Semi-supervised learning has proven to be a powerful paradigm for leveraging unlabeled data to mitigate the reliance on large labeled datasets. In this work, we unify the current dominant approaches for semi-supervised learning to produce a new algorithm, MixMatch, that works by guessing low-entropy labels for data-augmented unlabeled examples and mixing labeled and unlabeled data using MixUp. We show that MixMatch obtains state-of-the-art results by a large margin across many datasets and labeled data amounts. For example, on CIFAR-10 with 250 labels, we reduce error rate by a factor of 4 (from 38% to 11%) and by a factor of 2 on STL-10. We also demonstrate how MixMatch can help achieve a dramatically better accuracy-privacy trade-off for differential privacy. Finally, we perform an ablation study to tease apart which components of MixMatch are most important for its success.

Unsupervised domain adaptation (UDA) amounts to assigning class labels to the unlabeled instances of a dataset from a target domain, using labeled instances of a dataset from a related source domain. In this paper, we propose to cast this problem in a game-theoretic setting as a non-cooperative game and introduce a fully automatized iterative algorithm for UDA based on graph transduction games (GTG). The main advantages of this approach are its principled foundation, guaranteed termination of the iterative algorithms to a Nash equilibrium (which corresponds to a consistent labeling condition) and soft labels quantifying the uncertainty of the label assignment process. We also investigate the beneficial effect of using pseudo-labels from linear classifiers to initialize the iterative process. The performance of the resulting methods is assessed on publicly available object recognition benchmark datasets involving both shallow and deep features. Results of experiments demonstrate the suitability of the proposed game-theoretic approach for solving UDA tasks.

Reinforcement learning usually makes use of numerical rewards, which have nice properties but also come with drawbacks and difficulties. Using rewards on an ordinal scale (ordinal rewards) is an alternative to numerical rewards that has received more attention in recent years. In this paper, a general approach to adapting reinforcement learning problems to the use of ordinal rewards is presented and motivated. We show how to convert common reinforcement learning algorithms to an ordinal variation by the example of Q-learning and introduce Ordinal Deep Q-Networks, which adapt deep reinforcement learning to ordinal rewards. Additionally, we run evaluations on problems provided by the OpenAI Gym framework, showing that our ordinal variants exhibit a performance that is comparable to the numerical variations for a number of problems. We also give first evidence that our ordinal variant is able to produce better results for problems with less engineered and simpler-to-design reward signals.

Multiple Instance Learning (MIL) is a weak supervision learning paradigm that allows modeling of machine learning problems in which labels are available only for groups of examples called bags. A positive bag may contain one or more positive examples but it is not known which examples in the bag are positive. All examples in a negative bag belong to the negative class. Such problems arise frequently in fields of computer vision, medical image processing and bioinformatics. Many neural network based solutions have been proposed in the literature for MIL, however, almost all of them rely on introducing specialized blocks and connectivity in the architectures. In this paper, we present a novel and effective approach to Multiple Instance Learning in neural networks. Instead of making changes to the architectures, we propose a simple bag-level ranking loss function that allows Multiple Instance Classification in any neural architecture. We have demonstrated the effectiveness of our proposed method for popular MIL benchmark datasets. In addition, we have tested the performance of our method in convolutional neural networks used to model an MIL problem derived from the well-known MNIST dataset. Results have shown that despite being simpler, our proposed scheme is comparable or better than existing methods in the literature in practical scenarios. Python code files for all the experiments can be found at https://github.com/amina01/ESMIL.

Missing data imputation (MDI) is a fundamental problem in many scientific disciplines. Popular methods for MDI use global statistics computed from the entire data set (e.g., the feature-wise medians), or build predictive models operating independently on every instance. In this paper we propose a more general framework for MDI, leveraging recent work in the field of graph neural networks (GNNs). We formulate the MDI task in terms of a graph denoising autoencoder, where each edge of the graph encodes the similarity between two patterns. A GNN encoder learns to build intermediate representations for each example by interleaving classical projection layers and locally combining information between neighbors, while another decoding GNN learns to reconstruct the full imputed data set from this intermediate embedding. In order to speed-up training and improve the performance, we use a combination of multiple losses, including an adversarial loss implemented with the Wasserstein metric and a gradient penalty. We also explore a few extensions to the basic architecture involving the use of residual connections between layers, and of global statistics computed from the data set to improve the accuracy. On a large experimental evaluation, we show that our method robustly outperforms state-of-the-art approaches for MDI, especially for large percentages of missing values.

The application of deep learning to early detection and automated classification of Alzheimer's disease (AD) has recently gained considerable attention as rapid progress in neuroimaging techniques has generated large-scale multimodal neuroimaging data. Here we systematically reviewed publications, where deep learning approaches and neuroimaging data were used for diagnostic classification of AD. A PubMed and google scholar search was performed to find deep learning papers for AD published between January 2013 and July 2018, which were reviewed, evaluated, and classified by algorithms and neuroimaging types, and findings were summarized. The diagnostic classification of AD using deep learning approaches and neuroimaging data was examined in 16 studies. The approach to combine traditional machine learning for classification and stacked auto-encoder (SAE) for feature selection has produced accuracies of up to 98.8% for AD classification and 83.7% for prediction of conversion from mild cognitive impairment (MCI), a prodromal stage of AD, to AD. Deep learning approaches such as convolutional neural network (CNN) or recurrent neural network (RNN) using neuroimaging data without preprocessing for feature selection have yielded accuracies of up to 96.0% for AD classification and 84.2% for MCI conversion prediction. Furthermore, the best classification performance was obtained when multimodal neuroimaging data as well as fluid biomarkers were integrated. Deep learning approaches without preprocessing neuroimaging data for feature selection, a major bottleneck of traditional machining learning in high-dimensional data, continue to improve their performance and to show great promise in the diagnostic classification of AD using multimodal neuroimaging data.

Feature engineering and feature extraction are key -- and time-consuming -- parts of the machine learning workflow. They are about transforming training data and augmenting it with additional features in order to make machine learning algorithms more effective. Deep learning is changing that, according to its promoters. With deep learning, one can start with raw data, as features will be automatically created by the neural network when it learns. As usual with bold statements, this is both true and false.

What is the problem we are trying to solve with attention? There is a bottleneck at the end of encoder RNN. Gets worse the longer the sentence. Get dot product of encoder hidden states and first decoder hidden state (green). The more aligned (similar), the bigger the dot product.

It is very well known that the most fundamental unit of deep neural networks is called an artificial neuron/perceptron. But the very first step towards the perceptron we use today was taken in 1943 by McCulloch and Pitts, by mimicking the functionality of a biological neuron. Note: The concept, the content, and the structure of this article were largely based on the awesome lectures and the material offered by Prof. Mitesh M. Khapra on NPTEL's Deep Learning course. Basically, a neuron takes an input signal (dendrite), processes it like the CPU (soma), passes the output through a cable like structure to other connected neurons (axon to synapse to other neuron's dendrite). Now, this might be biologically inaccurate as there is a lot more going on out there but on a higher level, this is what is going on with a neuron in our brain -- takes an input, processes it, throws out an output.

AI can now write realistic-sounding text, give debating champs a run for their money, diagnose illnesses, and generate fake human faces--among much more. After training these systems on massive datasets, their creators essentially just let them do their thing to arrive at certain conclusions or outcomes. The problem is that more often than not, even the creators don't know exactly why they've arrived at those conclusions or outcomes. There's no easy way to trace a machine learning system's rationale, so to speak. The further we let AI go down this opaque path, the more likely we are to end up somewhere we don't want to be--and may not be able to come back from.