We present an active learning strategy for training parametric models of distance metrics, given triplet-based similarity assessments: object $x_i$ is more similar to object $x_j$ than to $x_k$. In contrast to prior work on class-based learning, where the fundamental goal is classification and any implicit or explicit metric is binary, we focus on perceptual metrics that express the degree of (dis)similarity between objects. We find that standard active learning approaches degrade when annotations are requested for batches of triplets at a time: our studies suggest that correlation among triplets is responsible. In this work, we propose a novel method to decorrelate batches of triplets, that jointly balances informativeness and diversity while decoupling the choice of heuristic for each criterion. Experiments indicate our method is general, adaptable, and outperforms the state-of-the-art.

We have previously described Tapestry Pooling, a scheme to enhance the capacity of RT-qPCR testing, and provided experimental evidence with spiked synthetic RNA to show that it can help to scale testing and restart the economy. Here we report on validation studies with Covid19 patient samples for the Tapestry Pooling scheme with prevalence in the range of 1% to 2%. We pooled RNA extracted from patient samples that were previously tested for Covid19, sending each sample to three pools. Following three different pooling schemes, we pipetted 320 samples into 48 pools with pool size of 20 at prevalence rate of 1.6%, 500 samples into 60 pools with pool size of 25 at prevalence rate of 2%, and 961 samples into 93 pools with pool size of 31 at prevalence rate of 1%. Of the 191 RT-qPCR experiments that we performed, only one pool was incorrect (false negative). Our recovery algorithm correctly called results for the individual samples, with a 100% sensitivity and a 99.9% specificity, with only one false positive across all the 1,781 blinded results required to be called. We show up to 10X savings in the number of tests required at a range of prevalence rates and pool sizes. These experiments establish that Tapestry Pooling is robust enough to handle the diversity of sample constitutions and viral loads seen in real-world samples.

No abstract