Search for: All records

Abstract Energy efficiency in computation is ultimately limited by noise, with quantum limits setting the fundamental noise floor. Analog physical neural networks hold promise for improved energy efficiency compared to digital electronic neural networks. However, they are typically operated in a relatively high-power regime so that the signal-to-noise ratio (SNR) is large (>10), and the noise can be treated as a perturbation. We study optical neural networks where all layers except the last are operated in the limit that each neuron can be activated by just a single photon, and as a result the noise on neuron activations is no longer merely perturbative. We show that by using a physics-based probabilistic model of the neuron activations in training, it is possible to perform accurate machine-learning inference in spite of the extremely high shot noise (SNR  ~ 1). We experimentally demonstrated MNIST handwritten-digit classification with a test accuracy of 98% using an optical neural network with a hidden layer operating in the single-photon regime; the optical energy used to perform the classification corresponds to just 0.038 photons per multiply-accumulate (MAC) operation. Our physics-aware stochastic training approach might also prove useful with non-optical ultra-low-power hardware. 

Abstract The HIV reservoir consists of infected cells in which the HIV-1 genome persists as provirus despite effective antiretroviral therapy (ART). Studies exploring HIV cure therapies often measure intact proviral DNA levels, time to rebound after ART interruption, or ex vivo stimulation assays of latently infected cells. This study utilizes barcoded HIV to analyze the reservoir in humanized mice. Using bulk PCR and deep sequencing methodologies, we retrieve 890 viral RNA barcodes and 504 proviral barcodes linked to 15,305 integration sites at the single RNA or DNA molecule in vivo. We track viral genetic diversity throughout early infection, ART, and rebound. The proviral reservoir retains genetic diversity despite cellular clonal proliferation and viral seeding by rebounding virus. Non-proliferated cell clones are likely the result of elimination of proviruses associated with transcriptional activation and viremia. Elimination of proviruses associated with viremia is less prominent among proliferated cell clones. Proliferated, but not massively expanded, cell clones contribute to proviral expansion and viremia, suggesting they fuel viral persistence. This approach enables comprehensive assessment of viral levels, lineages, integration sites, clonal proliferation and proviral epigenetic patterns in vivo. These findings highlight complex reservoir dynamics and the role of proliferated cell clones in viral persistence. 

Abstract Plant traits can be helpful for understanding grassland ecosystem responses to climate extremes, such as severe drought. However, intercontinental comparisons of how drought affects plant functional traits and ecosystem functioning are rare. The Extreme Drought in Grasslands experiment (EDGE) was established across the major grassland types in East Asia and North America (six sites on each continent) to measure variability in grassland ecosystem sensitivity to extreme, prolonged drought. At all sites, we quantified community‐weighted mean functional composition and functional diversity of two leaf economic traits, specific leaf area and leaf nitrogen content, in response to drought. We found that experimental drought significantly increased community‐weighted means of specific leaf area and leaf nitrogen content at all North American sites and at the wetter East Asian sites, but drought decreased community‐weighted means of these traits at moderate to dry East Asian sites. Drought significantly decreased functional richness but increased functional evenness and dispersion at most East Asian and North American sites. Ecosystem drought sensitivity (percentage reduction in aboveground net primary productivity) positively correlated with community‐weighted means of specific leaf area and leaf nitrogen content and negatively correlated with functional diversity (i.e., richness) on an intercontinental scale, but results differed within regions. These findings highlight both broad generalities but also unique responses to drought of community‐weighted trait means as well as their functional diversity across grassland ecosystems. 

We consider a non-adaptive controlled sensing sce-nario in which the actions of the decision maker are corrupted by an adversary. The objective of the decision maker is to either detect the presence of the corruption or make a correct decision. Accordingly, the performance of a controlled sensing strategy is measured in terms of the error probability when there is no adversary, denoted PE,0 , and the error probability when an adversary is present, denoted PE,1 . Our main result is Stein-lemma like characterization of the optimal achievable error exponent of PE,0 subject to a constraint on PE,1 . We also illustrate the result with numerical examples.