**AIJ Prominent and Classic Award Winning Papers**

**2016:**

The 2016 PROMINENT PAPER AWARD was given to:

*Monte-Carlo tree search and rapid action value estimation in computer Go **Sylvain Gelly and David Silver **Artificial Intelligence 175 (11), Pages 1856-1875, 2011*

Go is an ancient Chinese board game that has long been considered one of

the great challenges in AI.

While for a number of years computer programs have managed to beat the world's

leading human players in games like checkers and chess, the high level of intuition

and evaluation required by Go made it tough for AI search methods to crack. This has

changed in recent years, where the last milestone was the recent defeat of the

legendary Go player Lee Se-dol by AlphaGo, a program developed by Google DeepMind.

The previous milestone, however, that enabled this breakthrough, was achieved a decade

ago through the use of Monte-Carlo Tree Search augmented with a number of

enhancements. This led to programs that defeated human professional players and

achieved master (dan) level in 9x9 Go.

That work was reported in 2007 in two papers: one by Rémi Coulom in the

Int. Computer Games Association Journal, the other by Sylvain Gelly and David Silver

at the ICML 2007 conference. This AIJ paper is a follow-up of the latter, covering two of

the enhancements: rapid action value estimation and heuristic initialization. These extensions

led to a program that achieved master level in 19x19 Go for the first time.

The 2016 CLASSIC PAPER AWARD was given to:

*Real-time heuristic search **Richard E. Korf **Artificial intelligence 42 (2-3), pp 189-211, 1990*

This is the seminal paper in real-time heuristic search over the basic state

model considered in AI, where actions have deterministic effects and

information is complete. While standard heuristic search methods are

aimed at solving problems off-line, real-time search methods are used on-line

for selecting the next action to do after some form of lookahead.

Korf's minimin and real-time A* algorithms take inspiration in the ideas underlying

search in 2-player games, while Learning Real-time A* (LRTA*), where the heuristic

values are updated dynamically during the search, was the first to capture two key

properties: avoidance of loops and convergence to the optimal solution. LRTA* remains

a key reference in the field, where a number of variants have been developed that work

under slightly different assumptions, including the Real-time dynamic programming

algorithm (RTDP), that can be regarded as generalization of LRTA* to Markov Decision Processes.

**2015:**

The 2015 PROMINENT PAPER AWARD was given to:

*Label ranking by learning pairwise preferences Eyke Hüllermeier , Johannes Fürnkranz , Weiwei Cheng *

*, Klaus Brinker*

*Artificial Intelligence, Volume 172, issues 16-17, November 2008, pages 1897-1916*

It has been influential in the field of preferences and preference learning.

The 2015 CLASSIC PAPER AWARD was given to:

*Fusion, Propagation, and Structuring in Belief Networks ** Judear Pearl ** Artificial Intelligence 29 (3) (1986) 241-288*

This is the seminal journal paper that introduced Bayesian networks and the

distributed, linear-time, message-passing algorithm for belief propagation in

singly-connected networks (including trees) . This work along with Pearl's 1988 book,

"Probabilistic Reasoning in Intelligent Systems", sparked what some call the "probabilistic

revolution" in AI. The impact of Bayesian networks and Bayesian networks algorithms on

AI, Machine Learning, Information Theory, and Cognitive Science has been huge indeed, providing a

representational and computational framework that relates probabilistic reasoning with graphs, graph topology

with complexity bounds, and causal and evidential inference with directional information flow.

By showing "how to do with probabilities what people say that you can't", the paper introduced

key conceptual notions like the use of graphs for representing independence relations, and

the use of independence relations for making exact probabilistic inference tractable on tree and

tree-like graphs.

**2014:**

The 2014 PROMINENT PAPER AWARD was given to the following two papers:

*Reasoning about preferences in argumentation frameworksSanjay Modgil 173 (9–10), June 2009, Pages 901–934*

Argumentation is concerned with attempting to obtain rationally justifiable positions in the presence of conflicting evidence. Originating from the field of philosophy, argumentation research has now become a major topic for AI researchers. One of the key problems in argumentation is to develop a formal model and associated semantics for argumentation that can express the subtleties and nuances of argument and debate. Sanjay Modgil's paper made a major contribution to this problem.

His paper demonstrates how the canonical graph-based models used in abstract argumentation can be enriched to allow such notions as meta-argument, in which arguments can attack attacks. The paper motivates and presents this new model, and explores the relationship of the model to logic programming. Modgil's work represents a key contribution to the argumentation domain, and an outstanding exemplar of work in this area.

*Practical solution techniques for first-order MDPsScott Sanner and Craig Boutilier,Artificial Intelligence 173 (5–6), April 2009, Pages 748–788*

Decision-theoretic planning problems are naturally represented using probabilistic first-order logic (e.g. PDDL) but are traditionally solved by first 'grounding' the problem. Unfortunately, such a ground representation grows polynomially with the number of domain objects and exponentially in predicate arity. In this seminal paper first-order MDPs are solved without grounding. Although the paper is wide-ranging and could serve as an introduction to this area, it also has the necessary technical depth, providing a clear explanation of solving techniques based on (i) symbolic dynamic programming and (ii) first-order linear programs. Moreover these techniques are implemented and empirically evaluated, showing good results on a range of planning problems. Representing and reasoning with first-order probabilistic theories (often called "lifted inference") is a key research topic in AI; this paper constitutes a major advance to it.

The 2014 CLASSIC PAPER AWARD was given to:

*A logic for default reasoningRay Reiter Artificial Intelligence 13 (1-2), Pages 81-132 (1980) *

This seminal paper introduces and develops a mathematical theory of reasoning about defaults and exceptions that has become to be known as default logic. Reasoning about defaults is about drawing plausible conclusions in the absence of complete knowledge about a world. Default reasoning is a key component of everyday commonsense reasoning, and is essential in many computer systems.

The central element of default logic is the definition of the extensions to a first-order theory induced by a set of defaults. Default logic is nonmonotonic, in the sense that conclusions justified by defaults may need to be retracted when new axioms are added.

Reiter's approach to default reasoning has been immensely influential. Not only it has made a significant impact on the field of knowledge representation, including its application to the frame problem and to other difficult issues in the theory of commonsense reasoning, but it has also made a significant impact on logic programming and underlies much current work on default reasoning including answer set programming.

Overall, this article is one of the cornerstone publications of the knowledge representation research domain, and indeed of AI in general. The award committee is privileged to have the opportunity to recommend unanimously this paper as the recipient of the 2014 AIJ Classic Paper Award.