Deep convolutional neural networks require large amounts of labeled data samples. For many real-world applications, this is a major limitation which is commonly treated by augmentation methods. In this work, we address the problem of learning deep neural networks on small datasets. Our proposed architecture called ChimeraMix learns a data augmentation by generating compositions of instances. The generative model encodes images in pairs, combines the features guided by a mask, and creates new samples. For evaluation, all methods are trained from scratch without any additional data. Several experiments on benchmark datasets, e.g. ciFAIR-10, STL-10, and ciFAIR-100, demonstrate the superior performance of ChimeraMix compared to current state-of-the-art methods for classification on small datasets.
TorchXRayVision is an open source software library for working with chest X-ray datasets and deep learning models. It provides a common interface and common pre-processing chain for a wide set of publicly available chest X-ray datasets. In addition, a number of classification and representation learning models with different architectures, trained on different data combinations, are available through the library to serve as baselines or feature extractors.
Data-efficient image classification using deep neural networks in settings, where only small amounts of labeled data are available, has been an active research area in the recent past. However, an objective comparison between published methods is difficult, since existing works use different datasets for evaluation and often compare against untuned baselines with default hyper-parameters. We design a benchmark for data-efficient image classification consisting of six diverse datasets spanning various domains (e.g., natural images, medical imagery, satellite data) and data types (RGB, grayscale, multispectral). Using this benchmark, we re-evaluate the standard cross-entropy baseline and eight methods for data-efficient deep learning published between 2017 and 2021 at renowned venues. For a fair and realistic comparison, we carefully tune the hyper-parameters of all methods on each dataset. Surprisingly, we find that tuning learning rate, weight decay, and batch size on a separate validation split results in a highly competitive baseline, which outperforms all but one specialized method and performs competitively to the remaining one.
The wavelet scattering transform creates geometric invariants and deformation stability. In multiple signal domains, it has been shown to yield more discriminative representations compared to other non-learned representations and to outperform learned representations in certain tasks, particularly on limited labeled data and highly structured signals. The wavelet filters used in the scattering transform are typically selected to create a tight frame via a parameterized mother wavelet. In this work, we investigate whether this standard wavelet filterbank construction is optimal. Focusing on Morlet wavelets, we propose to learn the scales, orientations, and aspect ratios of the filters to produce problem-specific parameterizations of the scattering transform. We show that our learned versions of the scattering transform yield significant performance gains in small-sample classification settings over the standard scattering transform. Moreover, our empirical results suggest that traditional filterbank constructions may not always be necessary for scattering transforms to extract effective representations.
Self-supervised learning methods for computer vision have demonstrated the effectiveness of pre-training feature representations, resulting in well-generalizing Deep Neural Networks, even if the annotated data are limited. However, representation learning techniques require a significant amount of time for model training, with most of the time spent on precise hyper-parameter optimization and selection of augmentation techniques. We hypothesized that if the annotated dataset has enough morphological diversity to capture the diversity of the general population, as is common in medical imaging due to conserved similarities of tissue morphology, the variance error of the trained model is the dominant component of the Bias-Variance Trade-off. Therefore, we proposed the Variance Aware Training (VAT) method that exploits this data property by introducing the variance error into the model loss function, thereby, explicitly regularizing the model. Additionally, we provided a theoretical formulation and proof of the proposed method to aid interpreting the approach. Our method requires selecting only one hyper-parameter and matching or improving the performance of state-of-the-art self-supervised methods while achieving an order of magnitude reduction in the GPU training time. We validated VAT on three medical imaging datasets from diverse domains and for various learning objectives. These included a Magnetic Resonance Imaging (MRI) dataset for the heart semantic segmentation (MICCAI 2017 ACDC challenge), fundus photography dataset for ordinary regression of diabetic retinopathy progression (Kaggle 2019 APTOS Blindness Detection challenge), and classification of histopathologic scans of lymph node sections (PatchCamelyon dataset). Our code is available at https://github.com/DmitriiShubin/Variance-Aware-Training.
Training neural networks with auxiliary tasks is a common practice for improving the performance on a main task of interest. Two main challenges arise in this multi-task learning setting: (i) designing useful auxiliary tasks; and (ii) combining auxiliary tasks into a single coherent loss. Here, we propose a novel framework, AuxiLearn, that targets both challenges based on implicit differentiation. First, when useful auxiliaries are known, we propose learning a network that combines all losses into a single coherent objective function. This network can learn non-linear interactions between tasks. Second, when no useful auxiliary task is known, we describe how to learn a network that generates a meaningful, novel auxiliary task. We evaluate AuxiLearn in a series of tasks and domains, including image segmentation and learning with attributes in the low data regime, and find that it consistently outperforms competing methods.
In this study, we propose an ensemble long short-term memory (EnLSTM) network, which can be trained on a small dataset and process sequential data. The EnLSTM is built by combining the ensemble neural network (ENN) and the cascaded long short-term memory (C-LSTM) network to leverage their complementary strengths. In order to resolve the issues of over-convergence and disturbance compensation associated with training failure owing to the nature of small-data problems, model parameter perturbation and high-fidelity observation perturbation methods are introduced. The EnLSTM is compared with commonly-used models on a published dataset, and proven to be the state-of-the-art model in generating well logs with a mean-square-error (MSE) reduction of 34%. In the case study, 12 well logs that cannot be measured while drilling are generated based on logging-while-drilling (LWD) data. The EnLSTM is capable to reduce cost and save time in practice.
We present a new approach, called meta-meta classification, to learning in small-data settings. In this approach, one uses a large set of learning problems to design an ensemble of learners, where each learner has high bias and low variance and is skilled at solving a specific type of learning problem. The meta-meta classifier learns how to examine a given learning problem and combine the various learners to solve the problem. The meta-meta learning approach is especially suited to solving few-shot learning tasks, as it is easier to learn to classify a new learning problem with little data than it is to apply a learning algorithm to a small data set. We evaluate the approach on a one-shot, one-class-versus-all classification task and show that it is able to outperform traditional meta-learning as well as ensembling approaches.
The possibility to use widespread and simple chest X-ray (CXR) imaging for early screening of COVID-19 patients is attracting much interest from both the clinical and the AI community. In this study we provide insights and also raise warnings on what is reasonable to expect by applying deep-learning to COVID classification of CXR images. We provide a methodological guide and critical reading of an extensive set of statistical results that can be obtained using currently available datasets. In particular, we take the challenge posed by current small size COVID data and show how significant can be the bias introduced by transfer-learning using larger public non-COVID CXR datasets. We also contribute by providing results on a medium size COVID CXR dataset, just collected by one of the major emergency hospitals in Northern Italy during the peak of the COVID pandemic. These novel data allow us to contribute to validate the generalization capacity of preliminary results circulating in the scientific community. Our conclusions shed some light into the possibility to effectively discriminate COVID using CXR.
We revisit the long-standing problem of learning from a small sample, to which end we propose a novel method called GLICO (Generative Latent Implicit Conditional Optimization). GLICO learns a mapping from the training examples to a latent space and a generator that generates images from vectors in the latent space. Unlike most recent works, which rely on access to large amounts of unlabeled data, GLICO does not require access to any additional data other than the small set of labeled points. In fact, GLICO learns to synthesize completely new samples for every class using as little as 5 or 10 examples per class, with as few as 10 such classes without imposing any prior. GLICO is then used to augment the small training set while training a classifier on the small sample. To this end, our proposed method samples the learned latent space using spherical interpolation, and generates new examples using the trained generator. Empirical results show that the new sampled set is diverse enough, leading to improvement in image classification in comparison with the state of the art, when trained on small samples obtained from CIFAR-10, CIFAR-100, and CUB-200.
In this paper we challenge the common assumption that convolutional layers in modern CNNs are translation invariant. We show that CNNs can and will exploit the absolute spatial location by learning filters that respond exclusively to particular absolute locations by exploiting image boundary effects. Because modern CNNs filters have a huge receptive field, these boundary effects operate even far from the image boundary, allowing the network to exploit absolute spatial location all over the image. We give a simple solution to remove spatial location encoding which improves translation invariance and thus gives a stronger visual inductive bias which particularly benefits small data sets. We broadly demonstrate these benefits on several architectures and various applications such as image classification, patch matching, and two video classification datasets.
This paper studies the problem of post-hoc calibration of machine learning classifiers. We introduce the following desiderata for uncertainty calibration: (a) accuracy-preserving, (b) data-efficient, and (c) high expressive power. We show that none of the existing methods satisfy all three requirements, and demonstrate how Mix-n-Match calibration strategies (i.e., ensemble and composition) can help achieve remarkably better data-efficiency and expressive power while provably maintaining the classification accuracy of the original classifier. Mix-n-Match strategies are generic in the sense that they can be used to improve the performance of any off-the-shelf calibrator. We also reveal potential issues in standard evaluation practices. Popular approaches (e.g., histogram-based expected calibration error (ECE)) may provide misleading results especially in small-data regime. Therefore, we propose an alternative data-efficient kernel density-based estimator for a reliable evaluation of the calibration performance and prove its asymptotically unbiasedness and consistency. Our approaches outperform state-of-the-art solutions on both the calibration as well as the evaluation tasks in most of the experimental settings. Our codes are available at https://github.com/zhang64-llnl/Mix-n-Match-Calibration.
Cross-domain object detection is challenging, because object detection model is often vulnerable to data variance, especially to the considerable domain shift between two distinctive domains. In this paper, we propose a new Unbiased Mean Teacher (UMT) model for cross-domain object detection. We reveal that there often exists a considerable model bias for the simple mean teacher (MT) model in cross-domain scenarios, and eliminate the model bias with several simple yet highly effective strategies. In particular, for the teacher model, we propose a cross-domain distillation method for MT to maximally exploit the expertise of the teacher model. Moreover, for the student model, we alleviate its bias by augmenting training samples with pixel-level adaptation. Finally, for the teaching process, we employ an out-of-distribution estimation strategy to select samples that most fit the current model to further enhance the cross-domain distillation process. By tackling the model bias issue with these strategies, our UMT model achieves mAPs of 44.1%, 58.1%, 41.7%, and 43.1% on benchmark datasets Clipart1k, Watercolor2k, Foggy Cityscapes, and Cityscapes, respectively, which outperforms the existing state-of-the-art results in notable margins. Our implementation is available at https://github.com/kinredon/umt.
Extracting large amounts of data from biological samples is not feasible due to radiation issues, and image processing in the small-data regime is one of the critical challenges when working with a limited amount of data. In this work, we applied an existing algorithm named Variational Auto Encoder (VAE) that pre-trains a latent space representation of the data to capture the features in a lower-dimension for the small-data regime input. The fine-tuned latent space provides constant weights that are useful for classification. Here we will present the performance analysis of the VAE algorithm with different latent space sizes in the semi-supervised learning using the CIFAR-10 dataset.
We present an efficient coreset construction algorithm for large-scale Support Vector Machine (SVM) training in Big Data and streaming applications. A coreset is a small, representative subset of the original data points such that a models trained on the coreset are provably competitive with those trained on the original data set. Since the size of the coreset is generally much smaller than the original set, our preprocess-then-train scheme has potential to lead to significant speedups when training SVM models. We prove lower and upper bounds on the size of the coreset required to obtain small data summaries for the SVM problem. As a corollary, we show that our algorithm can be used to extend the applicability of any off-the-shelf SVM solver to streaming, distributed, and dynamic data settings. We evaluate the performance of our algorithm on real-world and synthetic data sets. Our experimental results reaffirm the favorable theoretical properties of our algorithm and demonstrate its practical effectiveness in accelerating SVM training.
Gaussian processes (GPs) provide flexible distributions over functions, with inductive biases controlled by a kernel. However, in many applications Gaussian processes can struggle with even moderate input dimensionality. Learning a low dimensional projection can help alleviate this curse of dimensionality, but introduces many trainable hyperparameters, which can be cumbersome, especially in the small data regime. We use additive sums of kernels for GP regression, where each kernel operates on a different random projection of its inputs. Surprisingly, we find that as the number of random projections increases, the predictive performance of this approach quickly converges to the performance of a kernel operating on the original full dimensional inputs, over a wide range of data sets, even if we are projecting into a single dimension. As a consequence, many problems can remarkably be reduced to one dimensional input spaces, without learning a transformation. We prove this convergence and its rate, and additionally propose a deterministic approach that converges more quickly than purely random projections. Moreover, we demonstrate our approach can achieve faster inference and improved predictive accuracy for high-dimensional inputs compared to kernels in the original input space.
Data-based discovery of effective, coarse-grained (CG) models of high-dimensional dynamical systems presents a unique challenge in computational physics and particularly in the context of multiscale problems. The present paper offers a data-based, probablistic perspective that enables the quantification of predictive uncertainties. One of the outstanding problems has been the introduction of physical constraints in the probabilistic machine learning objectives. The primary utility of such constraints stems from the undisputed physical laws such as conservation of mass, energy etc. that they represent. Furthermore and apart from leading to physically realistic predictions, they can significantly reduce the requisite amount of training data which for high-dimensional, multiscale systems are expensive to obtain (Small Data regime). We formulate the coarse-graining process by employing a probabilistic state-space model and account for the aforementioned equality constraints as virtual observables in the associated densities. We demonstrate how probabilistic inference tools can be employed to identify the coarse-grained variables in combination with deep neural nets and their evolution model without ever needing to define a fine-to-coarse (restriction) projection and without needing time-derivatives of state variables. Furthermore, it is capable of reconstructing the evolution of the full, fine-scale system and therefore the observables of interest need not be selected a priori. We demonstrate the efficacy of the proposed framework by applying it to systems of interacting particles and an image-series of a nonlinear pendulum.
For many NLP applications, such as question answering and summarisation, the goal is to select the best solution from a large space of candidates to meet a particular user's needs. To address the lack of user-specific training data, we propose an interactive text ranking approach that actively selects pairs of candidates, from which the user selects the best. Unlike previous strategies, which attempt to learn a ranking across the whole candidate space, our method employs Bayesian optimisation to focus the user's labelling effort on high quality candidates and integrates prior knowledge in a Bayesian manner to cope better with small data scenarios. We apply our method to community question answering (cQA) and extractive summarisation, finding that it significantly outperforms existing interactive approaches. We also show that the ranking function learned by our method is an effective reward function for reinforcement learning, which improves the state of the art for interactive summarisation.
Bayesian neural networks (BNN) are the probabilistic model that combines the strengths of both neural network (NN) and stochastic processes. As a result, BNN can combat overfitting and perform well in applications where data is limited. Earthquake rupture study is such a problem where data is insufficient, and scientists have to rely on many trial and error numerical or physical models. Lack of resources and computational expenses, often, it becomes hard to determine the reasons behind the earthquake rupture. In this work, a BNN has been used (1) to combat the small data problem and (2) to find out the parameter combinations responsible for earthquake rupture and (3) to estimate the uncertainty associated with earthquake rupture. Two thousand rupture simulations are used to train and test the model. A simple 2D rupture geometry is considered where the fault has a Gaussian geometric heterogeneity at the center, and eight parameters vary in each simulation. The test F1-score of BNN (0.8334), which is 2.34% higher than plain NN score. Results show that the parameters of rupture propagation have higher uncertainty than the rupture arrest. Normal stresses play a vital role in determining rupture propagation and are also the highest source of uncertainty, followed by the dynamic friction coefficient. Shear stress has a moderate role, whereas the geometric features such as the width and height of the fault are least significant and uncertain.
In drug-discovery-related tasks such as virtual screening, machine learning is emerging as a promising way to predict molecular properties. Conventionally, molecular fingerprints (numerical representations of molecules) are calculated through rule-based algorithms that map molecules to a sparse discrete space. However, these algorithms perform poorly for shallow prediction models or small datasets. To address this issue, we present SMILES Transformer. Inspired by Transformer and pre-trained language models from natural language processing, SMILES Transformer learns molecular fingerprints through unsupervised pre-training of the sequence-to-sequence language model using a huge corpus of SMILES, a text representation system for molecules. We performed benchmarks on 10 datasets against existing fingerprints and graph-based methods and demonstrated the superiority of the proposed algorithms in small-data settings where pre-training facilitated good generalization. Moreover, we define a novel metric to concurrently measure model accuracy and data efficiency.