Search methods are useful in hierarchical task network (HTN) planning to make performance less dependent on the domain knowledge provided, and to minimize plan costs. Here we investigate Monte-Carlo tree search (MCTS) as a new algorithmic alternative in HTN planning. We implement combinations of MCTS with heuristic search in PANDA. We furthermore investigate MCTS in JSHOP, to address lifted (non-grounded) planning, leveraging the fact that, in contrast to other search methods, MCTS does not require a grounded task representation. Our new methods yield coverage performance on par with the state of the art, but in addition can effectively minimize plan cost over time.

It has previously been observed that the verification of safety properties in deterministic model-checking frameworks can be compiled into classical planning. A similar connection exists between goal probability analysis on either side, yet that connection has not been explored. We fill that gap with a translation from Jani, an input language for quantitative model checkers including the Modest toolset and PRISM, into PPDDL. Our experiments motivate further cross-fertilization between both research areas, specifically the exchange of algorithms. Our study also initiates the creation of new benchmarks for goal probability analysis.

Variational encoder-decoders (VEDs) have shown promising results in dialogue generation. However, the latent variable distributions are usually approximated by a much simpler model than the powerful RNN structure used for encoding and decoding, yielding the KL-vanishing problem and inconsistent training objective. In this paper, we separate the training step into two phases: The first phase learns to autoencode discrete texts into continuous embeddings, from which the second phase learns to generalize latent representations by reconstructing the encoded embedding.  In this case, latent variables are sampled by transforming Gaussian noise through multi-layer perceptrons and are trained with a separate VED model, which has the potential of realizing a much more flexible distribution. We compare our model with current popular models and the experiment demonstrates substantial improvement in both metric-based and human evaluations.

This is coarse-grained in that, for each variable, it either remembers all past values (red), or remembers only the most recent one (black). We herein introduce limited-memory state variables, that remember a subset of their most recent values. It turns out that planning is still PSPACE-complete even when the memory is large enough to store all but a single value. Nevertheless, limited memory can be used to substantially broaden a known tractable fragment of red-black planning, yielding better heuristic functions in some domains.

Online social networks being extended to geographical space has resulted in large amount of user check-in data. Understanding check-ins can help to build appealing applications, such as location recommendation. In this paper, we propose DeepCity, a feature learning framework based on deep learning, to profile users and locations, with respect to user demographics and location category prediction. Both of the predictions are essential for social network companies to increase user engagement. The key contribution of DeepCity is the proposal of task-specific random walk which uses the location and user properties to guide the feature learning to be specific to each prediction task. Experiments conducted on 42M check-ins in three cities collected from Instagram have shown that DeepCity achieves a superior performance and outperforms state-of-the-art models significantly.

Maximizing goal probability is an important objective in probabilistic planning, yet algorithms for its optimal solution are severely underexplored. There is scant evidence of what the empirical state of the art actually is. Focusing on heuristic search, we close this gap with a comprehensive empirical analysis of known and adapted algorithms. We explore both, the general case where there may be 0-reward cycles, and the practically relevant special case of acyclic planning, like planning with a limited action-cost budget. We consider three different algorithmic objectives. We design suitable termination criteria, search algorithm variants, dead-end pruning methods using classical planning heuristics, and node selection strategies. Our evaluation on more than 1000 benchmark instances from the IPPC, resource-constrained planning, and simulated penetration testing reveals the behavior of heuristic search, and exhibits several improvements to the state of the art.

We introduce a state space search method that identifies dead-end states, analyzes the reasons for failure, and learns to avoid similar mistakes in the future. Our work is placed in classical planning. The key technique are critical-path heuristics h C , relative to a set C of conjunctions. These recognize a dead-end state s, returning h C (s) = infty, if s has no solution even when allowing to break up conjunctive subgoals into the elements of C. Our key idea is to learn C during search. Starting from a simple initial C, we augment search to identify unrecognized dead-ends s, where h C (s) < infinity. We design methods analyzing the situation at such s, adding new conjunctions into C to obtain h C (s) = infty, thus learning to recognize s as well as similar dead-ends search may encounter in the future. We furthermore learn clauses phi where s' not satisfying phi implies hC(s') = infty, to avoid the prohibitive overhead of computing h C on every search state. Arranging these techniques in a depth-first search, we obtain an algorithm approaching the elegance of clause learning in SAT, learning to refute search subtrees. Our experiments show that this can be quite powerful. On problems where dead-ends abound, the learning reliably reduces the search space by several orders of magnitude.

In optimal planning as heuristic search, admissible pruning techniques are paramount. One idea is dominance pruning, identifying states "better than" other states. Prior approaches are limited to simple dominance notions, like "more STRIPS facts true" or "higher resource supply". We apply simulation, well-known in model checking, to compute much more general dominance relations based on comparing transition behavior across states. We do so effectively by expressing state-space simulations through the composition of simulations on orthogonal projections. We show how simulation can be made more powerful by intertwining it with a notion of label dominance. Our experiments show substantial improvements across several IPC benchmark domains.

Many known planning tasks have inherent constraints concerning the best order in which to achieve the goals. A number of research efforts have been made to detect such constraints and use them for guiding search, in the hope to speed up the planning process. We go beyond the previous approaches by defining ordering constraints not only over the (top level) goals, but also over the sub-goals that will arise during planning. Landmarks are facts that must be true at some point in every valid solution plan. We show how such landmarks can be found, how their inherent ordering constraints can be approximated, and how this information can be used to decompose a given planning task into several smaller sub-tasks. Our methodology is completely domain- and planner-independent. The implementation demonstrates that the approach can yield significant performance improvements in both heuristic forward search and Graphplan-style planning.

RIFO, as has been proposed by Nebel et al., is a method that can automatically detect irrelevant information in planning tasks. The idea is to remove such irrelevant information as a pre-process to planning. While RIFO has been shown to be useful in a number of domains, its main disadvantage is that it is not completeness preserving. Furthermore, the pre-process often takes more running time than nowadays state-of-the-art planners, like FF, need for solving the entire planning task. We introduce the notion of relaxed irrelevance, concerning actions which are never needed within the relaxation that heuristic planners like FF and HSP use for computing their heuristic values. The idea is to speed up the heuristic functions by reducing the action sets considered within the relaxation. Starting from a sufficient condition for relaxed irrelevance, we introduce two preprocessing methods for filtering action sets. The first preprocessing method is proven to be completeness-preserving, and is empirically shown to terminate fast on most of our testing examples. The second method is fast on all our testing examples, and is empirically safe. Both methods have drastic pruning impacts in some domains, speeding up FF's heuristic function, and in effect the planning process.