving self‐critique. Furthermore, findings suggest the importance of intentionally eliciting and elevating PSTs' ideas both implicitly and explicitly, and for needing to attend to the emotional and relational aspects of learning environments to support students' language use. 

Constructing k-nearest neighbor (kNN) graphs is a fundamental component in many machine learning and scientific computing applications. Despite its prevalence, efficiently building all-nearest-neighbor graphs at scale on distributed heterogeneous HPC systems remains challenging, especially for large sparse non-integer datasets. We introduce optimizations for algorithms based on forests of random projection trees. Our novel GPU kernels for batched, within leaf, exact searches achieve 1.18× speedup over sparse reference kernels with less peak memory, and up to 19× speedup over CPU for memory-intensive problems. Our library,PyRKNN, implements distributed randomized projection forests for approximate kNN search. Optimizations to reduce and hide communication overhead allow us to achieve 5× speedup, in per iteration performance, relative to GOFMM (another projection tree, MPI-based kNN library), for a 64M 128d dataset on 1,024 processes. On a single-node we achieve speedup over FAISS-GPU for dense datasets and up to 10× speedup over CPU-only libraries.PyRKNNuniquely supports distributed memory kNN graph construction for both dense and sparse coordinates on CPU and GPU accelerators. 

Federated continual learning is a decentralized approach that enables edge devices to continuously learn new data, mitigating catastrophic forgetting while collaboratively training a global model. However, existing state-of-the-art approaches in federated continual learning focus primarily on learning continuously to classify discrete sets of images, leaving dense regression tasks such as depth estimation unaddressed. Furthermore, autonomous agents that use depth estimation to explore dynamic indoor environments inevitably encounter spatial and temporal shifts in data distributions. These shifts trigger a phenomenon called spatio-temporal catastrophic forgetting, a more complex and challenging form of catastrophic forgetting. In this paper, we address the fundamental research question: “Can we mitigate spatiotemporal catastrophic forgetting in federated continual learning for depth estimation in dynamic indoor environments?”. To address this question, we propose Local Online and Continual Adaptation (LOCA), the first approach to address spatio-temporal catastrophic forgetting in dynamic indoor environments. LOCA relies on two key algorithmic innovations: online batch skipping and continual local aggregation. Our extensive experiments show that LOCA mitigates spatio-temporal catastrophic forgetting and improves global model performance, while running on-device up to 3.35× faster and consuming 3.13× less energy compared to state-of-the-art. Thus, LOCA lays the groundwork for scalable autonomous systems that adapt in real time to learn private and dynamic indoor environments. 

Abstract Tides are an important factor shaping the sea ice system in the Arctic Ocean by altering vertical heat fluxes and advection patterns. Unfortunately, observations are sparse, and the analysis of tides is complicated by the proximity of wind-driven inertial oscillations to the semidiurnal frequencies. Furthermore, computational costs typically prohibit the inclusion of tides in ocean models, leaving a significant gap in our understanding. Motivated by summer observations showing elevated downward surface heat fluxes in the presence of tides, we analyzed simulations carried out with an eddy-permitting coupled ice–ocean model to quantify the impact of tidal effects on Arctic sea ice. In line with previous studies, we find an overall decrease in sea ice volume when tides are included in the simulations, associated with increased vertical mixing and the upward flux of heat from deeper layers of the Arctic Ocean, but this sea ice volume decrease is less pronounced than previously thought. Surprisingly, our simulations suggest that in summer, Arctic sea ice area is larger, by up to 1.5%, when tides are included in the simulations. This effect is partly caused by an increased downward surface heat flux and a consequently lower sea surface temperature, delaying sea ice melting predominantly in the Siberian Seas, where tides are moderately strong and the warm Atlantic Water core is located relatively deep and does not encroach on the wide continental shelf. Here, tidally enhanced downward heat flux from the surface in summer can dominate over the increased upward heat flux from the warm Atlantic Water layer. Significance StatementThis study sheds light on the complex and understudied role of tides in Arctic sea ice dynamics. By utilizing advanced computer models, our research uncovers that, contrary to common expectations, tides contribute to a seasonal increase in sea ice area by up to 1.5% in summer. This effect is attributed to enhanced advection of sea ice into the Siberian Seas and a local increase in downward heat flux reducing sea surface temperatures, thereby delaying sea ice melting in this region. Our findings challenge prevailing notions about the negative impact of tides on sea ice and highlight the importance of incorporating tidal impacts in ocean models to improve predictions of Arctic sea ice changes, key for our understanding of both Arctic and global climate dynamics. 

Abstract Triplet‐triplet annihilation photon upconversion (TTA‐UC) converts low‐energy photons to higher‐energy ones under low‐intensity incoherent excitation, thus enabling applications in fields ranging from medicine to solar energy conversion. Silylethynyl mono‐ and di‐substitution of acenes offers an attractive route to creating new annihilators that operate with minimal energy loss. Here, it is demonstrated that this approach can be extended to pyrene, yielding annihilators that display efficient red‐to‐blue upconversion. While pyrene is the namesake of P‐type delayed fluorescence, the original name for triplet‐triplet annihilation, it is known to be a poor annihilator due to its propensity for forming excimers. By tetra‐substituting pyrene with silylethynyl groups, excimer formation is substantially hindered while simultaneously minimizing the energy gap between the singlet and triplet pair states that participate in TTA‐UC, yielding outstanding annihilators for red‐to‐blue upconversion that operate with quantum yields of upward of 19% (29% when corrected for inner filter effects). Further, it is found that reducing the bulkiness of the silyl substituents is key to achieving high TTA‐UC quantum yields, which highlights the importance of annihilator side group selection when optimizing photon upconversion. 

Not AvailablDeploying monocular depth estimation on resource-constrained edge devices is a significant challenge, particularly when attempting to perform both training and inference concurrently. Current lightweight, self-supervised approaches typically rely on complex frameworks that are hard to implement and deploy in real-world settings. To address this gap, we introduce the first framework for Lightweight Training and Inference (LITI) that combines ready-to-deploy models with streamlined code and fully functional, parallel training and inference pipelines. Our experiments show various models being deployed for inference, training, or both inference and training, leveraging inputs from a real-time RGB camera sensor. Thus, our framework enables training and inference on resource-constrained edge devices for complex applications such as depth estimation. 

Abstract We introduce a new class of chemical probes for activity‐based sensing of proteases, termed cleavable, locked initiator probes (CLIPs). CLIPs contain a protease‐cleavable peptide linked between two programmable DNA strands—an “initiator” DNA and a shorter “blocking” DNA. These DNA sequences are designed to hybridize, creating a “locked” hairpin‐like structure. Upon proteolytic cleavage, the initiator strand is released, triggering the activation of CRISPR‐Cas12a enzymes and producing an amplified fluorescence response. CLIPs generate more than 20‐fold turn‐on signals at room temperature (25 °C), significantly outperforming commercial probes by yielding ∼40‐fold lower limits of detection (LOD) at 100‐fold lower concentrations. Their versatility enables the detection of various disease‐relevant proteases—including the SARS‐CoV‐2 main protease, caspase‐3, matrix metalloproteinase‐7, and cathepsin B—simply by altering the peptide sequence. Importantly, CLIPs detect cathepsin B in four different colorectal cancer cell lines, highlighting their clinical potential. Taken together, the sensitivity (LOD: ∼88 pM), selectivity, and rapid assay time (down to 35 min), combined with the ability to operate in complex biological media with minimal sample preparation, position CLIPs as powerful chemical tools for activity‐based sensing of functional enzymes. 

Speech conveys both linguistic messages and a wealth of social and identity information about a talker. This information arrives as complex variations across many acoustic dimensions. Ultimately, speech communication depends on experience within a language community to develop shared long-term knowledge of the mapping from acoustic patterns to the category distinctions that support word recognition, emotion evaluation, and talker identification. A great deal of research has focused on the learning involved in acquiring long-term knowledge to support speech categorization. Inadvertently, this focus may give the impression of a mature learning endpoint. Instead, there seems to be no firm line between perception and learning in speech. The contributions of acoustic dimensions are malleably reweighted continuously as a function of regularities evolving in short-term input. In this way, continuous learning across speech impacts the very nature of the mapping from sensory input to perceived category. This article presents a case study in understanding how incoming sensory input—and the learning that takes place across it—interacts with existing knowledge to drive predictions that tune the system to support future behavior.