
Graph Element Networks: adaptive, structured computation and memory
We explore the use of graph neural networks (GNNs) to model spatial processes in which there is a priori graphical structure. Similar to finite element analysis, we assign nodes of a GNN to spatial locations and use a computational process defined on the graph to model the relationship between an initial function defined over a space and a resulting function in the same space. We use GNNs as a computational substrate, and show that the locations of the nodes in space as well as their connectivity can be optimized to focus on the most complex parts of the space. Moreover, this representational strategy allows the learned inputoutput relationship to generalize over the size of the underlying space and run the same model at different levels of precision, trading computation for accuracy. We demonstrate this method on a traditional PDE problem, a physical prediction problem from robotics, and a problem of learning to predict scene images from novel viewpoints.
04/18/2019 ∙ by Ferran Alet, et al. ∙ 30 ∙ shareread it

Every Local Minimum is a Global Minimum of an Induced Model
For nonconvex optimization in machine learning, this paper proves that every local minimum achieves the global optimality of the perturbable gradient basis model at any differentiable point. As a result, nonconvex machine learning is theoretically as supported as convex machine learning with a handcrafted basis in terms of the loss at differentiable local minima, except in the case when a preference is given to the handcrafted basis over the perturbable gradient basis. The proofs of these results are derived under mild assumptions. Accordingly, the proven results are directly applicable to many machine learning models, including practical deep neural networks, without any modification of practical methods. Furthermore, as special cases of our general results, this paper improves or complements several stateoftheart theoretical results in the literature with a simple and unified proof technique.
04/07/2019 ∙ by Kenji Kawaguchi, et al. ∙ 16 ∙ shareread it

Learning to guide task and motion planning using scorespace representation
In this paper, we propose a learning algorithm that speeds up the search in task and motion planning problems. Our algorithm proposes solutions to three different challenges that arise in learning to improve planning efficiency: what to predict, how to represent a planning problem instance, and how to transfer knowledge from one problem instance to another. We propose a method that predicts constraints on the search space based on a generic representation of a planning problem instance, called scorespace, where we represent a problem instance in terms of the performance of a set of solutions attempted so far. Using this representation, we transfer knowledge, in the form of constraints, from previous problems based on the similarity in score space. We design a sequential algorithm that efficiently predicts these constraints, and evaluate it in three different challenging task and motion planning problems. Results indicate that our approach performs orders of magnitudes faster than an unguided planner
07/26/2018 ∙ by Beomjoon Kim, et al. ∙ 12 ∙ shareread it

FewShot Bayesian Imitation Learning with Logic over Programs
We describe an expressive class of policies that can be efficiently learned from a few demonstrations. Policies are represented as logical combinations of programs drawn from a small domainspecific language (DSL). We define a prior over policies with a probabilistic grammar and derive an approximate Bayesian inference algorithm to learn policies from demonstrations. In experiments, we study five strategy games played on a 2D grid with one shared DSL. After a few demonstrations of each game, the inferred policies generalize to new game instances that differ substantially from the demonstrations. We argue that the proposed method is an apt choice for policy learning tasks that have scarce training data and feature significant, structured variation between task instances.
04/12/2019 ∙ by Tom Silver, et al. ∙ 6 ∙ shareread it

Learning to select examples for program synthesis
Program synthesis is a class of regression problems where one seeks a solution, in the form of a sourcecode program, mapping the inputs to their corresponding outputs exactly. Due to its precise and combinatorial nature, it is commonly formulated as a constraint satisfaction problem, where inputoutput examples are encoded as constraints and solved with a constraint solver. A key challenge of this formulation is scalability: while constraint solvers work well with few wellchosen examples, a large set of examples can incur significant overhead in both time and memory. We address this challenge by constructing a representative subset of examples that is both small and able to constrain the solver sufficiently. We build the subset one example at a time, using a neural network to predict the probability of unchosen inputoutput examples conditioned on the chosen inputoutput examples, and adding the least probable example to the subset. Experiment on a diagram drawing domain shows our approach produces subsets of examples that are small and representative for the constraint solver.
11/09/2017 ∙ by Yewen Pu, et al. ∙ 0 ∙ shareread it

STRIPS Planning in Infinite Domains
Many robotic planning applications involve continuous actions with highly nonlinear constraints, which cannot be modeled using modern planners that construct a propositional representation. We introduce STRIPStream: an extension of the STRIPS language which can model these domains by supporting the specification of blackbox generators to handle complex constraints. The outputs of these generators interact with actions through possibly infinite streams of objects and static predicates. We provide two algorithms which both reduce STRIPStream problems to a sequence of finitedomain planning problems. The representation and algorithms are entirely domain independent. We demonstrate our framework on simple illustrative domains, and then on a highdimensional, continuous robotic task and motion planning domain.
01/01/2017 ∙ by Caelan Reed Garrett, et al. ∙ 0 ∙ shareread it

Learning to Rank for Synthesizing Planning Heuristics
We investigate learning heuristics for domainspecific planning. Prior work framed learning a heuristic as an ordinary regression problem. However, in a greedy bestfirst search, the ordering of states induced by a heuristic is more indicative of the resulting planner's performance than mean squared error. Thus, we instead frame learning a heuristic as a learning to rank problem which we solve using a RankSVM formulation. Additionally, we introduce new methods for computing features that capture temporal interactions in an approximate plan. Our experiments on recent International Planning Competition problems show that the RankSVM learned heuristics outperform both the original heuristics and heuristics learned through ordinary regression.
08/03/2016 ∙ by Caelan Reed Garrett, et al. ∙ 0 ∙ shareread it

Focused ModelLearning and Planning for NonGaussian Continuous StateAction Systems
We introduce a framework for model learning and planning in stochastic domains with continuous state and action spaces and nonGaussian 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 highvalue 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 multimodal pushing problem.
07/26/2016 ∙ by Zi Wang, et al. ∙ 0 ∙ shareread it

BackwardForward Search for Manipulation Planning
In this paper we address planning problems in highdimensional hybrid configuration spaces, with a particular focus on manipulation planning problems involving many objects. We present the hybrid backwardforward (HBF) planning algorithm that uses a backward identification of constraints to direct the sampling of the infinite action space in a forward search from the initial state towards a goal configuration. The resulting planner is probabilistically complete and can effectively construct long manipulation plans requiring both prehensile and nonprehensile actions in cluttered environments.
04/12/2016 ∙ by Caelan Reed Garrett, et al. ∙ 0 ∙ shareread it

Objectbased World Modeling in SemiStatic Environments with Dependent DirichletProcess Mixtures
To accomplish tasks in humancentric indoor environments, robots need to represent and understand the world in terms of objects and their attributes. We refer to this attributebased representation as a world model, and consider how to acquire it via noisy perception and maintain it over time, as objects are added, changed, and removed in the world. Previous work has framed this as multipletarget tracking problem, where objects are potentially in motion at all times. Although this approach is general, it is computationally expensive. We argue that such generality is not needed in typical world modeling tasks, where objects only change state occasionally. More efficient approaches are enabled by restricting ourselves to such semistatic environments. We consider a previouslyproposed clusteringbased world modeling approach that assumed static environments, and extend it to semistatic domains by applying a dependent Dirichletprocess (DDP) mixture model. We derive a novel MAP inference algorithm under this model, subject to data association constraints. We demonstrate our approach improves computational performance in semistatic environments.
12/02/2015 ∙ by Lawson L. S. Wong, et al. ∙ 0 ∙ shareread it

Generalization in Deep Learning
This paper explains why deep learning can generalize well, despite large capacity and possible algorithmic instability, nonrobustness, and sharp minima, effectively addressing an open problem in the literature. Based on our theoretical insight, this paper also proposes a family of new regularization methods. Its simplest member was empirically shown to improve base models and achieve stateoftheart performance on MNIST and CIFAR10 benchmarks. Moreover, this paper presents both datadependent and dataindependent generalization guarantees with improved convergence rates. Our results suggest several new open areas of research.
10/16/2017 ∙ by Kenji Kawaguchi, et al. ∙ 0 ∙ shareread it
Leslie Pack Kaelbling
is this you? claim profile
The American roboticist Leslie Pack Kaelbling is a panasonic professor of computer science and engineering at the Massachusetts Technology Institute. She is widely known for adapting Markov’s partially observable decisionmaking process for operational research for artificial intelligence and robotic application. In 1997, Kaelbling received the IJCAI Computers and Thought Award for enhanced learning in embedded control systems and the development of robot navigation programming tools. In 2000, she was elected Fellow of the Artificial Intelligence Association.
Kaelbling got an A. B. B. In 1983 in Philosophy and Ph.D. In 1990, both at Stanford University in Computer Science. During this time she also became a member of the Center for Language and Information Studies. Before joining Brown University, she then worked at SRI International and the associated robotics spinoff Teleos Research. In 1999, she left Brown to join the professorship at MIT. Her research focuses on policymaking, machine learning and sensing with robotics applications.
She and twothirds of the Kluwerowned journal Machine Learning, both in Spring 2000, resigned in protest at their paytoaccess archives, while the financial compensation for authors was simultaneously limited. Kaelbling cofounded and served as the Journal of Machine Learning Research’s first editorinchief, an open access journal reviewed by peers, covering the same topics. It permits researchers to freely publish and retain copyright in their online archives. In response to the mass resignation, Kluwer amended its publication policy to enable authors to archive their papers online after an online review by peers. Kaelbling replied that this policy was reasonable and would have made the creation of an alternative journal unnecessary, but it was not until the threat of resignation and the founding of the JMLR that the publication policy changed, that the Editorial Board made it clear that they wanted such a policy.