This work studies the (non)robustness of two-layer neural networks in various high-dimensional linearized regimes. We establish fundamental trade-offs between memorization and robustness, as measured by the Sobolev-seminorm of the model w.r.t the data distribution, i.e the square root of the average squared $L_2$-norm of the gradients of the model w.r.t the its input. More precisely, if $n$ is the number of training examples, $d$ is the input dimension, and $k$ is the number of hidden neurons in a two-layer neural network, we prove for a large class of activation functions that, if the model memorizes even a fraction of the training, then its Sobolev-seminorm is lower-bounded by (i) $\sqrt{n}$ in case of infinite-width random features (RF) or neural tangent kernel (NTK) with $d \gtrsim n$; (ii) $\sqrt{n}$ in case of finite-width RF with proportionate scaling of $d$ and $k$; and (iii) $\sqrt{n/k}$ in case of finite-width NTK with proportionate scaling of $d$ and $k$. Moreover, all of these lower-bounds are tight: they are attained by the min-norm / least-squares interpolator (when $n$, $d$, and $k$ are in the appropriate interpolating regime). All our results hold as soon as data is log-concave isotropic, and there is label-noise, i.e the target variable is not a deterministic function of the data / features. We empirically validate our theoretical results with experiments. Accidentally, these experiments also reveal for the first time, (iv) a multiple-descent phenomenon in the robustness of the min-norm interpolator.

It is well known that deep learning models have a propensity for fitting the entire training set even with random labels, which requires memorization of every training sample. In this paper, we investigate the memorization effect in adversarial training (AT) for promoting a deeper understanding of capacity, convergence, generalization, and especially robust overfitting of adversarially trained classifiers. We first demonstrate that deep networks have sufficient capacity to memorize adversarial examples of training data with completely random labels, but not all AT algorithms can converge under the extreme circumstance. Our study of AT with random labels motivates further analyses on the convergence and generalization of AT. We find that some AT methods suffer from a gradient instability issue, and the recently suggested complexity measures cannot explain robust generalization by considering models trained on random labels. Furthermore, we identify a significant drawback of memorization in AT that it could result in robust overfitting. We then propose a new mitigation algorithm motivated by detailed memorization analyses. Extensive experiments on various datasets validate the effectiveness of the proposed method.

This part of the gradient behaves similarly for permuted and unpermuted examples. In Eq. 25 we see that the contribution to the label dependent part of the gradient from permuted examples vanishes for large datasets, while the contribution from unpermuted examples does not provided the cross correlation between input features and labels is nonzero. This suggests that with small weight initialization, the gradient descent dynamics initially ignores the labels of permuted examples. Figure A.1 shows a breakdown of how the two components of the gradient computed on both unpermuted and permuted examples evolve over the course of training for the different layers of the VGG16 model trained on CIFAR-100. We see that the label dependent part behaves qualitatively differently for the unpermuted examples than for the permuted examples, as the permuted examples give close to zero contribution early in training in agreement with Eq. 25. The label independent part of the gradient shows similar trends between unpermuted and permuted examples, though in the final epochs, the unpermuted examples have a slightly larger label independent gradient indicating slightly greater model confidence on these examples. As the label dependent and label independent parts of the gradient have differing signs, they compete with each other and cancel when the loss is minimized, but are not independently zero and in fact grow during training. The slightly larger label independent gradient for unpermuted examples is balanced by a corresponding slightly larger label dependent gradient at the end of training.

The use case for Artificial Intelligence (AI) in the workplace is there. Deloitte's Tech Trends 2021 found AI and machine learning technologies are helping financial services firm Morgan Stanley use decades of data to supplement human insight with accurate models for fraud detection and prevention, sales and marketing automation, and personalized wealth management, among others. For marketing and customer experience, in particular, organizations are using AI and machine learning to improve internal business processes and workflow, automating repetitive tasks and to improve customer journeys and touchpoints, among other use cases. The CMO Survey by Duke University reports a steady increase as far as the extent to which companies are reporting implementing AI or ML into their marketing toolkits. However, the majority of marketers know AI is very important or critical to their success this year, according to Paul Roetzer, founder and CEO of the Marketing AI Institute and PR 20/20.

During the British summer, conversations about sport become almost ubiquitous. This year, however, one participant in those conversations was very different: IBM Watson, IBM's cognitive intelligence. The All England Lawn Tennis Club knew that 2016 would feature unusually fierce competition for attention, with the Tour de France and Euro 2016 taking place alongside Wimbledon. More than ever before, social media was going to be a vital tool in directing that conversation, and directing attention to SW19. Wimbledon's "Cognitive Command Centre" – powered by Watson's intelligence running on a hybrid, IBM-managed cloud - scanned social media for emerging news and trends.

Gartner predicts that "by 2022, 70 percent of white-collar workers will interact with conversational platforms on a daily basis." As a result, the research group found that more organizations are investing in chatbot development and deployment. IBM Business Partners like Sopra Steria are making chatbot and virtual assistant technology available to businesses. Sopra Steria, a European leader in digital transformation, has developed an intelligent virtual assistant for organizations across several industries who want to use an AI conversational interface to answer recurrent customer service questions. In developing our solution, we at Sopra Steria were looking for AI technology that was easy to configure and could support multiple languages and complex dialogs.

During the early stages of developing Case-Based Reasoning (CBR) systems the definition of similarity measures is challenging since this task requires transferring implicit knowledge of domain experts into knowledge representations. While an entire CBR system is very explanatory, the similarity measure determines the ranking but do not necessarily show which features contribute to high (or low) rankings. In this paper we present our work on opening the knowledge engineering process for similarity modelling. This work present is a result of an interdisciplinary research collaboration between AI and public health researchers developing e-Health applications. During this work explainability and transparency of the development process is crucial to allow in-depth quality assurance of the by the domain experts.

Recently, it has been proposed that fruitful synergies may exist between Deep Learning (DL) and Case Based Reasoning (CBR); that there are insights to be gained by applying CBR ideas to problems in DL (what could be called DeepCBR). In this paper, we report on a program of research that applies CBR solutions to the problem of Explainable AI (XAI) in the DL. We describe a series of twin-systems pairings of opaque DL models with transparent CBR models that allow the latter to explain the former using factual, counterfactual and semi-factual explanation strategies. This twinning shows that functional abstractions of DL (e.g., feature weights, feature importance and decision boundaries) can be used to drive these explanatory solutions. We also raise the prospect that this research also applies to the problem of Data Augmentation in DL, underscoring the fecundity of these DeepCBR ideas.

The world is going through a cybersecurity pandemic. No day passes without a hack or data theft being carried out, discovered, or begrudgingly announced. High-profile victims abound – from the PlayStation Network, hacked in 2011, to Dropbox's 2012 breach, to the 500-million-user data theft Yahoo! suffered in 2014, two years before going public about the hack. Those carrying out the attacks have honed their craft to create ever more sophisticated hacking tools. According to a recent study by security consultancy Juniper Research, cybercrime is expected to balloon into a $2.1 trillion (£1.7 trillion) industry by 2019.

It is often challenging for a system to solve a new complex problem from scratch, but much easier if the system can access other similar problems and description of their solutions -- a paradigm known as case-based reasoning (CBR). We propose a neuro-symbolic CBR approach for question answering over large knowledge bases (CBR-KBQA). While the idea of CBR is tempting, composing a solution from cases is nontrivial, when individual cases only contain partial logic to the full solution. To resolve this, CBR-KBQA consists of two modules: a non-parametric memory that stores cases (question and logical forms) and a parametric model which can generate logical forms by retrieving relevant cases from memory. Through experiments, we show that CBR-KBQA can effectively derive novel combination of relations not presented in case memory that is required to answer compositional questions. On several KBQA datasets that test compositional generalization, CBR-KBQA achieves competitive performance. For example, on the challenging ComplexWebQuestions dataset, CBR-KBQA outperforms the current state of the art by 11% accuracy. Furthermore, we show that CBR-KBQA is capable of using new cases \emph{without} any further training. Just by incorporating few human-labeled examples in the non-parametric case memory, CBR-KBQA is able to successfully generate queries containing unseen KB relations.