Summary: How to measure the degree and value of AI adoption among companies or even countries is hard. Here's a beginning proposal on how to get started. We talk a great deal about whether there are enough data scientists to go around, whether our advancements in AI techniques are better than others, if there's sufficient access to data, and whether our chips and clouds are up to the task. But to really understand where the rubber meets the road in achieving the promised benefits of AI it's all about who has adopted the most and adopted it to their greatest benefit. This is true whether you're trying to assess one company versus another or even one country versus another.

New artificial intelligence (AI) capabilities may make it possible to improve the effectiveness of brachytherapy for men with prostate cancer (PCa) by almost instantly generating dosage plans, according to investigators. In a typical high-dose rate (HDR) brachytherapy procedure for PCa, needle applicators are first inserted by the physician to the tumor target. A planner then develops a treatment plan manually. During this time the patient carries the needles, waiting for the planning to finish. With the current standard of care, it takes up to an hour or more to generate a high-quality plan.

We study minimax methods for off-policy evaluation (OPE) using value-functions and marginalized importance weights. Despite that they hold promises of overcoming the exponential variance in traditional importance sampling, several key problems remain: (1) They require function approximation and are generally biased. For the sake of trustworthy OPE, is there anyway to quantify the biases? (2) They are split into two styles ("weight-learning" vs "value-learning"). Can we unify them? In this paper we answer both questions positively. By slightly altering the derivation of previous methods (one from each style; Uehara et al., 2019), we unify them into a single confidence interval (CI) that automatically comes with a special type of double robustness: when either the value-function or importance weight class is well-specified, the CI is valid and its length quantifies the misspecification of the other class. We can also tell which class is misspecified, which provides useful diagnostic information for the design of function approximation. Our CI also provides a unified view of and new insights to some recent methods: for example, one side of the CI recovers a version of AlgaeDICE (Nachum et al., 2019b), and we show that the two sides need to be used together and either alone may incur doubled approximation error as a point estimate. We further examine the potential of applying these bounds to two long-standing problems: off-policy policy optimization with poor data coverage (i.e., exploitation), and systematic exploration. With a well-specified value-function class, we show that optimizing the lower and the upper bounds lead to effective exploitation and exploration, respectively. Our results also suggests an interesting assymetry between exploration and exploitation, that the former might require substantially weaker realizability assumptions than the latter.

Bandit learning algorithms typically involve the balance of exploration and exploitation. However, in many practical applications, worst-case scenarios needing systematic exploration are seldom encountered. In this work, we consider a smoothed setting for structured linear contextual bandits where the adversarial contexts are perturbed by Gaussian noise and the unknown parameter $\theta^*$ has structure, e.g., sparsity, group sparsity, low rank, etc. We propose simple greedy algorithms for both the single- and multi-parameter (i.e., different parameter for each context) settings and provide a unified regret analysis for $\theta^*$ with any assumed structure. The regret bounds are expressed in terms of geometric quantities such as Gaussian widths associated with the structure of $\theta^*$. We also obtain sharper regret bounds compared to earlier work for the unstructured $\theta^*$ setting as a consequence of our improved analysis. We show there is implicit exploration in the smoothed setting where a simple greedy algorithm works.

Current trends in machine learning rely on out-of-the-box gradient-based approaches. With the aim of mitigating numerical errors and to improve the convergence of the learning process, a common empirical practice is to standardize or normalize the data. However, there is a lack of theoretical analysis regarding why and when these methods result in an improvement of the learning process. In this work, we first study these methods in the context of black-box variational inference, specifically analyzing the effect that scaling the data has on the smoothness of the optimization landscape. Our analysis shows that no general rule applies in order to decide which of the existing data scaling methods, or even if they, will improve the learning process. Second, we highlight the issues that arise when dealing with multivariate data, due to the discrepancy in smoothness of the likelihood functions for different variables, and the inability to scale discrete data. Finally, we propose a novel Lipschitz standardization, and its extension for discrete data, which overcomes the aforementioned limitations. Specifically, as backed by our experiments, Lipschitz standardization i) favors a fairer learning across different variables in the data; and ii) results in faster and more accurate learning.

Molecular dynamics simulations use statistical mechanics at the atomistic scale to enable both the elucidation of fundamental mechanisms and the engineering of matter for desired tasks. The behavior of molecular systems at the microscale is typically simulated with differential equations parameterized by a Hamiltonian, or energy function. The Hamiltonian describes the state of the system and its interactions with the environment. In order to derive predictive microscopic models, one wishes to infer a molecular Hamiltonian that agrees with observed macroscopic quantities. From the perspective of engineering, one wishes to control the Hamiltonian to achieve desired simulation outcomes and structures, as in self-assembly and optical control, to then realize systems with the desired Hamiltonian in the lab. In both cases, the goal is to modify the Hamiltonian such that emergent properties of the simulated system match a given target. We demonstrate how this can be achieved using differentiable simulations where bulk target observables and simulation outcomes can be analytically differentiated with respect to Hamiltonians, opening up new routes for parameterizing Hamiltonians to infer macroscopic models and develop control protocols.

In this short note we observe that the sample complexity of PAC machine learning of various concepts, including learning the maximum (EMX), can be exactly determined when the support of the probability measures considered as models satisfies an a-priori bound. This result contrasts with the recently discovered undecidability of EMX within ZFC for finitely supported probabilities (with no a priori bound). Unfortunately, the decision procedure is at present, at least doubly exponential in the number of points times the uniform bound on the support size.

Interpreting neural network decisions and the information learned in intermediate layers is still a challenge due to the opaque internal state and shared non-linear interactions. Although (Kim et al, 2017) proposed to interpret intermediate layers by quantifying its ability to distinguish a user-defined concept (from random examples), the questions of robustness (variation against the choice of random examples) and effectiveness (retrieval rate of concept images) remain. We investigate these two properties and propose improvements to make concept activations reliable for practical use. Effectiveness: If the intermediate layer has effectively learned a user-defined concept, it should be able to recall --- at the testing step --- most of the images containing the proposed concept. For instance, we observed that the recall rate of Tiger shark and Great white shark from the ImageNet dataset with "Fins" as a user-defined concept was only 18.35% for VGG16. To increase the effectiveness of concept learning, we propose A-CAV --- the Adversarial Concept Activation Vector --- this results in larger margins between user concepts and (negative) random examples. This approach improves the aforesaid recall to 76.83% for VGG16. For robustness, we define it as the ability of an intermediate layer to be consistent in its recall rate (the effectiveness) for different random seeds. We observed that TCAV has a large variance in recalling a concept across different random seeds. For example, the recall of cat images (from a layer learning the concept of tail) varies from 18% to 86% with 20.85% standard deviation on VGG16. We propose a simple and scalable modification that employs a Gram-Schmidt process to sample random noise from concepts and learn an average "concept classifier". This approach improves the aforesaid standard deviation from 20.85% to 6.4%.

How perception and reasoning arise from neuronal network activity is poorly understood. This is reflected in the fundamental limitations of connectionist artificial intelligence, typified by deep neural networks trained via gradient-based optimization. Despite success on many tasks, such networks remain unexplainable black boxes incapable of symbolic reasoning and concept generalization. Here we show that a simple set of biologically consistent organizing principles confer these capabilities to neuronal networks. To demonstrate, we implement these principles in a novel machine learning algorithm, based on concept construction instead of optimization, to design deep neural networks that reason with explainable neuron activity. On a range of tasks including NP-hard problems, their reasoning capabilities grant additional cognitive functions, like deliberating through self-analysis, tolerating adversarial attacks, and learning transferable rules from simple examples to solve problems of unencountered complexity. The networks also naturally display properties of biological nervous systems inherently absent in current deep neural networks, including sparsity, modularity, and both distributed and localized firing patterns. Because they do not sacrifice performance, compactness, or training time on standard learning tasks, these networks provide a new black-box-free approach to artificial intelligence. They likewise serve as a quantitative framework to understand the emergence of cognition from neuronal networks.

Training deep neural networks is an important problem which is still far from solved. At the core of the problem is our still relatively poor understanding of what happens under the hood of a deep neural network. Practically, this translates to a wide variety of deep network architectures and activation functions used in them. They all, however, suffer from the same problem when it comes to interpretability. It is next to impossible to understand how and why even a single layer network performs a simple classification task, and this probelm only increases with the size and the depth of the network. Activation functions stem from Cybenko's seminal 1989 paper [1], which proved that sigmoidal functions are universal approximators. This gave rise to a number of sigmoidal activation functions, including the sigmoid, tanh, arctan, binary step, Elliott sign [2], SoftSign [3] [4], SQNL [5], soft clipping [6] and many others. Sigmoidal activations were useful in the early days of neural networks, but the most serious problem that they suffered from was vanishing gradients.