Search for: All records

Vision transformers (ViTs) have recently obtained success in many applications, but their intensive computation and heavy memory usage at both training and inference time limit their generalization. Previous compression algorithms usually start from the pre-trained dense models and only focus on efficient inference, while time-consuming training is still unavoidable. In contrast, this paper points out that the million-scale training data is redundant, which is the fundamental reason for the tedious training. To address the issue, this paper aims to introduce sparsity into data and proposes an end-to-end efficient training framework from three sparse perspectives, dubbed Tri-Level E-ViT. Specifically, we leverage a hierarchical data redundancy reduction scheme, by exploring the sparsity under three levels: number of training examples in the dataset, number of patches (tokens) in each example, and number of connections between tokens that lie in attention weights. With extensive experiments, we demonstrate that our proposed technique can noticeably accelerate training for various ViT architectures while maintaining accuracy. Remarkably, under certain ratios, we are able to improve the ViT accuracy rather than compromising it. For example, we can achieve 15.2% speedup with 72.6% (+0.4) Top-1 accuracy on Deit-T, and 15.7% speedup with 79.9% (+0.1) Top-1 accuracy on Deit-S. This proves the existence of data redundancy in ViT. Our code
is released at https://github.com/ZLKong/Tri-Level-ViT

 

Multimodal single-cell sequencing technologies provide unprecedented information on cellular heterogeneity from multiple layers of genomic readouts. However, joint analysis of two modalities without properly handling the noise often leads to overfitting of one modality by the other and worse clustering results than vanilla single-modality analysis. How to efficiently utilize the extra information from single cell multi-omics to delineate cell states and identify meaningful signal remains as a significant computational challenge. In this work, we propose a deep learning framework, named SAILERX, for efficient, robust, and flexible analysis of multi-modal single-cell data. SAILERX consists of a variational autoencoder with invariant representation learning to correct technical noises from sequencing process, and a multimodal data alignment mechanism to integrate information from different modalities. Instead of performing hard alignment by projecting both modalities to a shared latent space, SAILERX encourages the local structures of two modalities measured by pairwise similarities to be similar. This strategy is more robust against overfitting of noises, which facilitates various downstream analysis such as clustering, imputation, and marker gene detection. Furthermore, the invariant representation learning part enables SAILERX to perform integrative analysis on both multi- and single-modal datasets, making it an applicable and scalable tool for more general scenarios.

 

This paper presents a policy-driven sequential image augmentation approach for image-related tasks. Our approach applies a sequence of image transformations (e.g., translation, rotation) over a training image, one transformation at a time, with the augmented image from the previous time step treated as the input for the next transformation. This sequential data augmentation substantially improves sample diversity, leading to improved test performance, especially for data-hungry models (e.g., deep neural networks). However, the search for the optimal transformation of each image at each time step of the sequence has high complexity due to its combination nature. To address this challenge, we formulate the search task as a sequential decision process and introduce a deep policy network that learns to produce transformations based on image content. We also develop an iterative algorithm to jointly train a classifier and the policy network in the reinforcement learning setting. The immediate reward of a potential transformation is defined to encourage transformations producing hard samples for the current classifier. At each iteration, we employ the policy network to augment the training dataset, train a classifier with the augmented data, and train the policy net with the aid of the classifier. We apply the above approach to both public image classification benchmarks and a newly collected image dataset for material recognition. Comparisons to alternative augmentation approaches show that our policy-driven approach achieves comparable or improved classification performance while using significantly fewer augmented images. The code is available at https://github.com/Paul-LiPu/rl_autoaug. 

Abstract Motivation Single-cell sequencing assay for transposase-accessible chromatin (scATAC-seq) provides new opportunities to dissect epigenomic heterogeneity and elucidate transcriptional regulatory mechanisms. However, computational modeling of scATAC-seq data is challenging due to its high dimension, extreme sparsity, complex dependencies and high sensitivity to confounding factors from various sources. Results Here, we propose a new deep generative model framework, named SAILER, for analyzing scATAC-seq data. SAILER aims to learn a low-dimensional nonlinear latent representation of each cell that defines its intrinsic chromatin state, invariant to extrinsic confounding factors like read depth and batch effects. SAILER adopts the conventional encoder-decoder framework to learn the latent representation but imposes additional constraints to ensure the independence of the learned representations from the confounding factors. Experimental results on both simulated and real scATAC-seq datasets demonstrate that SAILER learns better and biologically more meaningful representations of cells than other methods. Its noise-free cell embeddings bring in significant benefits in downstream analyses: clustering and imputation based on SAILER result in 6.9% and 18.5% improvements over existing methods, respectively. Moreover, because no matrix factorization is involved, SAILER can easily scale to process millions of cells. We implemented SAILER into a software package, freely available to all for large-scale scATAC-seq data analysis. Availability and implementation The software is publicly available at https://github.com/uci-cbcl/SAILER. Supplementary information Supplementary data are available at Bioinformatics online. 

Characterizing genome-wide binding profiles of transcription factors (TFs) is essential for understanding biological processes. Although techniques have been developed to assess binding profiles within a population of cells, determining them at a single-cell level remains elusive. Here, we report scFAN (single-cell factor analysis network), a deep learning model that predicts genome-wide TF binding profiles in individual cells. scFAN is pretrained on genome-wide bulk assay for transposase-accessible chromatin sequencing (ATAC-seq), DNA sequence, and chromatin immunoprecipitation sequencing (ChIP-seq) data and uses single-cell ATAC-seq to predict TF binding in individual cells. We demonstrate the efficacy of scFAN by both studying sequence motifs enriched within predicted binding peaks and using predicted TFs for discovering cell types. We develop a new metric “TF activity score” to characterize each cell and show that activity scores can reliably capture cell identities. scFAN allows us to discover and study cellular identities and heterogeneity based on chromatin accessibility profiles.