We investigate two causes for adversarial vulnerability in deep neural networks: bad data and (poorly) trained models. When trained with SGD, deep neural networks essentially achieve zero training error, even in the presence of label noise, while also exhibiting good generalization on natural test data, something referred to as benign overfitting [2, 10]. However, these models are vulnerable to adversarial attacks. We identify label noise as one of the causes for adversarial vulnerability, and provide theoretical and empirical evidence in support of this. Surprisingly, we find several instances of label noise in datasets such as MNIST and CIFAR, and that robustly trained models incur training error on some of these, i.e. they don't fit the noise. However, removing noisy labels alone does not suffice to achieve adversarial robustness. Standard training procedures bias neural networks towards learning "simple" classification boundaries, which may be less robust than more complex ones. We observe that adversarial training does produce more complex decision boundaries. We conjecture that in part the need for complex decision boundaries arises from sub-optimal representation learning. By means of simple toy examples, we show theoretically how the choice of representation can drastically affect adversarial robustness.

Graph neural networks (GNNs) extends the functionality of traditional neural networks to graph-structured data. Similar to CNNs, an optimized design of graph convolution and pooling is key to success. Borrowing ideas from physics, we propose a path integral based graph neural networks (PAN) for classification and regression tasks on graphs. Specifically, we consider a convolution operation that involves every path linking the message sender and receiver with learnable weights depending on the path length, which corresponds to the maximal entropy random walk. It generalizes the graph Laplacian to a new transition matrix we call maximal entropy transition (MET) matrix derived from a path integral formalism. Importantly, the diagonal entries of the MET matrix are directly related to the subgraph centrality, thus providing a natural and adaptive pooling mechanism. PAN provides a versatile framework that can be tailored for different graph data with varying sizes and structures. We can view most existing GNN architectures as special cases of PAN. Experimental results show that PAN achieves state-of-the-art performance on various graph classification/regression tasks, including a new benchmark dataset from statistical mechanics we propose to boost applications of GNN in physical sciences.

In this paper, we focus on the task of multi-view multi-source geo-localization, which serves as an important auxiliary method of GPS positioning by matching drone-view image and satellite-view image with pre-annotated GPS tag. To solve this problem, most existing methods adopt metric loss with an weighted classification block to force the generation of common feature space shared by different view points and view sources. However, these methods fail to pay sufficient attention to spatial information (especially viewpoint variances). To address this drawback, we propose an elegant orientation-based method to align the patterns and introduce a new branch to extract aligned partial feature. Moreover, we provide a style alignment strategy to reduce the variance in image style and enhance the feature unification. To demonstrate the performance of the proposed approach, we conduct extensive experiments on the large-scale benchmark dataset. The experimental results confirm the superiority of the proposed approach compared to state-of-the-art alternatives.

When it comes to AI ethics around the use of facial recognition, China does not have a good record. As India has banned Chinese apps including TikTok, one that went viral in 2019 and 2020 that uses AI to recommend micro videos, Australia and the U.S. are likely to be next. Kevin Mayer left Disney recently to join ByteDance, as CEO of TikTok, but you cannot separate TikTok, from its parent company with an HQ located in Beijing. If this company isn't helping export China's police surveillance capitalism play, I don't know what is. It's the greatest PR stunt by ByteDance I've seen yet.

A new study has explored whether AI can provide more attractive answers to humanity's most profound questions than history's most influential thinkers. Researchers from the University of New South Wales first fed a series of moral questions to Salesforce's CTRL system, a text generator trained on millions of documents and websites, including all of Wikipedia. They added its responses to a collection of reflections from the likes of Plato, Jesus Christ, and, err, Elon Musk. The team then asked more than 1,000 people which musings they liked best -- and whether they could identify the source of the quotes. In worrying results for philosophers, the respondents preferred the AI's answers to almost half the questions. And only a small minority recognized that CTRL's statements were computer-generated.

Arrangements for the OECD's role as host will be finalised in the coming days. The GPAI will bring together experts from industry, government, civil society and academia to conduct research and pilot projects on AI. Its objective, as set out by founding members Australia, Canada, the European Union, France, Germany, India, Italy, Japan, Korea, Mexico, New Zealand, Singapore, Slovenia, the United Kingdom and the United States, is to bridge the gap between theory and practice on AI policy. An example would be looking at how AI could help societies respond to and recover from the Covid-19 crisis. Basing its Secretariat at the OECD will allow the GPAI to create a strong link between international policy development and technical discourse on AI, taking advantage of the OECD's expertise on AI policy and its leadership in setting out the first international standard for trustworthy AI – the OECD Principles on Artificial Intelligence.

Convolutional neural network (CNN) models for computer vision are powerful but lack explainability in their most basic form. This deficiency remains a key challenge when applying CNNs in important domains. Recent work for explanations through feature importance of approximate linear models has moved from input-level features (pixels or segments) to features from mid-layer feature maps in the guise of concept activation vectors (CAVs). CAVs contain concept-level information and could be learnt via Clustering. In this work, we rethink the ACE algorithm of Ghorbani et~al., proposing an alternative concept-based explanation framework. Based on the requirements of fidelity (approximate models) and interpretability (being meaningful to people), we design measurements and evaluate a range of dimensionality reduction methods for alignment with our framework. We find that non-negative concept activation vectors from non-negative matrix factorization provide superior performance in interpretability and fidelity based on computational and human subject experiments. Our framework provides both local and global concept-level explanations for pre-trained CNN models.

In this contribution we study social network modelling by using human interaction as a basis. To do so, we propose a new set of functions, affinities, designed to capture the nature of the local interactions among each pair of actors in a network. By using these functions, we develop a new community detection algorithm, the Borgia Clustering, where communities naturally arise from the multi-agent interaction in the network. We also discuss the effects of size and scale for communities regarding this case, as well as how we cope with the additional complexity present when big communities arise. Finally, we compare our community detection solution with other representative algorithms, finding favourable results.

The ability to conduct logical reasoning is a fundamental aspect of intelligent human behavior, and thus an important problem along the way to human-level artificial intelligence. Traditionally, logic-based symbolic methods from the field of knowledge representation and reasoning have been used to equip agents with capabilities that resemble human logical reasoning qualities. More recently, however, there has been an increasing interest in using machine learning rather than logic-based symbolic formalisms to tackle these tasks. In this paper, we employ state-of-the-art methods for training deep neural networks to devise a novel model that is able to learn how to effectively perform logical reasoning in the form of basic ontology reasoning. This is an important and at the same time very natural logical reasoning task, which is why the presented approach is applicable to a plethora of important real-world problems. We present the outcomes of several experiments, which show that our model is able to learn to perform highly accurate ontology reasoning on very large, diverse, and challenging benchmarks. Furthermore, it turned out that the suggested approach suffers much less from different obstacles that prohibit logic-based symbolic reasoning, and, at the same time, is surprisingly plausible from a biological point of view.

Question answering biases in video QA datasets can mislead multimodal model to overfit to QA artifacts and jeopardize the model's ability to generalize. Understanding how strong these QA biases are and where they come from helps the community measure progress more accurately and provide researchers insights to debug their models. In this paper, we analyze QA biases in popular video question answering datasets and discover pretrained language models can answer 37-48% questions correctly without using any multimodal context information, far exceeding the 20% random guess baseline for 5-choose-1 multiple-choice questions. Our ablation study shows biases can come from annotators and type of questions. Specifically, annotators that have been seen during training are better predicted by the model and reasoning, abstract questions incur more biases than factual, direct questions. We also show empirically that using annotator-non-overlapping train-test splits can reduce QA biases for video QA datasets.