32151 research outputs found
Sort by
Hadamard domain training with integers for class incremental quantized learning
Continual learning (CL) implemented directly on-device is crucial for practical deployment of applications to battery powered devices, where privacy needs must be balanced with personalization, and agility in adapting to new data. Existing CL techniques can be cost-prohibitive on such devices requiring quantized operations to enable practical deployment. However, as we show commonly used fully quantized training (FQT) solutions do not converge when applied to CL with low-precision hardware. We propose Hadamard Domain Quantized Training (HDQT), that uses the Hadamard transform to facilitate FQT with 4-bit integer operands. HDQT enables low-precision, on-device training where other FQT solutions fail. An examination of gradient alignment reveals that for early feature detection layers, HDQT gradients are better aligned to the unquantized baselines than those generated by other FQT methods. This improved alignment translates to consistently better performance over the course of learning, reflected in the training trajectories through the model loss-landscape. Numerical experiments conducted on Human Activity Recognition (HAR) datasets reveal a ≪ 1% average accuracy reduction across various competitive CL methods, even under aggressive 4-bit quantization with 8-bit accumulators. Additionally, on CIFAR100 there is no loss of accuracy observed when the accumulator precision is relaxed to 12 bits for competitive non-dynamic architecture CL methods
Scribal assimilation as inter-textual allusion
In this essay I examine instances of scribal assimilation in ancient Jewish scripture and identify the literary features that triggered scribal recollections of other texts. I argue that some instances of assimilation can be understood as inter-textual allusion, and should be analyzed as such with reference to their formal features and possible exegetical functions. Finally, I conclude with reflections on the relationship between assimilation and the formation of a scripture collection.Peer reviewe
Sedimentology and stratigraphy of lower Cretaceous fluvial to shallow marine deposits on the central Atlantic passive margin : the Aaiun-Tarfaya Basin, Morocco
This paper presents the first integrated regional outcrop-based sedimentological study of the northern Aaiun-Tarfaya Basin located in Morocco (NW Africa). The Lower Cretaceous Tan-Tan Formation has been subdivided into six new members and placed within a sequence stratigraphic framework that includes two incomplete depositional sequences. Strong thickness variations of individual lithostratigraphic units from north to south suggest differential subsidence during sedimentation and/or the existence of major topography on the basal unconformity that the succession onlaps. The results provide valuable insights into the timing of local tectonics in the Western Anti-Atlas and the control on the evolution of the sedimentary system. Deposition of each of these six units is interpreted to be the result of a complex interplay between an overall eustatic sea-level rise during the early Cretaceous, sediment delivery controlled by tectonic movements in the Western Anti-Atlas and Reguibat Shield and periods of differential subsidence in the basin. The results document the style of evolution of a back-stepping wave-dominated system feeding into the Central Atlantic during the passive margin phase. The improved facies and depositional models together with improved understanding of the evolution of the delta have significant implication for exploring the deep-water equivalents offshore.Peer reviewe
Radio and millimeter observations of the environments of young stars undergoing accretion bursts
Studying the formation and evolution of Sun-like stars is crucial for understanding the origins of the Solar System. Young stellar objects (YSOs) undergo episodic outbursts that may impact disk chemistry and planet formation. The FU Orionis- and EX Lupi-type objects (FUors and EXors) experience dramatic brightness increases due to enhanced accretion, and while these outbursts may have lasting effects on the disk chemistry and surrounding environments, their impact remains poorly explored, especially in the radio and millimeter wavelengths.
In this PhD thesis, I conducted the first dedicated radio surveys of FUors and EXors, using ammonia (NH₃) with the Effelsberg 100-m telescope to trace their dense cores and water (H₂O) maser emission as an accretion burst indicator. Additionally, I performed the first wideband millimeter line survey of the evolved FUor V1057 Cyg with the IRAM 30-m telescope, complemented with additional data from the APEX 12-m telescope.
The NH3 survey detected (1,1) emission in over half of the sampled objects (28/51), revealing cold regions (below 25 K) with column densities similar to those found in infrared dark clouds. Archival Herschel SPIRE data provided H₂ column densities and dust temperatures, identifying dense material in at least seven sources. The detection of the (2,2) emission suggests some FUors and EXors may be younger than previously classified.
The 22.2 GHz H₂O maser survey detected emission in three sources (6%), including two FUors and one EXor. Notably, masers were observed in the FUor HH 354 IRS for the very first time, and multiple flares were identified in archival data for other sources, suggesting episodic variability.
The millimeter survey of V1057 Cyg identified over 35 molecular species, including complex organic molecules. Position-velocity diagrams of ¹²CO suggest past episodic outbursts, with a dynamical timescale of the outflow on the orders of tens of thousands of years, indicating that it cannot represent the ongoing outburst alone. This work demonstrates the potential of molecular line studies to refine the classification and chemical characterization of eruptive stars
Applying distance sampling to estimate densities of fin whale calls recorded by ocean bottom seismometers in the Marianas region
Funding: This research was funded by both the Office of Naval Research under Award No. N00014-21-1-2564 and Living Marine Resources under Award No. N39430-21-C-2208.Ocean bottom seismometers (OBSs) are opportunistic instruments that can be used for passive acoustic monitoring of fin whales. To estimate animal population densities from acoustic data using distance sampling, a standard density estimation method, it is essential to estimate ranges to detected calls. Regional OBS networks are too widely spaced to localize fin whale calls through arrival time differences on multiple instruments. This motivates the development of ranging methods using single instruments. We previously developed and published a single-OBS ranging method for fin whale calls that uses the relative timing of multipath arrivals for calls reverberating in the water column to estimate horizontal ranges to fin whales. In this study, we applied multipath ranging and distance sampling methods to fin whale calls recorded on 7 OBSs deployed over a 200 × 300 km2 area east of the Marianas Trench from February 2012 to January 2013. We detected and ranged to calls within 40 km at each OBS, then applied point-transect distance sampling methods to fit detection functions and estimate the probability of call detection. We calculated monthly fin whale call densities during the Marianas calling season from December to April, except for February when frequent airgun shots prevented ranging. We discuss sources of uncertainty in call density estimations, and identify encounter rate variance as the largest contributor. Increasing the number of OBSs used for ranging would improve the precision of call density estimates. Looking forward, converting call densities into estimates of animal abundance will require independent estimates of fin whale call rates.Peer reviewe
Real-life and fictional role models as a resource for rights education for and about girls
Funding: This work was supported by the Arts and Humanities Research Council.Mae Jemison, Little People Big Dreams by Maria Isabel Sánchez Vegara and Janna Morton, tells the story of the first African American woman astronaut. In the novel Johnny Mad Dog by Emmanuel Dongala,16-year-old schoolgirl Laokolé is inspired by Mae’s story as she struggles to survive conflict in an unnamed African country. This article takes an interdisciplinary approach to analyse how narratives of real and fictional female role models are deployed in both texts and offer material for human rights and global citizenship education. As a form of utopian ideation, role model narratives can inspire transformative agency, but also raise questions around the relationship between individual striving and societal change, and the value of ‘representation’. Applying critical literacy methodology, this article explores how these stories shed light both on the significance of female role models as an educational resource and on connections and tensions between human rights and global citizenship values.Peer reviewe
Precision molecular editing : predicting substrate scope and regiochemistry for CHEESY1, a flavin dependent halogenase
Funding: EPSRC PRIMED for Diversification EP/Z533610/1 and BBSRC BB/Y513738/1 under the Super Follow on Fund Scheme, BA Researchers at Risk Fellowship Program (O.H.), UKRI EP-X030008 (initially selected for funding by ERC PoC), China Scholarship Council (Y.Z.), Criticat (J.D.). We thank the European Research Council under the European Union’s Seventh Framework Programme (FP7/2007-2013/ERC grant agreement no 614779 GenoChemetics to R.J.M.G.).The ability to carry out C–H activation at any site on any heteroaromatic scaffold is a holy grail, offering the potential to revolutionize molecule making. Precision editing, activating, and replacing a C–H bond with a carbon halogen bond opens the way to almost any diversification imaginable. Here, through genome mining and in silico analysis, we present a previously undescribed halogenase tool for precision C–H activation and halogenation. While many halogenated metabolites have been found in the marine environment, indicating the operation of a vast array of halogenases, the presence of such halogenases in other salty environments is less well-known. Here, we describe the first discovery and utilization of a halogenase from a microbe associated with salty and fermented food: specifically brined cheese. Most halogenases explored so far have been identified through their association with a biosynthetic gene cluster of a known natural product. Based on their role in the biosynthesis of that natural product, their native substrate is predicted. Many exciting and unexplored halogenases exist discretely of identified biosynthetic clusters. We describe an approach of carrying out halogenase discovery (unrelated to and unguided by known biosynthetic pathways and their encoded natural products) and predicting non-native substrates that the enzyme can and cannot process as well as the regiochemistry of each biotransformation. Following carrying out discovery in silico, we demonstrate the validation of the discovery results in the laboratory. CHEESY1 (Chemistry Helper Enzyme Enabling SelectivitY1) is shown to regioselectively halogenate a broad structural range of medicinally relevant heterocycles, including quinolines, isoquinoline, phenylpyrazole, and flavonoids. Being able to predict from gene to substrate to product for this powerful class of enzymes, which together afford the opportunity for providing a general solution for precision molecule editing, ahead of carrying out wet experimentation opens up exciting opportunities for the future of chemical catalysis and synthesis.Peer reviewe
Size and body condition drive the energetic cost of a baleen whale foraging in shallow habitat
Funding: Data collection was supported by the NOAA National Marine Fisheries Service Office of Science and Technology Ocean Acoustics Program [2016 and 2017; 50-27], the Oregon Sea Grant Program Development funds [2018; RECO-40-PD], the Oregon State University Marine Mammal Institute [2019], and the Office of Naval Research Marine Mammals and Biology program [2020-2022; #N00014-20-1-2760].Energy expenditure strongly influences an animal’s foraging decisions and activity budgets. Diving animals especially need to be energetically efficient because they exercise while oxygen is limited. By estimating the energetics of behavior, we can better understand the cascading effects of individual responses to disturbance and environmental change. Pacific Coast Feeding Group (PCFG) gray whales use a variety of foraging tactics in shallow habitats (<20 m), which present challenges associated with maneuverability and buoyancy. We use a seven-year dataset of concurrent individual behavior, morphology, and breath-by-breath respiration data collected via drone paired with two years of tri-axial accelerometry tag data to study patterns and correlates of respiration. We assess how several respiration metrics (acting as proxies for oxygen consumption) are associated with individual length, body condition and behavior (forage and travel), and test whether respiration reflects recovery from, or anticipation of, a foraging dive using Bayesian linear mixed effects models. Given model results, we simulated daily field metabolic rate (FMR) to explore how diving costs may affect energetics at a daily scale. We find that respiration reflects recovery from the preceding dive and that dives are more energetically expensive for longer, more buoyant whales. Longer dives and the most common foraging tactics also incur higher energetic costs. FMR simulations show that individual size and dive duration have the largest effects on energy expenditure. Thus, PCFG gray whale foraging success may be limited by the energetic costs associated with size and buoyancy, highlighting the costs of a shallow habitat foraging niche.Peer reviewe
Elevation-dependent warming : observations, models, and energetic mechanisms
Observational data and numerical models suggest that, under climate change, elevated land surfaces warm faster than non-elevated ones. Proposed drivers of this “elevation-dependent warming” (EDW) include surface albedo and water vapour feedbacks, the temperature dependence of longwave emission, and aerosols. Yet the relative importance of each proposed mechanism both regionally and at large scales is unclear, highlighting an incomplete physical understanding of EDW. Here we expand on previous regional studies and use gridded observations, atmospheric reanalysis, and a range of climate model simulations to investigate EDW over the historical period across the tropics and subtropics (40° S to 40° N). Observations, reanalysis, and fully coupled models exhibit annual mean warming trends (1959–2014), binned by surface elevation, which are larger over elevated surfaces and broadly consistent across datasets. EDW varies by season, with stronger observed signals in local winter and autumn. Analysis of large ensembles of single-forcing simulations (1959–2005) suggests historical EDW is likely a forced response of the climate system rather than an artefact of internal variability and is primarily driven by increasing greenhouse gas concentrations. To gain quantitative insight into the mechanisms contributing to large-scale EDW, a forcing–feedback framework based on top-of-atmosphere energy balance is applied to the fully coupled models. This framework identifies the Planck and surface albedo feedbacks as being robust drivers of EDW (i.e. enhancing warming over elevated surfaces), with energy transport by the atmospheric circulation also playing an important role. In contrast, water vapour and cloud feedbacks along with weaker radiative forcing in elevated regions oppose EDW. Implications of the results for understanding future EDW are discussed.Peer reviewe
Intraguild predation in sympatric seals and the effect on a declining population
Funding: Natural Environment Research Council(NERC), Grant/Award Number: NE/R015007/1; The Scottish Universities Partnership for Environmental Research,Grant/Award Number: NE/S007342/1.1. Evidence of intraguild predation (IGP) is found across taxonomic groups but varies in its prevalence within predator populations. The effects of IGP can be similar to traditional predation, but when exploited by few, specialist predators the impact on intraguild prey populations is less clear. 2. Grey seals (Halichoerus grypus) and harbour seals (Phoca vitulina) are marine top predators with similar diets of a broad range of fish and cephalopods. However, in recent years, adult male grey seals have been observed predating both grey seals and harbour seals. 3. Combining 11 years of strandings data and direct observations of predation from citizen-science across Scotland, we examined the prevalence, spatial extent and temporal trends of IGP by grey seals. These results informed realistic predation scenarios and the effect of IGP on a small, protected population of harbour seals was explored. 4. IGP was geographically widespread, involving multiple, seemingly specialist adult males across distinct subpopulations. The prevalence of IGP, as revealed by strandings data, appeared to increase through the study. Predation was most pronounced on adult harbour seals during the breeding season, amplifying the population level impact of each predation event through the loss of future reproductive potential. Indeed, in a depleted population in southeast Scotland, the estimated peak predation level was projected to cause declines of 7%–11%, well within the rate of the current harbour seal decline. 5. This study demonstrated the utility of integrating disparate datasets to address conservation challenges and highlights how IGP, while seemingly rare, can hold sympatric species in a predator pit and in small populations may contribute to declines.Peer reviewe