Automated learning and decision making systems in public-facing applications are vulnerable to malicious attacks. Examples of such systems include spam detectors, credit card fraud detectors, and network intrusion detection systems. These systems are at further risk of attack when money is directly involved, such as market forecasters or decision systems used in determining insurance or loan rates. In this paper, we consider the setting where a predictor Bob has a fixed model, and an unknown attacker Alice aims to perturb (or poison) future test instances so as to alter Bob's prediction to her benefit. We focus specifically on Bob's optimal defense actions to limit Alice's effectiveness. We define a general framework for determining Bob's optimal defense action against Alice's worst-case attack. We then demonstrate our framework by considering linear predictors, where we provide tractable methods of determining the optimal defense action. Using these methods, we perform an empirical investigation of optimal defense actions for a particular class of linear models -- autoregressive forecasters -- and find that for ten real world futures markets, the optimal defense action reduces the Bob's loss by between 78 and 97%.

We develop a general framework for agent abstraction based on the situation calculus and the ConGolog agent programming language. We assume that we have a high-level specification and a low-level specification of the agent, both represented as basic action theories. A refinement mapping specifies how each high-level action is implemented by a low-level ConGolog program and how each high-level fluent can be translated into a low-level formula. We define a notion of sound abstraction between such action theories in terms of the existence of a suitable bisimulation between their respective models. Sound abstractions have many useful properties that ensure that we can reason about the agent's actions (e.g., executability, projection, and planning) at the abstract level, and refine and concretely execute them at the low level. We also characterize the notion of complete abstraction where all actions (including exogenous ones) that the high level thinks can happen can in fact occur at the low level.

In this paper, we present a Virtual-Suspect system which can be used to train inexperienced law enforcement personnel in interrogation strategies. The system supports different scenario configurations based on historical data. The responses presented by the Virtual-Suspect are selected based on the psychological state of the suspect, which can be configured as well. Furthermore, each interrogator's statement affects the Virtual-Suspect's current psychological state, which may lead the interrogation in different directions. In addition, the model takes into account the context in which the statements are made. Experiments with 24 subjects demonstrate that the Virtual-Suspect's behavior is similar to that of a human who plays the role of the suspect.

This paper studies the benefit for information providers in free public information disclosure in settings where the prospective information buyers are people. The underlying model, which applies to numerous real-life situations, considers a standard decision making setting where the decision maker is uncertain about the outcomes of her decision. The information provider can fully disambiguate this uncertainty and wish to maximize her profit from selling such information. We use a series of AMT-based experiments with people to test the benefit for the information provider from reducing some of the uncertainty associated with the decision maker's problem, for free. Free information disclosure of this kind can be proved to be ineffective when the buyer is a fully-rational agent. Yet, when it comes to people we manage to demonstrate that a substantial improvement in the information provider's profit can be achieved with such an approach. The analysis of the results reveals that the primary reason for this phenomena is people's failure to consider the strategic nature of the interaction with the information provider. Peoples' inability to properly calculate the value of information is found to be secondary in its influence.

The generation of near-optimal plans for multi-agent systems with numerical states and temporal actions is computationally challenging. Current off-the-shelf planners can take a very long time before generating a near-optimal solution. In an effort to reduce plan computation time, increase the quality of the resulting plans, and make them more interpretable by humans, we explore collaborative planning techniques that actively involve human users in plan generation. Specifically, we explore a framework in which users provide high-level strategies encoded as soft preferences to guide the low-level search of the planner. Through human subject experimentation, we empirically demonstrate that this approach results in statistically significant improvements to plan quality, without substantially increasing computation time. We also show that the resulting plans achieve greater similarity to those generated by humans with regard to the produced sequences of actions, as compared to plans that do not incorporate user-provided strategies.

Security agencies have found security games to be useful models to understand how to better protect their assets. The key practical elements in this work are: (i) the attacker can simultaneously attack multiple targets, and (ii) different targets exhibit different types of dependencies based on the assets being protected (e.g., protection of critical infrastructure, network security, etc.). However, little is known about the computational complexity of these problems, especially when there exist dependencies among the targets. Moreover, previous security game models do not in general scale well. In this paper, we investigate a general security game where the utility function is defined on a collection of subsets of all targets, and provide a novel theoretical framework to show how to compactly represent such a game, efficiently compute the optimal (minimax) strategies, and characterize the complexity of this problem. We apply our theoretical framework to the network security game. We characterize settings under which we find a polynomial time algorithm for computing optimal strategies. In other settings we prove the problem is NP-hard and provide an approximation algorithm.

Computational mechanism analysis is a recent approach to economic analysis in which a mechanism design setting is analyzed entirely by a computer. For games with non-trivial numbers of players and actions, the approach is only feasible when these games can be encoded compactly, e.g., as Action-Graph Games. Such encoding is currently a manual process requiring expert knowledge; our aim is to simplify and automate it. Our contribution, the Positronic Economist is a software system having two parts: (1) a Python-based language for succinctly describing mechanisms; and (2) a system that takes such descriptions as input, automatically identifies computationally useful structure, and produces a compact Action-Graph Game.

In this paper, we analyze an emerging economic form, called fans economy, in which a fan donates money to the host and gets allocated proportional to the amount of his donation (normalized by the overall amount of donation). Fans economy is the major way live streaming apps monetize and includes a number of popular economic forms ranging from crowdfunding to mutual fund. We propose an auction game, coined all-pay auctions with proportional allocation (APAPA), to model the fans economy and analyze the auction from the perspective of revenue. Comparing to the standard all-pay auction, which normally has no pure Nash-Equilibrium in the complete information setting, we solve the pure Nash-Equilibrium of the APAPA in closed form and prove its uniqueness. Motivated by practical concerns, we then analyze the case where APAPA is equipped with a reserve and show that there might be multiple equilibria in this case. We give an efficient algorithm to compute all equilibria in this case. For either case, with or without reserve, we show that APAPA always extracts revenue that 2-approximates the second-highest valuation. Furthermore, we conduct experiments to show how revenue changes with respect to different reserves.

We study multi-type housing markets, where there are p ≥ 2 types of items, each agent is initially endowed one item of each type, and the goal is to design mechanisms without monetary transfer to (re)allocate items to the agents based on their preferences over bundles of items, such that each agent gets one item of each type. In sharp contrast to classical housing markets, previous studies in multi-type housing markets have been hindered by the lack of natural solution concepts, because the strict core might be empty. We break the barrier in the literature by leveraging AI techniques and making natural assumptions on agents’ preferences. We show that when agents’ preferences are lexicographic, even with different importance orders, the classical top-trading-cycles mechanism can be extended while preserving most of its nice properties. We also investigate computational complexity of checking whether an allocation is in the strict core and checking whether the strict core is empty. Our results convey an encouragingly positive message: it is possible to design good mechanisms for multi-type housing markets under natural assumptions on preferences.

A repeated game is a formal model for analyzing cooperation in long-term relationships, e.g., in the prisoner's dilemma. Although the case where each player observes her opponent's action with some observation errors (imperfect private monitoring) is difficult to analyze, a special type of an equilibrium called belief-free equilibrium is identified to make the analysis in private monitoring tractable. However, existing works using a belief-free equilibrium show that cooperative relations can be sustainable only in ideal situations. We deal with a generic problem that can model both the prisoner's dilemma and the team production problem. We examine a situation with an additional action that is dominated by another action. To our surprise, by adding this seemingly irrelevant action, players can achieve sustainable cooperative relations far beyond the ideal situations. More specifically, we identify a class of strategies called one-shot punishment strategy that can constitute a belief-free equilibrium in a wide range of parameters. Moreover, for a two-player case, the obtained welfare matches a theoretical upper bound.