The analysis and optimization of complex systems can be reduced to mathematical problems collectively known as combinatorial optimization. Many such problems can be mapped onto ground-state search problems of the Ising model, and various artificial spin systems are now emerging as promising approaches. However, physical Ising machines have suffered from limited numbers of spin-spin couplings because of implementations based on localized spins, resulting in severe scalability problems. We report a 2000-spin network with all-to-all spin-spin couplings. Using a measurement and feedback scheme, we coupled time-multiplexed degenerate optical parametric oscillators to implement maximum cut problems on arbitrary graph topologies with up to 2000 nodes.

We consider the problem of influence maximization in fixed networks, for both stochastic and adversarial contagion models. The common goal is to select a subset of nodes of a specified size to infect so that the number of infected nodes at the conclusion of the epidemic is as large as possible. In the stochastic setting, the epidemic spreads according to a general triggering model, which includes the popular linear threshold and independent cascade models. We establish upper and lower bounds for the influence of an initial subset of nodes in the network, where the influence is defined as the expected number of infected nodes. Although the problem of exact influence computation is NP-hard in general, our bounds may be evaluated efficiently, leading to scalable algorithms for influence maximization with rigorous theoretical guarantees. In the adversarial spreading setting, an adversary is allowed to specify the edges through which contagion may spread, and the player chooses sets of nodes to infect in successive rounds. Both the adversary and player may behave stochastically, but we limit the adversary to strategies that are oblivious of the player's actions. We establish upper and lower bounds on the minimax pseudo-regret in both undirected and directed networks.

Real-time agent-centered heuristic search is a well-studied problem where an agent that can only reason locally about the world must travel to a goal location using bounded computation and memory at each step. Many algorithms have been proposed for this problem and theoretical results have also been derived for the worst-case performance with simple examples demonstrating worst-case performance in practice. Lower bounds, however, have not been widely studied. In this paper we study best-case performance more generally and derive theoretical lower bounds for reaching the goal using LRTA*, a canonical example of a real-time agent-centered heuristic search algorithm. The results show that, given some reasonable restrictions on the state space and the heuristic function, the number of steps an LRTA*-like algorithm requires to reach the goal will grow asymptotically faster than the state space, resulting in ``scrubbing'' where the agent repeatedly visits the same state. We then show that while the asymptotic analysis does not hold for more complex real-time search algorithms, experimental results suggest that it is still descriptive of practical performance.

Suboptimal heuristic search algorithms such as weighted A* and greedy best-first search are widely used to solve problems for which guaranteed optimal solutions are too expensive to obtain. These algorithms crucially rely on a heuristic function to guide their search. However, most research on building heuristics addresses optimal solving. In this paper, we illustrate how established wisdom for constructing heuristics for optimal search can fail when considering suboptimal search. We consider the behavior of greedy best-first search in detail and we test several hypotheses for predicting when a heuristic will be effective for it. Our results suggest that a predictive characteristic is a heuristic's goal distance rank correlation (GDRC), a robust measure of whether it orders nodes according to distance to a goal. We demonstrate that GDRC can be used to automatically construct abstraction-based heuristics for greedy best-first search that are more effective than those built by methods oriented toward optimal search. These results reinforce the point that suboptimal search deserves sustained attention and specialized methods of its own.

Unavoidable dead-ends are common in many probabilistic planning problems, e.g. when actions may fail or when operating under resource constraints. An important objective in such settings is MaxProb, determining the maximal probability with which the goal can be reached, and a policy achieving that probability. Yet algorithms for MaxProb probabilistic planning are severely underexplored, to the extent that there is scant evidence of what the empirical state of the art actually is. We close this gap with a comprehensive empirical analysis. We design and explore a large space of heuristic search algorithms, systematizing known algorithms and contributing several new algorithm variants. We consider MaxProb, as well as weaker objectives that we baptize AtLeastProb (requiring to achieve a given goal probabilty threshold) and ApproxProb (requiring to compute the maximum goal probability up to a given accuracy). We explore both the general case where there may be 0-reward cycles, and the practically relevant special case of acyclic planning, such as planning with a limited action-cost budget. We design suitable termination criteria, search algorithm variants, dead-end pruning methods using classical planning heuristics, and node selection strategies. We design a benchmark suite comprising more than 1000 instances adapted from the IPPC, resource-constrained planning, and simulated penetration testing. Our evaluation clarifies the state of the art, characterizes the behavior of a wide range of heuristic search algorithms, and demonstrates significant benefits of our new algorithm variants.

Check out how three Galvanize students put together an IBM-inspired chess AI, and get the files for free on Github. This project focuses on computer science concepts such as data structures and algorithms. Chessnut is the chess engine we are using for all the moves and chess logic. Currently trying to implement multiprocessing as our recursive function uses a lot of computing power so calculating heuristics on board states more than 4 levels deep takes a lot of time. With a depth of 3 levels, our AI makes pretty good moves but also makes a lot of ill-advised ones as well. The AI's chess intelligence is estimated to be at a level 3 out of 9.

We introduce a framework for model learning and planning in stochastic domains with continuous state and action spaces and non-Gaussian transition models. It is efficient because (1) local models are estimated only when the planner requires them; (2) the planner focuses on the most relevant states to the current planning problem; and (3) the planner focuses on the most informative and/or high-value actions. Our theoretical analysis shows the validity and asymptotic optimality of the proposed approach. Empirically, we demonstrate the effectiveness of our algorithm on a simulated multi-modal pushing problem.

Ever since Google announced RankBrain, a machine learning algo, there has been much speculation about what other parts of the search algo Google might use machine learning in. While they have said Penguin is not a machine learning algorithm, there is speculation that other parts of the search algo could incorporate machine learning, especially with comments John Mueller made recently about machine learning being used in a lot of Google's systems. During the breakfast keynote at Pubcon last week with Gary Illyes and Eric Enge, the use of machine learning in the Google algo came up with RankBrain being discussed. Enge then asked "It hasn't taken over the whole algorithm?" I can probably say that it will never take over the whole algorithm, or the core algorithm.

One of the biggest buzzwords around Google and the overall technology market is machine learning. Google uses it with RankBrain for search and in other ways. We asked Gary Illyes from Google in part two of our interview how Google uses machine learning with search. Illyes said that Google uses it mostly for "coming up with new signals and signal aggregations." So they may look at two or more different existing non-machine-learning signals and see if adding machine learning to the aggregation of them can help improve search rankings and quality.