Policy Optimization with Second-Order Advantage Information

Policy optimization on high-dimensional continuous control tasks exhibits its difficulty caused by the large variance of the policy gradient estimators. We present the action subspace dependent gradient (ASDG) estimator which incorporates the Rao-Blackwell theorem (RB) and Control Variates (CV) into a unified framework to reduce the variance. To invoke RB, our proposed algorithm (POSA) learns the underlying factorization structure among the action space based on the second-order advantage information. POSA captures the quadratic information explicitly and efficiently by utilizing the wide \& deep architecture. Empirical studies show that our proposed approach demonstrates the performance improvements on high-dimensional synthetic settings and OpenAI Gym's MuJoCo continuous control tasks.

Enhanced tensor factorization framework using non-negative and probabilistic tensor factorization approaches for microblogging content propagation modelling

With the aim of identifying the user’s preferences, Content propagation modeling from the micro-blogging sites aids diverse organizations. In existing studies, four user behavior aspects were used by the content propagation model that is to say topic virality, user’s position, user susceptibility and user virality. The propagation occurrences are signified as a tensor factorization model so-called V2S is presented with the aim of deriving the behavioral aspects via which the content propagation is designed. On the other hand, it doesn’t comprise the linguistic patterns in the content that decreases the performance of the content propagation. Moreover, by utilizing advanced tensor approaches, the factorization structure is improved. Therefore the performance of the complete system is decreased meaningfully. With the aim of overcoming the aforesaid problems, Enhanced V2S (EV2S) Tensor Factorization framework is presented in this research that make use of the Probabilistic Latent Tensor Factorization (PLTF) as well as Non-negative Tensor Factorization (NTF) in order to derive the behavioral facets. NTF is presented for decreasing the content propagation errors. By making use of fast gradient descent technique, the unrestrained issue, which happens in this model is solved. This research system identifies the reposts as well as re-tweets in huge datasets proficiently with minimum processing time. From the experimentation outcomes, it is proved that the EV2S-PLTF tensor factorization performs better when compared to the previous tensor frameworks.

Weak Homomorphisms of Coalgebras Beyond Set

We study the notion of weak homomorphisms between coalgebras of different types generalizing thereby that of homomorphisms for similarly typed coalgebras. This helps extend some results known so far in the theory of Universal coalgebra over Set. We find conditions under which coalgebras of a set of types and weak homomorphisms between them form a category. Moreover, we establish an Isomorphism Theorem that extends the so-called First Isomorphism Theorem, showing thereby that this category admits a canonical factorization structure for morphisms

Dispersed Factorization Structures

Factorization structures on a category form a useful categorical tool. As is known, any , satisfying suitable completeness—and smallness—conditions, has a sufficient supply of factorization structures; in fact, there is a bijection between the class of all epireflective (full and isomorphism- closed) subcategories of and the class of all so called perfect factorizationstructures of In this paper, for an arbitrary category supplied with a fixed factorization structure (E, M), a similar bijection between the class of all E-reflective (full and isomorphism-closed) subcategories of and the class of all (E, M)-dispersed factorization structures on , introduced in this paper, will be established.