Search for: All records

Deep learning is increasingly moving towards a transfer learning paradigm whereby large foundation models are fine-tuned on downstream tasks, starting from an initialization learned on the source task. But an initialization contains relatively little information about the source task. Instead, we show that we can learn highly informative posteriors from the source task, through supervised or self-supervised approaches, which then serve as the basis for priors that modify the whole loss surface on the downstream task. This simple modular approach enables significant performance gains and more data-efficient learning on a variety of downstream classification and segmentation tasks, serving as a drop-in replacement for standard pre-training strategies. These highly informative priors also can be saved for future use, similar to pre-trained weights, and stand in contrast to the zero-mean isotropic uninformative priors that are typically used in Bayesian deep learning. 

Abstract Watershed resilience is the ability of a watershed to maintain its characteristic system state while concurrently resisting, adapting to, and reorganizing after hydrological (for example, drought, flooding) or biogeochemical (for example, excessive nutrient) disturbances. Vulnerable waters include non-floodplain wetlands and headwater streams, abundant watershed components representing the most distal extent of the freshwater aquatic network. Vulnerable waters are hydrologically dynamic and biogeochemically reactive aquatic systems, storing, processing, and releasing water and entrained (that is, dissolved and particulate) materials along expanding and contracting aquatic networks. The hydrological and biogeochemical functions emerging from these processes affect the magnitude, frequency, timing, duration, storage, and rate of change of material and energy fluxes among watershed components and to downstream waters, thereby maintaining watershed states and imparting watershed resilience. We present here a conceptual framework for understanding how vulnerable waters confer watershed resilience. We demonstrate how individual and cumulative vulnerable-water modifications (for example, reduced extent, altered connectivity) affect watershed-scale hydrological and biogeochemical disturbance response and recovery, which decreases watershed resilience and can trigger transitions across thresholds to alternative watershed states (for example, states conducive to increased flood frequency or nutrient concentrations). We subsequently describe how resilient watersheds require spatial heterogeneity and temporal variability in hydrological and biogeochemical interactions between terrestrial systems and down-gradient waters, which necessitates attention to the conservation and restoration of vulnerable waters and their downstream connectivity gradients. To conclude, we provide actionable principles for resilient watersheds and articulate research needs to further watershed resilience science and vulnerable-water management. 

The mitochondrial carrier family protein SLC25A3 transports both copper and phosphate in mammals, yet in Saccharomyces cerevisiae the transport of these substrates is partitioned across two paralogs: PIC2 and MIR1. To understand the ancestral state of copper and phosphate transport in mitochondria, we explored the evolutionary relationships of PIC2 and MIR1 orthologs across the eukaryotic tree of life. Phylogenetic analyses revealed that PIC2-like and MIR1-like orthologs are present in all major eukaryotic supergroups, indicating an ancient gene duplication created these paralogs. To link this phylogenetic signal to protein function, we used structural modeling and site-directed mutagenesis to identify residues involved in copper and phosphate transport. Based on these analyses, we generated an L175A variant of mouse SLC25A3 that retains the ability to transport copper but not phosphate. This work highlights the utility of using an evolutionary framework to uncover amino acids involved in substrate recognition by mitochondrial carrier family proteins.