We are back with some highlights from the second day of NIPS. A lot of fascinating research was showcased today, and we are excited to share some of our favorites with you. If you missed them, feel free to check our Day 1 and Day 3 Highlights! One of the most memorable sessions of the first two days was today's invited talk by Kate Crawford, about bias in Machine Learning. We recommend taking a look at the feature image of this post, representing modern Machine Learning datasets as an attempt at creating a taxonomy of the world.

Games that model real world interactions are often complex, with huge numbers of possible strategies and information states. We are interested in better understanding the effect of abstraction in game-theoretic analysis. In particular, we focus on the strategy selection problem: how should an agent choose a strategy to play in a game, based on an abstracted game model? This problem has three interacting Figure 1: 2-players asymmetric abstractions components: (1) the method for abstracting the game, (2) the method for selecting a strategy based on the abstraction, and An example of an abstraction meta-game is shown in Figure (3) the method for mapping this strategy back to the original 1. In this example, we have two players who are playing game. This approach has been studied extensively for the one-shot normal form game shown at the top of the poker, which is a 2-player, zero-sum game. However, much figure; this is the base game. They each perform their own less is known about how abstraction interacts with strategy (unspecified) abstraction to reduce the game.

Normal form games are one of the most familiar representations for modeling interactions among multiple agent. However, modeling many realistic interactions between agents results in games that are extremely large. In these cases computing standard solutions like Nash equilibrium may be intractable. To overcome this issue the idea of abstraction has been investigated, most prominently in research on computer Poker. Solving a game using abstraction requires using some method to simplify the game before it is analyzed. We study a new variation for solving normal form games using abstraction that is based on finding and solving suitable sub games. We compare this method with several variations of a common type of abstraction based on clustering similar strategies.

Equilibrium or near-equilibrium solutions to very large extensive form games are often computed by using abstractions to reduce the game size. A common abstraction technique for games with a large number of available actions is to restrict the number of legal actions in every state. This method has been used to discover equilibrium solutions for the game of no-limit heads-up Texas Hold'em. When using a solution to an abstracted game to play one side in the un-abstracted (real) game, the real opponent actions may not correspond to actions in the abstracted game. The most popular method for handling this situation is to translate opponent actions in the real game to the closest legal actions in the abstracted game. We show that this approach can result in a very exploitable player and propose an alternative solution. We use probabilistic mapping to translate a real action into a probability distribution over actions, whose weights are determined by a similarity metric. We show that this approach significantly reduces the exploitability when using an abstract solution in the real game.