Quantifying the value of data is a fundamental problem in machine learning. Data valuation has multiple important use cases: (1) building insights about the learning task, (2) domain adaptation, (3) corrupted sample discovery, and (4) robust learning. To adaptively learn data values jointly with the target task predictor model, we propose a meta learning framework which we name Data Valuation using Reinforcement Learning (DVRL). We employ a data value estimator (modeled by a deep neural network) to learn how likely each datum is used in training of the predictor model. We train the data value estimator using a reinforcement signal of the reward obtained on a small validation set that reflects performance on the target task. We demonstrate that DVRL yields superior data value estimates compared to alternative methods across different types of datasets and in a diverse set of application scenarios. The corrupted sample discovery performance of DVRL is close to optimal in many regimes (i.e. as if the noisy samples were known apriori), and for domain adaptation and robust learning DVRL significantly outperforms state-of-the-art by 14.6% and 10.8%, respectively.

In this work we address the problem of transferring knowledge obtained from a vast annotated source domain to a low labeled or unlabeled target domain. We propose Adversarial Variational Domain Adaptation (AVDA), a semi-supervised domain adaptation method based on deep variational embedded representations. We use approximate inference and adversarial methods to map samples from source and target domains into an aligned semantic embedding. We show that on a semi-supervised few-shot scenario, our approach can be used to obtain a significant speed-up in performance when using an increasing number of labels on the target domain.

ROBEL is an open-source platform of cost-effective robots designed for reinforcement learning in the real world. ROBEL introduces two robots, each aimed to accelerate reinforcement learning research in different task domains: D'Claw is a three-fingered hand robot that facilitates learning dexterous manipulation tasks, and D'Kitty is a four-legged robot that facilitates learning agile legged locomotion tasks. These low-cost, modular robots are easy to maintain and are robust enough to sustain on-hardware reinforcement learning from scratch with over 14000 training hours registered on them to date. To leverage this platform, we propose an extensible set of continuous control benchmark tasks for each robot. These tasks feature dense and sparse task objectives, and additionally introduce score metrics as hardware-safety. We provide benchmark scores on an initial set of tasks using a variety of learning-based methods. Furthermore, we show that these results can be replicated across copies of the robots located in different institutions. Code, documentation, design files, detailed assembly instructions, final policies, baseline details, task videos, and all supplementary materials required to reproduce the results are available at www.roboticsbenchmarks.org.

Rapid intensification (RI) is when a sudden and considerable increase in tropical cyclone (TC) intensity occurs. Accurate early prediction of RI from TC images is important for preventing the possible damages caused by TCs. The main difficulty of RI prediction is to extract important features that are effective for RI prediction, which is challenging even for experienced meteorologists. Inspired by the success of deep learning models for automatic feature extraction and strong predictive performance, we initiate this study that experiments with multiple domain-knowledge guided deep learning models. The goal is to evaluate the potential use of these models for RI prediction. Furthermore, we examine the internal states of the models to obtain visualizable insights for RI prediction. Our model is efficient in training while achieving state-of-the-art performance on the benchmark dataset on HSS metric. The results showcase the success of adapting deep learning to solve complex meteorology problems.

Deep Graph Neural Networks (GNNs) are instrumental in graph classification and graph-based regression tasks. In these tasks, graph pooling is a critical ingredient by which GNNs adapt to input graphs of varying size and structure. We propose a new graph pooling operation based on compressive Haar transforms, called HaarPooling. HaarPooling is computed following a chain of sequential clusterings of the input graph. The input of each pooling layer is transformed by the compressive Haar basis of the corresponding clustering. HaarPooling operates in the frequency domain by the synthesis of nodes in the same cluster and filters out fine detail information by compressive Haar transforms. Such transforms provide an effective characterization of the data and preserve the structure information of the input graph. By the sparsity of the Haar basis, the computation of HaarPooling is of linear complexity. The GNN with HaarPooling and existing graph convolution layers achieves state-of-the-art performance on diverse graph classification problems.

While the process of feature learning by deep neural networks is still poorly understood, numerous techniques have been developed for visualizing these features in trained networks in an attempt to better understand the learning process [10, 29, 26, 21]. This is usually achieved by optimizing over the input space in order to maximize the activation of a single neuron or filter, with appropriate regularization. These techniques are commonly applied to pre-trained state-of-the-art models, yet they can also be used to explore the effects of various architectural choices on the learned representations. In this work, we specifically investigate the effect of the number of neurons on the quality of the learned representations. It has been shown that increasing the width of a neural network will generally not affect performance on a given task or even improve it [23, 13], and it is natural to study the effect of this modification on the learned representations. We use the recently developed activation atlases technique [3] in order to visualize the inputs that activate an entire hidden state, as opposed to a single neuron. In the case of a wide network, we find that the resulting visualizations differ dramatically. Our main findings are: - As we increase the number of neurons in a network, the features that an individual neuron responds to appear less like natural images. However, there exist directions in the hidden state space that are activated by natural-looking images.

Overparameterized transformer networks have obtained state of the art results in various natural language processing tasks, such as machine translation, language modeling, and question answering. These models contain hundreds of millions of parameters, necessitating a large amount of computation and making them prone to overfitting. In this work, we explore LayerDrop, a form of structured dropout, which has a regularization effect during training and allows for efficient pruning at inference time. In particular, we show that it is possible to select sub-networks of any depth from one large network without having to finetune them and with limited impact on performance. We demonstrate the effectiveness of our approach by improving the state of the art on machine translation, language modeling, summarization, question answering, and language understanding benchmarks. Moreover, we show that our approach leads to small BERT-like models of higher quality compared to training from scratch or using distillation.

Pairwise Choice Markov Chains (PCMC) have been recently introduced to overcome limitations of choice models based on traditional axioms unable to express empirical observations from modern behavior economics like framing effects and asymmetric dominance. The inference approach that estimates the transition rates between each possible pair of alternatives via maximum likelihood suffers when the examples of each alternative are scarce and is inappropriate when new alternatives can be observed at test time. In this work, we propose an amortized inference approach for PCMC by embedding its definition into a neural network that represents transition rates as a function of the alternatives' and individual's features. We apply our construction to the complex case of airline itinerary booking where singletons are common (due to varying prices and individual-specific itineraries), and asymmetric dominance and behaviors strongly dependent on market segments are observed. Experiments show our network significantly outperforming, in terms of prediction accuracy and logarithmic loss, feature engineered standard and latent class Multinomial Logit models as well as recent machine learning approaches.

It has been widely recognized that adversarial examples can be easily crafted to fool deep networks, which mainly root from the locally non-linear behavior nearby input examples. Applying mixup in training provides an effective mechanism to improve generalization performance and model robustness against adversarial perturbations, which introduces the globally linear behavior in-between training examples. However, in previous work, the mixup-trained models only passively defend adversarial attacks in inference by directly classifying the inputs, where the induced global linearity is not well exploited. Namely, since the locality of the adversarial perturbations, it would be more efficient to actively break the locality via the globality of the model predictions. Inspired by simple geometric intuition, we develop an inference principle, named mixup inference (MI), for mixup-trained models. MI mixups the input with other random clean samples, which can shrink and transfer the equivalent perturbation if the input is adversarial. Our experiments on CIFAR-10 and CIFAR-100 demonstrate that MI can further improve the adversarial robustness for the models trained by mixup and its variants.

Regularising Deep Networks with DGMsMatthew Willetts, Alexander Camuto, Stephen Roberts, Chris Holmes University of Oxford Alan Turing Institute {mwilletts, acamuto, sroberts, cholmes }@turing.ac.uk Abstract Here we develop a new method for regularising neural networks where we learn a density estimator over the activations of all layers of the model. We extend recent work in data imputation using V AEs (Ivanov et al., 2018) so that we can obtain a posterior for an arbitrary subset of activations conditioned on the remainder. Our method has links both to dropout and to data augmentation. We demonstrate that our training method leads to lower cross-entropy test set loss for 2-hidden-layer neural networks trained on CIFAR-10 and SVHN compared to standard regularisation baselines, but our model does not improve test-set accuracy compared to our baselines. This implies that although decisions are broadly similar, our approach provides a network with better calibrated uncertainty measures over the class posteriors. 1 Introduction Methods such a dropout [1], batch norm [2], L 2 regularisation [3] and data augmentation [3, 4] improve generalisation and robustness of deep discriminative models p Ψ(y x).