Search for: All records

This study presents a comprehensive taxonomic revision of the family Suberitidae (Porifera: Demospongiae) for California, USA. We include the three species previously known from the region, document two additional species previously known from other regions, and formally describe four new species as Pseudosuberites latke sp. nov., Suberites californiana sp. nov., Suberites kumeyaay sp. nov., and Suberites agaricus sp. nov. Multi-locus DNA sequence data is presented for seven of the nine species, and was combined with all publicly available data to produce the most comprehensive global phylogeny for the family to date. By integrating morphological and genetic data, we show that morphological characters may be sufficient for regional species identification but are likely inadequate for global classification into genera that reflect the evolutionary history of the family. We therefore propose that DNA sequencing is a critical component to support future taxonomic revisions. 

The maximum parsimony phylogenetic reconciliation problem seeks to explain incongruity between a gene phylogeny and a species phylogeny with respect to a set of evolutionary events. While the reconciliation problem is well-studied for species and gene trees subject to events such as duplication, transfer, loss, and deep coalescence, recent work has examined species phylogenies that incorporate hybridization and are thus represented by networks rather than trees. In this paper, we show that the problem of computing a maximum parsimony reconciliation for a gene tree and species network is NP-hard even when only considering deep coalescence. This result suggests that future work on maximum parsimony reconciliation for species networks should explore approximation algorithms and heuristics. 

Deep learning models are increasingly used for end-user applications, supporting both novel features such as facial recognition, and traditional features, e.g. web search. To accommodate high inference throughput, it is common to host a single pre-trained Convolutional Neural Network (CNN) in dedicated cloud-based servers with hardware accelerators such as Graphics Processing Units (GPUs). However, GPUs can be orders of magnitude more expensive than traditional Central Processing Unit (CPU) servers. These resources could also be under-utilized facing dynamic workloads, which may result in inflated serving costs. One potential way to alleviate this problem is by allowing hosted models to share the underlying resources, which we refer to as multi-tenant inference serving. One of the key challenges is maximizing the resource efficiency for multi-tenant serving given hardware with diverse characteristics, models with unique response time Service Level Agreement (SLA), and dynamic inference workloads. In this paper, we present PERSEUS, a measurement framework that provides the basis for understanding the performance and cost trade-offs of multi-tenant model serving. We implemented PERSEUS in Python atop a popular cloud inference server called Nvidia TensorRT Inference Server. Leveraging PERSEUS, we evaluated the inference throughput and cost for serving various models and demonstrated that multi-tenant model serving led to up to 12% cost reduction.