We present a supervised sequence to sequence transduction model with a hard attention mechanism which combines the more traditional statistical alignment methods with the power of recurrent neural networks. We evaluate the model on the task of morphological inflection generation and show that it provides state of the art results in various setups compared to the previous neural and non-neural a...

The main aim of this paper is to provide a new analysis of licensers of negative polarity items (NPIs). The problems with Fauconnier-Ladusaw's downward entailment analysis have been argued since Linebarger (1980). I will show that there exists a class of weak NPI licensers characterized by non-monotonicity and exclusivity. Weak negation, which is monotone decreasing, has been known to license w...

We study two important properties for the scalability of a replicated system: genuine partial replication (GPR) and snapshot isolation (SI). We prove that these properties are incompatible. To side step this impossibility result, we propose a novel consistency criterion called Non-Monotonic Snapshot Isolation (NMSI). NMSI retains the most important properties of SI: read-only transactions alway...

Today’s models for propagation-based constraint solvers require propagators as implementations of constraints to be at least contracting and monotonic. These models do not comply with reality: today’s constraint programming systems actually use non-monotonic propagators. This paper introduces the first realistic model of constraint propagation by assuming a propagator to be weakly monotonic (co...

Non-monotonic inference is inference that is defeasible: in contrast with deductive inference, the conclusions drawn may be withdrawn in the light of further information, even though all the original premises are retained. Much of our everyday reasoning is like this, and a non-monotonic approach has applications to a number of technical problems in artificial intelligence. Work on formalizing n...

A key problem in reinforcement learning for control with general function approximators (such as deep neural networks and other nonlinear functions) is that, for many algorithms employed in practice, updates to the policy or Q-function may fail to improve performance—or worse, actually cause the policy performance to degrade. Prior work has addressed this for policy iteration by deriving tight ...

Statistical Rate Monotonic Scheduling (SRMS) is a generalization of the classical RMS results of Liu and Layland [10] for periodic tasks with highly variable execution times and statistical QoS requirements. The main tenet of SRMS is that the variability in task resource requirements could be smoothed through aggregation to yield guaranteed QoS. This aggregation is done over time for a given ta...

Anisotropic diffusion affords an efficient, adaptive signal smoothing technique that can be used for signal enhancement, signal segmentation, and signal scale-space creation. This paper introduces a novel partial differential equation (PDE)-based diffusion method for generating locally monotonic signals. Unlike previous diffusion techniques that diverge or converge to trivial signals, locally m...

In a monotonic sequence game, two players alternately choose elements of a sequence from some fixed ordered set. The game ends when the resulting sequence contains either an ascending subsequence of length a or a descending one of length d. We investigate the behaviour of this game when played on finite linear orders or Q and provide some general observations for play on arbitrary ordered sets.

We characterize which scoring rules are Maskin-monotonic for each social choice problem as a function of the number of agents and the number of alternatives. We show that a scoring rule is Maskin-monotonic if and only if it satisfies a certain unanimity condition. Since scoring rules are neutral, Maskin-monotonicity turns out to be equivalent to Nash-implementability within the class of scoring...