Google

About: Google is a company organization based out in Mountain View, California, United States. It is known for research contribution in the topics: Reinforcement learning & Artificial neural network. The organization has 25678 authors who have published 39857 publications receiving 2161734 citations. The organization is also known as: Google Inc. & Google LLC.

Parallelized egocentric fields for autonomous navigation

Abstract: In this paper, we propose a general framework for local path-planning and steering that can be easily extended to perform high-level behaviors. Our framework is based on the concept of affordances: the possible ways an agent can interact with its environment. Each agent perceives the environment through a set of vector and scalar fields that are represented in the agent’s local space. This egocentric property allows us to efficiently compute a local space-time plan and has better parallel scalability than a global fields approach. We then use these perception fields to compute a fitness measure for every possible action, defined as an affordance field. The action that has the optimal value in the affordance field is the agent’s steering decision. We propose an extension to a linear space-time prediction model for dynamic collision avoidance and present our parallelization results on multicore systems. We analyze and evaluate our framework using a comprehensive suite of test cases provided in SteerBench and demonstrate autonomous virtual pedestrians that perform steering and path planning in unknown environments along with the emergence of high-level responses to never seen before situations.

Two-stage Robust Network Design with Exponential Scenarios

Abstract: We study two-stage robust variants of combinatorial optimization problems on undirected graphs, like Steiner tree, Steiner forest, and uncapacitated facility location. Robust optimization problems, previously studied by Dhamdhere et al. (Proc. of 46th Annual IEEE Symposium on Foundations of Computer Science (FOCS'05), pp. 367---378, 2005), Golovin et al. (Proc. of the 23rd Annual Symposium on Theoretical Aspects of Computer Science (STACS), 2006), and Feige et al. (Proc. of the 12th International Integer Programming and Combinatorial Optimization Conference, pp. 439---453, 2007), are two-stage planning problems in which the requirements are revealed after some decisions are taken in Stage 1. One has to then complete the solution, at a higher cost, to meet the given requirements. In the robust k-Steiner tree problem, for example, one buys some edges in Stage 1. Then k terminals are revealed in Stage 2 and one has to buy more edges, at a higher cost, to complete the Stage 1 solution to build a Steiner tree on these terminals. The objective is to minimize the total cost under the worst-case scenario. In this paper, we focus on the case of exponentially many scenarios given implicitly. A scenario consists of any subset of k terminals (for k-Steiner tree), or any subset of k terminal-pairs (for k-Steiner forest), or any subset of k clients (for facility location). Feige et al. (Proc. of the 12th International Integer Programming and Combinatorial Optimization Conference, pp. 439---453, 2007) give an LP-based general framework for approximation algorithms for a class of two stage robust problems. Their framework cannot be used for network design problems like k-Steiner tree (see later elaboration). Their framework can be used for the robust facility location problem, but gives only a logarithmic approximation. We present the first constant-factor approximation algorithms for the robust k-Steiner tree (with exponential number of scenarios) and robust uncapacitated facility location problems. Our algorithms are combinatorial and are based on guessing the optimum cost and clustering to aggregate nearby vertices. For the robust k-Steiner forest problem on trees and with uniform multiplicative increase factor for Stage 2 (also known as inflation), we present a constant approximation. We show APX-hardness of the robust min-cut problem (even with singleton-set scenarios), resolving an open question of (Dhamdhere et al. in Proc. of 46th Annual IEEE Symposium on Foundations of Computer Science (FOCS'05), pp. 367---378, 2005) and (Golovin et al. in Proc. of the 23rd Annual Symposium on Theoretical Aspects of Computer Science (STACS), 2006).

User interface for selecting options

Abstract: Icons can be combined to form toolbars such as those appearing on computer desktops and within application windows such as web browsers. Small icons are desired because desktop real estate is expensive. Small icons can be difficult to select or open because it can be hard to confirm which icon is indicated. An icon is indicated when the desktop cursor hovers over it. Altering the indicated icon provides good user feedback. Icons can be changed varying their size. Non-indicated icons can revert to a minimum size while an indicated one grows to a maximum size. Different images can be displayed for different sized icons, producing a more pleasing appearance. Alternatively, icons can be presented as display images produced by morphing two images together using morphing functions and icon weights. Morphing includes growing and shrinking as well as color changes, animating, and so forth.

Event correlation in cloud computing

Abstract: The present invention pertains to cloud computing systems and handling of events that occur in the cloud. A correlation system obtains and analyzes various events to obtain comprehensive information about the behavior of the cloud. An event repository receives and maintains time-stamped events, which may be obtained from the cloud itself or from external sources reporting on the cloud. A query engine is employed to select, filter and aggregate events from the event repository. The query engine may take into account metadata which describes relationships between different parts of the cloud. Results from the query engine may be presented on a display or otherwise reported. Using such information, the system may fix known problems or change certain parameters to improve the cloud computing process.

Machine Learning-Based Restart Policy for CDCL SAT Solvers

Abstract: Restarts are a critically important heuristic in most modern conflict-driven clause-learning (CDCL) SAT solvers. The precise reason as to why and how restarts enable CDCL solvers to scale efficiently remains obscure. In this paper we address this question, and provide some answers that enabled us to design a new effective machine learning-based restart policy. Specifically, we provide evidence that restarts improve the quality of learnt clauses as measured by one of best known clause quality metrics, namely, literal block distance (LBD). More precisely, we show that more frequent restarts decrease the LBD of learnt clauses, which in turn improves solver performance. We also note that too many restarts can be harmful because of the computational overhead of rebuilding the search tree from scratch too frequently. With this trade-off in mind, between that of learning better clauses vs. the computational overhead of rebuilding the search tree, we introduce a new machine learning-based restart policy that predicts the quality of the next learnt clause based on the history of previously learnt clauses. The restart policy erases the solver’s search tree during its run, if it predicts that the quality of the next learnt clause is below some dynamic threshold that is determined by the solver’s history on the given input. Our machine learning-based restart policy is based on two observations gleaned from our study of LBDs of learnt clauses. First, we discover that high LBD percentiles can be approximated with z-scores of the normal distribution. Second, we find that LBDs, viewed as a sequence, are correlated and hence the LBDs of past learnt clauses can be used to predict the LBD of future ones. With these observations in place, and techniques to exploit them, our new restart policy is shown to be effective over a large benchmark from the SAT Competition 2014 to 2017.