With increase in capabilities of artificial intelligence, over the last decade, a significant number of researchers have realized importance in creating not only capable intelligent systems, but also making them safe and secure [1-6]. Unfortunately, the field of AI Safety is very young, and researchers are still working to identify its main challenges and limitations. Impossibility results are well known in many fields of inquiry [7-13], and some have now been identified in AI Safety [14-16]. In this paper, we concentrate on a poorly understood concept of unpredictability of intelligent systems [17], which limits our ability to understand impact of intelligent systems we are developing and is a challenge for software verification and intelligent system control, as well as AI Safety in general. In theoretical computer science and in software development in general, many well-known impossibility results are well established, some of them are strongly related to the subject of this paper, for example: Rice's Theorem states that no computationally effective method can decide if a program will exhibit a particular nontrivial behavior, such as producing a specific output [18].

Undirected neural sequence models such as BERT [Devlin et al., 2019] have received renewed interest due to their success on discriminative natural language understanding tasks such as question-answering and natural language inference. The problem of generating sequences directly from these models has received relatively little attention, in part because generating from such models departs significantly from the conventional approach of monotonic generation in directed sequence models. We investigate this problem by first proposing a generalized model of sequence generation that unifies decoding in directed and undirected models. The proposed framework models the process of generation rather than a resulting sequence, and under this framework, we derive various neural sequence models as special cases, such as autoregressive, semi-autoregressive, and refinement-based non-autoregressive models. This unification enables us to adapt decoding algorithms originally developed for directed sequence models to undirected models. We demonstrate this by evaluating various decoding strategies for the recently proposed cross-lingual masked translation model [Lample and Conneau, 2019]. Our experiments reveal that generation from undirected sequence models, under our framework, is competitive with the state of the art on WMT'14 English-German translation. We furthermore observe that the proposed approach enables constant-time translation while remaining within 1 BLEU score compared to linear-time translation from the same undirected neural sequence model.

We consider the combinatorial multi-armed bandit (CMAB) problem, where the reward function is nonlinear. In this setting, the agent chooses a batch of arms on each round and receives feedback from each arm of the batch. The reward that the agent aims to maximize is a function of the selected arms and their expectations. In many applications, the reward function is highly nonlinear, and the performance of existing algorithms relies on a global Lipschitz constant to encapsulate the function's nonlinearity. This may lead to loose regret bounds, since by itself, a large gradient does not necessarily cause a large regret, but only in regions where the uncertainty in the reward's parameters is high. To overcome this problem, we introduce a new smoothness criterion, which we term \emph{Gini-weighted smoothness}, that takes into account both the nonlinearity of the reward and concentration properties of the arms. We show that a linear dependence of the regret in the batch size in existing algorithms can be replaced by this smoothness parameter. This, in turn, leads to much tighter regret bounds when the smoothness parameter is batch-size independent. For example, in the probabilistic maximum coverage (PMC) problem, that has many applications, including influence maximization, diverse recommendations and more, we achieve dramatic improvements in the upper bounds. We also prove matching lower bounds for the PMC problem and show that our algorithm is tight, up to a logarithmic factor in the problem's parameters.

To combine explicit and implicit generative models, we introduce semi-implicit generator (SIG) as a flexible hierarchical model that can be trained in the maximum likelihood framework. Both theoretically and experimentally, we demonstrate that SIG can generate high quality samples especially when dealing with multi-modality. By introducing SIG as an unbiased regularizer to the generative adversarial network (GAN), we show the interplay between maximum likelihood and adversarial learning can stabilize the adversarial training, resist the notorious mode collapsing problem of GANs, and improve the diversity of generated random samples.

We propose a set of compositional design patterns to describe a large variety of systems that combine statistical techniques from machine learning with symbolic techniques from knowledge representation. As in other areas of computer science (knowledge engineering, software engineering, ontology engineering, process mining and others), such design patterns help to systematize the literature, clarify which combinations of techniques serve which purposes, and encourage re-use of software components. We have validated our set of compositional design patterns against a large body of recent literature.

In importance sampling (IS)-based reinforcement learning algorithms such as Proximal Policy Optimization (PPO), IS weights are typically clipped to avoid large variance in learning. However, policy update from clipped statistics induces large bias in tasks with high action dimensions, and bias from clipping makes it difficult to reuse old samples with large IS weights. In this paper, we consider PPO, a representative on-policy algorithm, and propose its improvement by dimension-wise IS weight clipping which separately clips the IS weight of each action dimension to avoid large bias and adaptively controls the IS weight to bound policy update from the current policy. This new technique enables efficient learning for high action-dimensional tasks and reusing of old samples like in off-policy learning to increase the sample efficiency. Numerical results show that the proposed new algorithm outperforms PPO and other RL algorithms in various Open AI Gym tasks.

A fundamental result in differential privacy states that the privacy guarantees of a mechanism are preserved by any post-processing of its output. In this paper we investigate under what conditions stochastic post-processing can amplify the privacy of a mechanism. By interpreting post-processing as the application of a Markov operator, we first give a series of amplification results in terms of uniform mixing properties of the Markov process defined by said operator. Next we provide amplification bounds in terms of coupling arguments which can be applied in cases where uniform mixing is not available. Finally, we introduce a new family of mechanisms based on diffusion processes which are closed under post-processing, and analyze their privacy via a novel heat flow argument. As applications, we show that the rate of "privacy amplification by iteration" in Noisy SGD introduced by Feldman et al. [FOCS'18] admits an exponential improvement in the strongly convex case, and propose a simple mechanism based on the Ornstein-Uhlenbeck process which has better mean squared error than the Gaussian mechanism when releasing a bounded function of the data.

Major internet companies routinely perform tens of thousands of A/B tests each year. Such large-scale sequential experimentation has resulted in a recent spurt of new algorithms that can provably control the false discovery rate (FDR) in a fully online fashion. However, current state-of-the-art adaptive algorithms can suffer from a significant loss in power if null p-values are conservative (stochastically larger than the uniform distribution), a situation that occurs frequently in practice. In this work, we introduce a new adaptive discarding method called ADDIS that provably controls the FDR and achieves the best of both worlds: it enjoys appreciable power increase over all existing methods if nulls are conservative (the practical case), and rarely loses power if nulls are exactly uniformly distributed (the ideal case). We provide several practical insights on robust choices of tuning parameters, and extend the idea to asynchronous and offline settings as well.

We introduce a variant of the $k$-nearest neighbor classifier in which $k$ is chosen adaptively for each query, rather than supplied as a parameter. The choice of $k$ depends on properties of each neighborhood, and therefore may significantly vary between different points. (For example, the algorithm will use larger $k$ for predicting the labels of points in noisy regions.) We provide theory and experiments that demonstrate that the algorithm performs comparably to, and sometimes better than, $k$-NN with an optimal choice of $k$. In particular, we derive bounds on the convergence rates of our classifier that depend on a local quantity we call the `advantage' which is significantly weaker than the Lipschitz conditions used in previous convergence rate proofs. These generalization bounds hinge on a variant of the seminal Uniform Convergence Theorem due to Vapnik and Chervonenkis; this variant concerns conditional probabilities and may be of independent interest.

Grogan et al [11,12] have recently proposed a solution to colour transfer by minimising the Euclidean distance L2 between two probability density functions capturing the colour distributions of two images (palette and target). It was shown to be very competitive to alternative solutions based on Optimal Transport for colour transfer. We show that in fact Grogan et al's formulation can also be understood as a new robust Optimal Transport based framework with entropy regularisation over marginals.