An excellent representation is crucial for reinforcement learning (RL) performance, especially in vision-based reinforcement learning tasks. The quality of the environment representation directly influences the achievement of the learning task. Previous vision-based RL typically uses explicit or implicit ways to represent environments, such as images, points, voxels, and neural radiance fields. However, these representations contain several drawbacks. They cannot either describe complex local geometries or generalize well to unseen scenes, or require precise foreground masks. Moreover, these implicit neural representations are akin to a "black box", significantly hindering interpretability. 3D Gaussian Splatting (3DGS), with its explicit scene representation and differentiable rendering nature, is considered a revolutionary change for reconstruction and representation methods. In this paper, we propose a novel Generalizable Gaussian Splatting framework to be the representation of RL tasks, called GSRL. Through validation in the RoboMimic environment, our method achieves better results than other baselines in multiple tasks, improving the performance by 10%, 44%, and 15% compared with baselines on the hardest task. This work is the first attempt to leverage generalizable 3DGS as a representation for RL.

Recently, humanoid robots have made significant advances in their ability to perform challenging tasks due to the deployment of Reinforcement Learning (RL), however, the inherent complexity of humanoid robots, including the difficulty of designing complicated reward functions and training entire sophisticated systems, still poses a notable challenge. To conquer these challenges, after many iterations and in-depth investigations, we have meticulously developed a full-size humanoid robot, "Adam", whose innovative structural design greatly improves the efficiency and effectiveness of the imitation learning process. In addition, we have developed a novel imitation learning framework based on an adversarial motion prior, which applies not only to Adam but also to humanoid robots in general. Using the framework, Adam can exhibit unprecedented human-like characteristics in locomotion tasks. Our experimental results demonstrate that the proposed framework enables Adam to achieve human-comparable performance in complex locomotion tasks, marking the first time that human locomotion data has been used for imitation learning in a full-size humanoid robot. For more video demonstrations, please visit our YouTube channel: https://www.youtube.com/watch?v=7hK2ySYBa1I

Emergent software container technology is now available on any cloud and opens up new opportunities to execute and scale data intensive applications wherever data is located. However, many traditional relational databases hosted on clouds have not scaled well. In this paper, a framework and deployment methodology to containerize relational SQL queries is presented, so that, a single SQL query can be scaled and executed by a network of cooperating containers, achieving intra-operator parallelism and other significant performance gains. Results of container prototype experiments are reported and compared to a real-world RDBMS baseline. Preliminary result on a research cloud shows up to 3-orders of magnitude performance gain for some queries when compared to running the same query on a single VM hosted RDBMS baseline.

Personal data sharing has emerged as a popular activity on online social networks such as Facebook, Google+, Twitter. As a result, privacy issues have received significant attention in both the research literature and the mainstream media. In this study, we designed a set of questions aimed to learn about user views of online privacy, user knowledge about OSNs privacy settings, and user awareness of privacy disclosure. Our goal is to find out from the users whether and how well users are knowledgable of, satisfied with, and able to effectively use available privacy settings. The information obtained from this study can be used to help OSNs adjust their privacy settings to better match user expectations, and help privacy advocates design better ways to help users control the disclosure of their online information. We collected answers to the questions from a group of 377 users, selected via several methods, who have experiences with multiple OSNs, including Facebook, Google+, and LinkedIn. We analyzed the data with respect to user demographics. Our study shows that 44% of the users lack the knowledge about privacy policies and mechanisms of their OSNs; 34% and 41% of the users, respectively, are seriously and somewhat concern about their privacy protection; and 80% of the users do not think their OSNs have provided sufficient privacy control or default privacy settings that match their expectations. Based on our analysis, we propose several options for OSNs and OSN users to improve the user privacy.

In a fuzzy relational database where a relation is a fuzzy set of tuples and ill-known data are represented by possibility distributions, nested fuzzy queries can be expressed in the Fuzzy SQL language. Although it provides a very convenient way for users to express complex queries, a nested fuzzy query may be very inefficient to process with the naive evaluation method based on its semantics. In conventional databases, nested queries are unnested to improve the efficiency of their evaluation. In this paper, we extend the unnesting techniques to process several types of nested fuzzy queries. An extended merge-join is used to evaluate the unnested fuzzy queries. As shown by both theoretical analysis and experimental results, the unnesting techniques with the extended merge-join significantly improve the performance of evaluating nested fuzzy queries.

Proposes a new measure of fuzzy equality (FE) comparison based on the similarity of possibility distributions. We define a type of fuzzy equi-join based on the new FE comparison and allow threshold values to be associated with predicates of the join condition. A sort-merge join algorithm based on a partial order of intervals is used to evaluate the fuzzy equi-join. In order for the evaluation to be efficient, we identify various mappings, called FE indicators, that determine appropriate intervals for fuzzy data with different characteristics. Experimental results from our preliminary simulation of the algorithm show a significant improvement of efficiency when FE indicators are used with the sort-merge join algorithm.

Given a relational database and a set of fuzzy terms defined for some attributes we consider the problem of mining fuzzy quantitative association rules that may contain crisp values, intervals, and fuzzy terms in both antecedent and consequent. We present an algorithm extended from the equi-depth partition (EDP) algorithm for solving this problem. Our approach combines interval partition with pre-defined fuzzy terms and is more general.

Approaches are proposed to allow fuzzy terms to be interpreted according to the context within which they are used. Such an interpretation is natural and useful. A query-dependent interpretation is proposed to allow a fuzzy term to be interpreted relative to a partial answer of a query. A scaling process is used to transform a pre-defined meaning of a fuzzy term into on appropriate meaning in the given context. Sufficient conditions are given for a nested fuzzy query with RELATIVE quantifiers to be unnested for an efficient evaluation. An attribute-dependent interpretation is proposed to model the applications in which the meaning of a fuzzy term in an attribute must be interpreted with respect to values in other related attributes. Two necessary and sufficient conditions for a tuple to have a unique attribute-dependent interpretation are provided. We describe an interpretation system that allows queries to be processed based on the attribute-dependent interpretation of the data. Two techniques, grouping and shifting, are proposed to improve the implementation.>

The magic rewriting focuses on relevant data but suffers from additional rules, predicates, and tuples that are generated in search for the relevant data. Reducing the arity of predicates can cut down the number of such rules, predicates, and tuples by an exponential factor. In this paper, we consider a subclass of linear single-IDB programs and show that the magic rewriting can be decomposed in such a way that it is applied to only programs having smaller arities and fewer recursive rules, without losing the binding capacity. The decomposed rewriting is shown to be much more efficient than the standard one and amenable to distributed and parallel environments. The considered subclass significantly generalizes recursions previously proposed for efficient implementation. The decomposed rewriting and the standard generalized magic rewriting are extended to multi-binding queries in such a way that data relevant to one binding is not necessarily considered as relevant to other bindings. The work in this paper shows the use of tuple ID as an important technique in optimizing logic programs.

Parallel bottom-up evaluation provides an alternative for the efficient evaluation of logic programs. Existing parallel evaluation strategies are neither effective nor efficient in determining the data to be transmitted among processors. In this paper, we propose re different strategy, for general Datalog programs, that is based on the partitioning of data rather than that of rule instantiations. The partition and processing schemes defined in this paper are more general than those in existing strategies. A parallel evaluation algorithm is given based on the semi-naive bottom-up evaluation. A notion of potential usefulness is recognized as a data transmission criterion to reduce, both effectively and efficiently, the amount of data transmitted. Heuristics and algorithms are proposed for designing the partition and processing schemes for a given program. Results from an experiment show that the strategy proposed in this paper has many promising features.>

A framework has been proposed for evaluating queries in a deductive database system. The framework is based on modeling the intentional database rules and the extensional database relations by using a special class of petri nets called predicate/transition nets. A system of communicating processes is then derived from the predicate/transition net model. Thus, parallel query evaluation becomes the issue of how concurrent processes compute intermediate relations based on the given extensional database relations, communicate and synchronize with each other through message passing to produce the evaluation result. In the framework, the AND/OR parallelisms are exploited with the augmentation of the heuristics of maximizing the bound arguments and utilizing the sideways information passing to restrict the computation to generating those portions of intermediate relations that are relevant to the query. The research results have indicated that the framework is sound and complete and lends itself to a parallel implementation.

Optical Character Recognition (OCR) has become a highly demanded information transfer technology in recent years. A problem of current OCR technology is that texts produced by the state-of-the-art OCR software contain an unacceptable frequency of errors. This prevents the OCR technology from being efficiently used for vast-volume information transfer or daily office operation applications. To correct these errors in a conventional way requires a significant amount of costly human-machine interaction. In this article, we identify and classify the types and distributions of optical recognition errors. We proposed a novel post-processing strategy, based on machine learing techniques, to correct errors resulted from unrecognized process

Linearization of nonlinear recursive programs is an important issue in logic databases for both practical and theoretical reasons. If a nonlinear recursive program can be transformed into an equivalent linear recursive program, then it may be computed more efficiently than when the tranformation is not possible. We provide a set of necessary and sufficient conditions for a simple doubly recursive program to be equivalent to a simple linear recursive program. The necessary and sufficient conditions can be verified effectively.

The problem of converting a simple nonlinear recursive logic query into an equivalent linear one is considered. A general method is given to transform a nonlinear rule into a sequence of linear ones. For efficient processing it is necessary to convert a nonlinear rule into a single linear rule. For such a conversion, a necessary and sufficient condition is provided for a type of doubly recursive rule to be equivalent to the resulting linear rule. It is also shown that a restricted type of higher order recursive rule is equivalent to the linear rule obtained by its conversion.>

In this paper, we study the optimization of linear recursive queries using wavefront methods. The following results are obtained.(i) In spite of seemingly reasonable approach of the wavefront methods, certain linear recursive queries are not processed efficiently or correctly.(ii) A characterization of the expressions generated by linear recursive rules is given. Properties of the expressions will be useful for efficient processing of linear recursive queries.(iii) Conditions for efficient processing of linear recursive rules using wavefront methods with the properties given in (ii) are provided.

Nonlinear recursive queries are usually less efficient in processing than linear recursive queries. It is therefore of interest to transform non-linear recursive queries into linear ones. We obtain a necessary and sufficient condition for a doubly recursive rule of a certain type to be logically equivalent to a single linear recursive rule obtained in a specific way.

We propose a scheme to make use of semantic information to process distributed queries locally without data transfer with respect to the join clauses of the query. Since not all queries can be processed without data transfer, we give an algorithm to recognize the "locally processable queries." For nonlocally processable queries, a simple "fragment and replicate" algorithm is used. The algorithm chooses a relation to remain fragmented at the sites where they are situated while replicating the other relations at those sites. Our algorithm determines the chosen relation and the chosen copy of every fragment of the chosen relation such that the minimum response time is obtained. The algorithm runs in linear time. If the fragments of the relation are allowed to be processed in other sites, then the problem is NP hard. Two heuristics are given for that situation. They are compared to the optimal situation. Experimental results show that the strategies produced by the heuristics have small errors relative to the optimal strategy.