Recent progress in Generative Adversarial Networks (GANs) has shown promising signs of improving GAN training via architectural change. Despite some early success, at present the design of GAN architectures requires human expertise, laborious trial-and-error testings, and often draws inspiration from its image classification counterpart. In the current paper, we present the first neural architecture search algorithm, automated neural architecture search for deep generative models, or AGAN for abbreviation, that is specifically suited for GAN training. For unsupervised image generation tasks on CIFAR-10, our algorithm finds architecture that outperforms state-of-the-art models under same regularization techniques. For supervised tasks, the automatically searched architectures also achieve highly competitive performance, outperforming best human-invented architectures at resolution $32\times32$. Moreover, we empirically demonstrate that the modules learned by AGAN are transferable to other image generation tasks such as STL-10.

Neural networks are increasingly employed in safety-critical domains. This has prompted interest in verifying or certifying logically encoded properties of neural networks. Prior work has largely focused on checking existential properties, wherein the goal is to check whether there exists any input that violates a given property of interest. However, neural network training is a stochastic process, and many questions arising in their analysis require probabilistic and quantitative reasoning, i.e., estimating how many inputs satisfy a given property. To this end, our paper proposes a novel and principled framework to quantitative verification of logical properties specified over neural networks. Our framework is the first to provide PAC-style soundness guarantees, in that its quantitative estimates are within a controllable and bounded error from the true count. We instantiate our algorithmic framework by building a prototype tool called NPAQ that enables checking rich properties over binarized neural networks. We show how emerging security analyses can utilize our framework in 3 concrete point applications: quantifying robustness to adversarial inputs, efficacy of trojan attacks, and fairness/bias of given neural networks.

In recent years, reinforcement learning has been successful in solving video games from Atari to Star Craft II. However, the end-to-end model-free reinforcement learning (RL) is not sample efficient and requires a significant amount of computational resources to achieve superhuman level performance. Model-free RL is also unlikely to produce human-like agents for playtesting and gameplaying AI in the development cycle of complex video games. In this paper, we present a hierarchical approach to training agents with the goal of achieving human-like style and high skill level in team sports games. While this is still work in progress, our preliminary results show that the presented approach holds promise for solving the posed multi-agent learning problem.

Designing neural architectures for edge devices is subject to constraints of accuracy, inference latency, and computational cost. Traditionally, researchers manually craft deep neural networks to meet the needs of mobile devices. Neural Architecture Search (NAS) was proposed to automate the neural architecture design without requiring extensive domain expertise and significant manual efforts. Recent works utilized NAS to design mobile models by taking into account hardware constraints and achieved state-of-the-art accuracy with fewer parameters and less computational cost measured in Multiply-accumulates (MACs). To find highly compact neural architectures, existing works relies on predefined cells and directly applying width multiplier, which may potentially limit the model flexibility, reduce the useful feature map information, and cause accuracy drop. To conquer this issue, we propose GRAM(GRAph propagation as Meta-knowledge) that adopts fine-grained (node-wise) search method and accumulates the knowledge learned in updates into a meta-graph. As a result, GRAM can enable more flexible search space and achieve higher search efficiency. Without the constraints of predefined cell or blocks, we propose a new structure-level pruning method to remove redundant operations in neural architectures. SwiftNet, which is a set of models discovered by GRAM, outperforms MobileNet-V2 by 2.15x higher accuracy density and 2.42x faster with similar accuracy. Compared with FBNet, SwiftNet reduces the search cost by 26x and achieves 2.35x higher accuracy density and 1.47x speedup while preserving similar accuracy. SwiftNetcan obtain 63.28% top-1 accuracy on ImageNet-1K with only 53M MACs and 2.07M parameters. The corresponding inference latency is only 19.09 ms on Google Pixel 1.

A remarkable feature of human beings is their capacity for creative behaviour, referring to their ability to react to problems in ways that are novel, surprising, and useful. Transformational creativity is a form of creativity where the creative behaviour is induced by a transformation of the actor's conceptual space, that is, the representational system with which the actor interprets its environment. In this report, we focus on ways of adapting systems of learned representations as they switch from performing one task to performing another. We describe an experimental comparison of multiple strategies for adaptation of learned features, and evaluate how effectively each of these strategies realizes the adaptation, in terms of the amount of training, and in terms of their ability to cope with restricted availability of training data. We show, among other things, that across handwritten digits, natural images, and classical music, adaptive strategies are systematically more effective than a baseline method that starts learning from scratch.

Intratumor heterogeneity in lung cancer may influence outcomes. CT radiomics seeks to assess tumor features to provide detailed imaging features. However, CT radiomic features vary according to the reconstruction kernel used for image generation. To investigate the effect of different reconstruction kernels on radiomic features and assess whether image conversion using a convolutional neural network (CNN) could improve reproducibility of radiomic features between different kernels. In this retrospective analysis, patients underwent non–contrast material–enhanced and contrast material–enhanced axial chest CT with soft kernel (B30f) and sharp kernel (B50f) reconstruction using a single CT scanner from April to June 2017.

Plate from Muybridge's Animal Locomotion series published in 1887. Deep learning has become the dominate lens through which machines understand video. Yet video files consume huge amounts of storage space and are extremely computationally demanding to analyze using deep learning. Certain use cases can benefit from converting videos to sequences of still images for analysis, enabling full data parallelism and vast reductions in data storage and computation. Representing video as still imagery also presents unique opportunities for non-consumptive analysis similar to the use of ngrams for text.

As deep learning has become ubiquitous, evaluations of its accuracy typically compare its performance against an idealized baseline of flawless human results that bear no resemblance to the actual human workflow those algorithms are being designed to replace. For example, the accuracy of real-time algorithmic speech recognition is frequently compared against human captioning produced in offline multi-coder reconciled environments and subjected to multiple reviews to generate flawless content that looks absolutely nothing like actual real-time human transcription. If we really wish to understand the usability of AI today we should be comparing it against the human workflows it is designed to replace, not an impossible vision of nonexistent human perfection. While the press is filled with the latest superhuman exploits of bleeding-edge research AI systems besting humans at yet another task, the reality of production AI systems is far more mundane. Most commercial applications of deep learning can achieve higher accuracy than their human counterparts at some tasks and worse performance on others. Instead of comparing deep learning algorithms against trained humans placed in the same situations, we as a society have developed the habit of comparing them against idealized but entirely non-existent utopian baselines that look nothing like actual human performance on those tasks.

When asked why he robbed banks, Willie Sutton famously replied, "Because that's where the money is". And so much of artificial antelligence evolved in the United States – because that's where the computers were. However with Europe's strong educational institutions, the path to advanced AI technologies has been cleared by European computer scientists, neuroscientists, and engineers – many of whom were later poached by US universities and companies. From backpropagation to Google Translate, deep learning, and the development of more advanced GPUs permitting faster processing and rapid developments in AI over the past decade, some of the greatest contributions to AI have come from European minds. Modern AI can be traced back to the work of the English mathematician Alan Turing, who in early 1940 designed the bombe – an electromechanical precursor to the modern computer (itself based on previous work by Polish scientists) that broke the German military codes in World War II.

It seems like using these pre-trained models have become a new standard for industry best practices. After all, why wouldn't you take advantage of a model that's been trained on more data and compute than you could ever muster by yourself? Advances within the NLP space have also encouraged the use of pre-trained language models like GPT and GPT-2, AllenNLP's ELMo, Google's BERT, and Sebastian Ruder and Jeremy Howard's ULMFiT (for an excellent over of these models, see this TOPBOTs post). One common technique for leveraging pretrained models is feature extraction, where you're retrieving intermediate representations produced by the pretrained model and using those representations as inputs for a new model. These final fully-connected layers are generally assumed to capture information that is relevant for solving a new task.