AAAI-18 Whats Hot Talks
Sunday, February 4
Camp, Third floor, Hilton
The AAAI-18 “What’s Hot” track aims to present exciting recent advances and current challenges in subareas of Artificial Intelligence with major conferences or competitions. Five “What’s Hot” presentations will be presented, including:
Hot Trends in Autonomous Agents and Multiagent Systems (AAMAS)
Sanmay Das, Edmund H. Durfee
The International Conference on Autonomous Agents and Multiagent Systems (AAMAS) brings together researchers in all areas of agent technology, and provides a single high-profile forum for research in the theory and practice of autonomous agents and multiagent systems. In this brief note, we use papers presented at AAMAS 2017 to identify “hot” topics in the field, including the integration of concepts from multiple AI subfields into intelligent agents, the use of concepts from social science for improving multiagent interactions, and the application of past and emerging AAMAS advances to solve problems of societal importance.
What’s Hot at CP2017
J. Christopher Beck
The 23rd International Conference on Principles and Practice of Constraint Programming was held in Melbourne Australia from August 28 to September 1, 2017, co-located with SAT2017 and ICLP2017. The conferences shared five invited talks covering the intersection of SAT, CP and logic programming; parallelization; machine learning for automated solver design; advances in MaxSAT, and the commercialization of solver technology. The passing of Alain Colmerauer was marked with a technical session and a reception in his honor. In addition to the main technical program and the application track, the conference featured thematic tracks on Machine Learning, Operations Research, Satisfiability, and Test and Verification. Highlights included continued strong research on the integration of CP with SAT and mathematical programming, the traditional strength of global constraint formulation and propagation, and applications in scheduling and planning, building infrastructure design, and vehicle routing. The Machine Learning track contained papers on declaring and learning structured models, constrained data mining, and parameter learning while the Operations Research track incorporated work based on problem decompositions and hybridization of CP with mathematical programming, using integer programming for weighted constraint satisfaction problems, and generating explanations for constraint inference.
What’s Hot at ICAPS 2017
Automated planning has become quite rich and diverse. We survey recent developments in foundations, algorithms, applications, and emerging research.
AI planning has roots in synthesizing simple sequence of discrete, deterministic actions. However, many problems require more sophisticated models of action, time, and change. Recent research has focused on these issues, including: (1) continuous time and resources, employing hybrid algorithms that integrate OR and SMT techniques with traditional planning, (2) plan quality improvement and optimization, using sophisticated cost/reward models, (3) planning under uncertainty, where recent innovations make creative use of occupation measures for LP-based heuristic functions, and (4) planning when domain models are incomplete or inaccurate. Additionally, there are exciting new results in bidirectional search algorithms, and a renewed push towards incorporating learning within planning. Finally, there is increased attention on devising planning systems that work with people, involving work on human interfaces for planning, explainability, cooperative planning, and plan recognition.
Planning technology is also being applied in diverse areas. In addition to traditional applications like robotics, gaming, and logistics, emerging applications include smart management of traffic, energy grids, and fleets of sharable vehicles like bicycles and cars. Many of these have already been deployed in real systems.
Hot Trends in Satisfiability Testing (SAT)
Serge Gaspers, Toby Walsh
The 20th edition of the International Conference on Theory and Applications of Satisfiability Testing (SAT 2017) took place in Melbourne, Australia from August 28 until September 1st 2017. It was co-located with CP 2017 and ICLP 2017 and immediately followed IJCAI 2017.
We report on some of the highlights of SAT 2017, focusing on recent trends and hot topics. These include the use of dependency schemes in solving quantified Boolean formulas, minimal correction sets for MaxSAT solving, evaluations of branching heuristics for SAT solving, and the use of graph width measures and backdoors for SAT and #SAT.
What’s Hot in Combinatorial Search? (SoCS)
Alex Fukunaga, Akihiro Kishimoto
We give an overview of recent trends in the field of heuristic search as observed in the 10th Annual Symposium on Combinatorial Search (SoCS), held in June 2017.
Combinatorial search plays a crucial part in solving many tasks in AI. SoCS, which was established in 2008 as an annual conference, brings together researchers working in this area to present their results and ideas and to foster collaboration.
The majority of the papers has been on heuristic search and its application. Major topics include heuristics search in planning, any-angle pathfinding, multi-agent pathfinding, real-time search, construction of heuristics (e.g., pattern database), and puzzle solving. In addition, papers on related topics including parallel simulated annealing, SAT solving, solving graph optimization problems with Monte Carlo search, and large-scale parallel pattern mining have been presented at SoCS.
The SoCS 2017 program reflects recent trends in the combinatorial search community. Bidirectional search continues to be investigated. The recent explosion of interest in deep learning in the AI community was reflected by some poster presentations on the application of neural networks to search. Furthermore, the best paper awards indicate that understanding the behavior of greedy best-first search, and better utilization of multiple heuristics are topics of interest.