Recent works have developed several methods of defending neural networks against adversarial attacks with certified guarantees. However, these techniques can be computationally costly due to the use of certification during training. We develop a new regularizer that is both more efficient than existing certified defenses, requiring only one additional forward propagation through a network, and can be used to train networks with similar certified accuracy. Through experiments on MNIST and CIFAR-10 we demonstrate improvements in training speed and comparable certified accuracy compared to state-of-the-art certified defenses.

In recent years, extending variational autoencoder's framework to learn disentangled representations has received much attention. We address this problem by proposing a framework capable of disentangling class-related and class-independent factors of variation in data. Our framework employs an attention mechanism in its latent space in order to improve the process of extracting class-related factors from data. We also deal with the multimodality of data distribution by utilizing mixture models as learnable prior distributions, as well as incorporating the Bhattacharyya coefficient in the objective function to prevent highly overlapping mixtures. Our model's encoder is further trained in a semi-supervised manner, with a small fraction of labeled data, to improve representations' interpretability. Experiments show that our framework disentangles class-related and class-independent factors of variation and learns interpretable features. Moreover, we demonstrate our model's performance with quantitative and qualitative results on various datasets.

In recent years, deep neural networks have emerged as the state-of-the-art method for classification in numerous domains, in particular computer vision (Krizhevsky et al., 2012; Zeiler & Fergus, 2014; Simonyan & Zisserman, 2015; Szegedy et al., 2015; He et al., 2016). However, a large-scale dataset with clean annotations, such as ImageNet (Deng et al., 2009), is usually required for the networks to show superior performance. In practice, obtaining such a large amount of data with manually verified labels is prohibitively laborious or expensive. On the other hand, various crowdsourcing platforms make it possible (and cheaper) to collect large amounts of annotated data, even though the labels may be incorrect. This calls for robust methods that can still effectively learn from the copious data in the presence of noisy labels. When the training set contains noisy labels, it has been observed in Zhang et al. (2017); Arpit et al. (2017) that deep neural networks exhibit the early learning phenomenon (Liu et al., 2020), in which the clean instances are fit to the network before the noisy ones are (over-)fit to the network. On the theoretical front, it is proved in Liu et al. (2020) that even simple linear models in high dimensions exhibit such behavior as well. This intriguing phenomenon, as a double-edged sword, begs the question: Is there a way to make use of the discriminating power the network learns during the early stage of the training, in order to prevent itself from memorizing the noisy labels at the later stage of the training? In this paper, we propose MARVEL (MARgins Via Early Learning), a method that attempts to exclude noisy instances from future training by leveraging this early learning phenomenon.

To cope with high annotation costs, training a classifier only from weakly supervised data has attracted a great deal of attention these days. Among various approaches, strengthening supervision from completely unsupervised classification is a promising direction, which typically employs class priors as the only supervision and trains a binary classifier from unlabeled (U) datasets. While existing risk-consistent methods are theoretically grounded with high flexibility, they can learn only from two U sets. In this paper, we propose a new approach for binary classification from m U-sets for $m\ge2$. Our key idea is to consider an auxiliary classification task called surrogate set classification (SSC), which is aimed at predicting from which U set each observed data is drawn. SSC can be solved by a standard (multi-class) classification method, and we use the SSC solution to obtain the final binary classifier through a certain linear-fractional transformation. We built our method in a flexible and efficient end-to-end deep learning framework and prove it to be classifier-consistent. Through experiments, we demonstrate the superiority of our proposed method over state-of-the-art methods.

A high-precision manipulation task, such as needle threading, is challenging. Physiological studies have proposed connecting low-resolution peripheral vision and fast movement to transport the hand into the vicinity of an object, and using high-resolution foveated vision to achieve the accurate homing of the hand to the object. The results of this study demonstrate that a deep imitation learning based method, inspired by the gaze-based dual resolution visuomotor control system in humans, can solve the needle threading task. First, we recorded the gaze movements of a human operator who was teleoperating a robot. Then, we used only a high-resolution image around the gaze to precisely control the thread position when it was close to the target. We used a low-resolution peripheral image to reach the vicinity of the target. The experimental results obtained in this study demonstrate that the proposed method enables precise manipulation tasks using a general-purpose robot manipulator and improves computational efficiency.

With strong marketing advocacy of the benefits of cannabis use for improved mental health, cannabis legalization is a priority among legislators. However, preliminary scientific research does not conclusively associate cannabis with improved mental health. In this study, we explore the relationship between depression and consumption of cannabis in a targeted social media corpus involving personal use of cannabis with the intent to derive its potential mental health benefit. We use tweets that contain an association among three categories annotated by domain experts - Reason, Effect, and Addiction. The state-of-the-art Natural Langauge Processing techniques fall short in extracting these relationships between cannabis phrases and the depression indicators. We seek to address the limitation by using domain knowledge; specifically, the Drug Abuse Ontology for addiction augmented with Diagnostic and Statistical Manual of Mental Disorders lexicons for mental health. Because of the lack of annotations due to the limited availability of the domain experts' time, we use supervised contrastive learning in conjunction with GPT-3 trained on a vast corpus to achieve improved performance even with limited supervision. Experimental results show that our method can significantly extract cannabis-depression relationships better than the state-of-the-art relation extractor. High-quality annotations can be provided using a nearest neighbor approach using the learned representations that can be used by the scientific community to understand the association between cannabis and depression better.

We consider the problem of molecular graph generation using deep models. While graphs are discrete, most existing methods use continuous latent variables, resulting in inaccurate modeling of discrete graph structures. In this work, we propose GraphDF, a novel discrete latent variable model for molecular graph generation based on normalizing flow methods. GraphDF uses invertible modulo shift transforms to map discrete latent variables to graph nodes and edges. We show that the use of discrete latent variables reduces computational costs and eliminates the negative effect of dequantization. Comprehensive experimental results show that GraphDF outperforms prior methods on random generation, property optimization, and constrained optimization tasks.

Characterizing and removing motion blur caused by camera shake or object motion remains an important task for image restoration. In recent years, removal of motion blur in photographs has seen impressive progress in the hands of deep learning-based methods, trained to map directly from blurry to sharp images. Characterization of motion blur, on the other hand, has received less attention and progress in model-based methods for restoration lags behind that of data-driven end-to-end approaches. In this paper, we propose a general, non-parametric model for dense non-uniform motion blur estimation. Given a blurry image, we estimate a set of adaptive basis kernels as well as the mixing coefficients at pixel level, producing a per-pixel map of motion blur. This rich but efficient forward model of the degradation process allows the utilization of existing tools for solving inverse problems. We show that our method overcomes the limitations of existing non-uniform motion blur estimation and that it contributes to bridging the gap between model-based and data-driven approaches for deblurring real photographs.

In tumor segmentation, inter-observer variation is acknowledged to be a significant problem. This is even more significant in clinical target volume (CTV) segmentation, specifically, in post-operative settings, where a gross tumor does not exist. In this scenario, CTV is not an anatomically established structure but rather one determined by the physician based on the clinical guideline used, the preferred trade off between tumor control and toxicity, their experience, training background etc... This results in high inter-observer variability between physicians. Inter-observer variability has been considered an issue, however its dosimetric consequence is still unclear, due to the absence of multiple physician CTV contours for each patient and the significant amount of time required for dose planning. In this study, we analyze the impact that these physician stylistic variations have on organs-at-risk (OAR) dose by simulating the clinical workflow using deep learning. For a given patient previously treated by one physician, we use DL-based tools to simulate how other physicians would contour the CTV and how the corresponding dose distributions should look like for this patient. To simulate multiple physician styles, we use a previously developed in-house CTV segmentation model that can produce physician style-aware segmentations. The corresponding dose distribution is predicted using another in-house deep learning tool, which, averaging across all structures, is capable of predicting dose within 3% of the prescription dose on the test data. For every test patient, four different physician-style CTVs are considered and four different dose distributions are analyzed. OAR dose metrics are compared, showing that even though physician style variations results in organs getting different doses, all the important dose metrics except Maximum Dose point are within the clinically acceptable limit.

Generating an image from its textual description requires both a certain level of language understanding and common sense knowledge about the spatial relations of the physical entities being described. In this work, we focus on inferring the spatial relation between entities, a key step in the process of composing scenes based on text. More specifically, given a caption containing a mention to a subject and the location and size of the bounding box of that subject, our goal is to predict the location and size of an object mentioned in the caption. Previous work did not use the caption text information, but a manually provided relation holding between the subject and the object. In fact, the used evaluation datasets contain manually annotated ontological triplets but no captions, making the exercise unrealistic: a manual step was required; and systems did not leverage the richer information in captions. Here we present a system that uses the full caption, and Relations in Captions (REC-COCO), a dataset derived from MS-COCO which allows to evaluate spatial relation inference from captions directly. Our experiments show that: (1) it is possible to infer the size and location of an object with respect to a given subject directly from the caption; (2) the use of full text allows to place the object better than using a manually annotated relation. Our work paves the way for systems that, given a caption, decide which entities need to be depicted and their respective location and sizes, in order to then generate the final image.