This post will help you set up a GPU enabled Docker container on an AWS EC2 instance for Deep Learning. We'll also cover how to access a Jupyter server running inside the container from your local machine. One of my favourite things about Docker is that it allows you to easily reproduce your local development environment on a remote machine. This comes in real handy whenever you need to train a model on a GPU. This post assumes some familiarity with building Docker images and launching EC2 instances on AWS.

Apache Spark is a widely used technology for data processing and is used heavily by machine learning users. Spark can be used to classify products, forecast demand, and personalize recommendations. While Spark supports a variety of programming languages, the preferred Spark SDK is implemented for Scala, which is not well supported by most deep learning frameworks. Most machine learning frameworks favor Python with their SDKs, leaving Spark developers with suboptimal options: porting their code to Python or implementing a custom Scala wrapper. These options impact developer velocity and threaten production environments with brittle code.

OpenAI, the machine learning nonprofit co-founded by Elon Musk, has released its first commercial product: a rentable version of a text generation tool the organisation once deemed too dangerous to release. Dubbed simply "the API", the new service lets businesses directly access the most powerful version of GPT-3, OpenAI's general purpose text generation AI. The tool is already a more than capable writer. Feeding an earlier version of the opening line of George Orwell's Nineteen Eighty-Four – "It was a bright cold day in April, and the clocks were striking thirteen" – the system recognises the vaguely futuristic tone and the novelistic style, and continues with: "I was in my car on my way to a new job in Seattle. I put the gas in, put the key in, and then I let it run. I just imagined what the day would be like. In 2045, I was a teacher in some school in a poor part of rural China. I started with Chinese history and history of science."

Identifying sleep problem severity from overnight polysomnography (PSG) recordings plays an important role in diagnosing and treating sleep disorders such as the Obstructive Sleep Apnea (OSA). This analysis traditionally is done by specialists manually through visual inspections, which can be tedious, time-consuming, and is prone to subjective errors. One of the solutions is to use Convolutional Neural Networks (CNN) where the convolutional and pooling layers behave as feature extractors and some fully-connected (FCN) layers are used for making final predictions for the OSA severity. In this paper, a CNN architecture with 1D convolutional and FCN layers for classification is presented. The PSG data for this project are from the Cleveland Children's Sleep and Health Study database and classification results confirm the effectiveness of the proposed CNN method. The proposed 1D CNN model achieves excellent classification results without manually preprocesssing PSG signals such as feature extraction and feature reduction.

In this work, we propose MixMOOD - a systematic approach to mitigate effect of class distribution mismatch in semi-supervised deep learning (SSDL) with MixMatch. This work is divided into two components: (i) an extensive out of distribution (OOD) ablation test bed for SSDL and (ii) a quantitative unlabelled dataset selection heuristic referred to as MixMOOD. In the first part, we analyze the sensitivity of MixMatch accuracy under 90 different distribution mismatch scenarios across three multi-class classification tasks. These are designed to systematically understand how OOD unlabelled data affects MixMatch performance. In the second part, we propose an efficient and effective method, called deep dataset dissimilarity measures (DeDiMs), to compare labelled and unlabelled datasets. The proposed DeDiMs are quick to evaluate and model agnostic. They use the feature space of a generic Wide-ResNet and can be applied prior to learning. Our test results reveal that supposed semantic similarity between labelled and unlabelled data is not a good heuristic for unlabelled data selection. In contrast, strong correlation between MixMatch accuracy and the proposed DeDiMs allow us to quantitatively rank different unlabelled datasets ante hoc according to expected MixMatch accuracy. This is what we call MixMOOD. Furthermore, we argue that the MixMOOD approach can aid to standardize the evaluation of different semi-supervised learning techniques under real world scenarios involving out of distribution data.

Mathematical proofs can be mechanised using proof assistants to eliminate gaps and errors. However, mechanisation still requires intensive labour. To promote automation, it is essential to capture high-level human mathematical reasoning, which we address as the problem of generating suitable propositions. We build a non-synthetic dataset from the largest repository of mechanised proofs and propose a task on causal reasoning, where a model is required to fill in a missing intermediate proposition given a causal context. Our experiments (using various neural sequence-to-sequence models) reveal that while the task is challenging, neural models can indeed capture non-trivial mathematical reasoning. We further propose a hierarchical transformer model that outperforms the transformer baseline.

Several works have proposed Simplicity Bias (SB)---the tendency of standard training procedures such as Stochastic Gradient Descent (SGD) to find simple models---to justify why neural networks generalize well [Arpit et al. 2017, Nakkiran et al. 2019, Valle-Perez et al. 2019]. However, the precise notion of simplicity remains vague. Furthermore, previous settings that use SB to justify why neural networks generalize well do not simultaneously capture the brittleness of neural networks---a widely observed phenomenon in practice [Goodfellow et al. 2014, Jo and Bengio 2017]. To this end, we introduce a collection of piecewise-linear and image-based datasets that (a) naturally incorporate a precise notion of simplicity and (b) capture the subtleties of neural networks trained on real datasets. Through theory and experiments on these datasets, we show that SB of SGD and variants is extreme: neural networks rely exclusively on the simplest feature and remain invariant to all predictive complex features. Consequently, the extreme nature of SB explains why seemingly benign distribution shifts and small adversarial perturbations significantly degrade model performance. Moreover, contrary to conventional wisdom, SB can also hurt generalization on the same data distribution, as SB persists even when the simplest feature has less predictive power than the more complex features. We also demonstrate that common approaches for improving generalization and robustness---ensembles and adversarial training---do not mitigate SB and its shortcomings. Given the central role played by SB in generalization and robustness, we hope that the datasets and methods in this paper serve as an effective testbed to evaluate novel algorithmic approaches aimed at avoiding the pitfalls of extreme SB.

The prevalence of location-based services contributes to the explosive growth of individual-level trajectory data and raises public concerns about privacy issues. In this research, we propose a novel LSTM-TrajGAN approach, which is an end-to-end deep learning model to generate privacy-preserving synthetic trajectory data for data sharing and publication. We design a loss metric function TrajLoss to measure the trajectory similarity losses for model training and optimization. The model is evaluated on the trajectory-user-linking task on a real-world semantic trajectory dataset. Compared with other common geomasking methods, our model can better prevent users from being re-identified, and it also preserves essential spatial, temporal, and thematic characteristics of the real trajectory data. The model better balances the effectiveness of trajectory privacy protection and the utility for spatial and temporal analyses, which offers new insights into the GeoAI-powered privacy protection.

Recent progress in quantum algorithms and hardware indicates the potential importance of quantum computing in the near future. However, finding suitable application areas remains an active area of research. Quantum machine learning is touted as a potential approach to demonstrate quantum advantage within both the gate-model and the adiabatic schemes. For instance, the Quantum-assisted Variational Autoencoder has been proposed as a quantum enhancement to the discrete VAE. We extend on previous work and study the real-world applicability of a QVAE by presenting a proof-of-concept for similarity search in large-scale high-dimensional datasets. While exact and fast similarity search algorithms are available for low dimensional datasets, scaling to high-dimensional data is non-trivial. We show how to construct a space-efficient search index based on the latent space representation of a QVAE. Our experiments show a correlation between the Hamming distance in the embedded space and the Euclidean distance in the original space on the Moderate Resolution Imaging Spectroradiometer (MODIS) dataset. Further, we find real-world speedups compared to linear search and demonstrate memory-efficient scaling to half a billion data points.

A key challenge in understanding the sensory transformations of the visual system is to obtain a highly predictive model of responses from visual cortical neurons. Deep neural networks (DNNs) provide a promising candidate for such a model. However, DNNs require orders of magnitude more training data than neuroscientists can collect from real neurons because experimental recording time is severely limited. This motivates us to find images that train highly-predictive DNNs with as little training data as possible. We propose gaudy images---high-contrast binarized versions of natural images---to efficiently train DNNs. In extensive simulation experiments, we find that training DNNs with gaudy images substantially reduces the number of training images needed to accurately predict the simulated responses of visual cortical neurons. We also find that gaudy images, chosen before training, outperform images chosen during training by active learning algorithms. Thus, gaudy images overemphasize features of natural images, especially edges, that are the most important for efficiently training DNNs. We believe gaudy images will aid in the modeling of visual cortical neurons, potentially opening new scientific questions about visual processing, as well as aid general practitioners that seek ways to improve the training of DNNs.