In the past decade, convolutional neural networks (CNNs) have been widely adopted as the main building block for endto-end audio classification models, which aim to learn a direct mapping from audio spectrograms to corresponding labels. To better capture long-range global context, a recent trend is to add a self-attention mechanism on top of the CNN, forming a CNN-attention hybrid model. However, it is unclear whether the reliance on a CNN is necessary, and if neural networks purely based on attention are sufficient to obtain good performance in audio classification. In this paper, we answer the question by introducing the Audio Spectrogram Transformer (AST), the first convolution-free, purely attention-based model for audio classification. We evaluate AST on various audio classification benchmarks, where it achieves new state-of-the-art results of 0.485 mAP on AudioSet, 95.6% accuracy on ESC-50, and 98.1% accuracy on Speech Commands V2.

No abstract

The relationship between the expression of particular genes in cells and their impact on phenotypic characteristics is important for understanding how cells regulate responses to their environment. We have developed a microwell-based method to detect copies of mRNA transcripts directly from individual cells by one-step, single-cell, reverse transcription polymerase chain reaction (RT-PCR). Our approach permits the detection of mRNA transcripts of interest for more than 6000 single cells in parallel per assay with high sensitivity and specificity for constitutively active genes. This simple method was also combined with microengraving and image-based cytometry to examine the relationships between gene expression and cellular secretion of antibodies in a clonal population. We observed that most individual human B cell hybridomas transcribed a requisite gene for their antibodies, but only a subset of those cells secreted the antibody. The technique should also allow the detection of replicating intracellular pathogens such as retroviruses.

Audio event classification is an active research area and has a wide range of applications. Since the release of AudioSet, great progress has been made in advancing the classification accuracy, which mostly comes from the development of novel model architectures and attention modules. However, we find that appropriate training techniques are equally important for building audio event classification models with AudioSet, but have not received the attention they deserve. To fill the gap, in this work, we present PSLA, a collection of training techniques that can noticeably boost the model accuracy including ImageNet pretraining, balanced sampling, data augmentation, label enhancement, model aggregation and their design choices.By training an EfficientNet with these techniques, we obtain a model that achieves a new state-of-the-art mean average precision (mAP) of 0.474 on AudioSet, outperforming the previous best system of 0.439.

We present here a new method to enhance the detection of secreted cytokines and chemokines from single human mononuclear cells. The technique uses a hybridization chain reaction (HCR) to amplify signals resulting from sandwich immunoassays. This immuno-HCR employs oligonucleotide-based initiators covalently linked to antibodies to propagate a chain reaction of hybridization events involving a pair of complementary hairpin oligomers bearing fluorescent labels. Integrating this strategy for signal amplification with microengraving—a soft lithographic method for printing arrays of secreted proteins from thousands of single cells—improves both the limits of detection and sensitivity for cytokines and chemokines captured from individual cells by an average of 200-fold relative to methods for direct detection by fluoresence. This approach should enhance the utility of microengraving for defining the immunological signatures of diseases and responses to interventional therapies based on multiplexed single-cell analysis.