Independence-aware Advantage Estimation

Most of existing advantage function estimation methods in reinforcement learning suffer from the problem of high variance, which scales unfavorably with the time horizon. To address this challenge, we propose to identify the independence property between current action and future states in environments, which can be further leveraged to effectively reduce the variance of the advantage estimation. In particular, the recognized independence property can be naturally utilized to construct a novel importance sampling advantage estimator with close-to-zero variance even when the Monte-Carlo return signal yields a large variance. To further remove the risk of the high variance introduced by the new estimator, we combine it with existing Monte-Carlo estimator via a reward decomposition model learned by minimizing the estimation variance. Experiments demonstrate that our method achieves higher sample efficiency compared with existing advantage estimation methods in complex environments.

High-Accuracy Parameter Estimation for Damped Signal via Three Weighted DFT Samples

Recently, accurate parameter estimation of the damped complex exponential plays an increasingly important role in the field of precise measurement. However, the estimation variance of interpolation-based algorithms for the parameter estimation cannot be asymptotic to the Crámer-Rao lower bound (CRLB). This paper originally proposes a generalized, fast, and the accurate two-iteration estimator (TIE) based on the discrete Fourier transform (DFT). It can be operated by an arbitrary window (symmetric or asymmetric window). Theoretical estimation variances of the frequency and the damping factor for arbitrary windows are derived, respectively. Furthermore, extensive computer simulations are performed to compare the performance of the TIE with other state-of-the-art algorithms in the literature. The results support the theoretical findings and verify that high-accuracy parameter estimation can be ensured by the proposed algorithm. More importantly, the estimation variances returned by the TIE with the rectangle window exactly track the CRLB for a damped single tone.

PERBANDINGAN TIGA PENDEKATAN GEOSTATISTIK UNTUK MEMODELKAN KETIDAKPASTIAN DALAM ESTIMASI SUMBERDAYA TIMAH DAN MINERAL IKUTAN TIMAH PADA ENDAPAN ALUVIAL

Blok C merupakan salah satu blok endapan aluvial di Pulau Bangka yang memiliki prospek timah dan mineral ikutan timah seperti ilmenite, rutile, anatase, zircon, dan monazite. Endapan aluvial umumnya memiliki variabilitas yang tinggi sehingga faktor ketidakpastian akan sumberdaya timah dan mineral ikutan timah juga tinggi. Pada penelitian ini dilakukan perbandingan antara 3 (tiga) pendekatan geostatistik untuk memodelkan ketidakpastian sumberdaya dengan studi kasus pada endapan aluvial di Blok C di Pulau Bangka. Untuk mengetahui variabilitas global di daerah penelitian dilakukan dengan menggunakan metode Global Estimation Variance (GEV), sedangkan untuk mengetahui variabilitas lokal dilakukan menggunakan Sequential Gaussian Simulation (SGS) dan Discrete Gaussian Model (DGM). Hasil dari metode GEV dibandingkan dengan metode SGS dan hasil dari metode SGS juga akan dibandingkan dengan metode DGM. Dari hasil perbandingan GEV dan SGS menunjukkan bahwa hasil GEV cenderung less confidence jika dibandingkan dengan hasil SGS. Less confidence pada hasil GEV disebabkan oleh efek proporsional di daerah penelitian. Hasil perbandingan SGS dan DGM menunjukkan pola yang hampir sama untuk Sn (timah) dan ilmenite+rutile+anatase serta pola yang cukup berbeda untuk zircon. Perbedaan ini disebabkan oleh pemusatan data yang merupakan bagian dari metode DGM. Selain itu, mayoritas nilai minimum hasil DGM lebih besar daripada nilai minimum hasil SGS dan nilai maksimum hasil DGM lebih kecil daripada nilai maksimum hasil SGS. Hal ini disebabkan oleh change of support coefficient (r) yang mempengaruhi fungsi dari transformasi

Remittances and corruption in Nigeria

PurposeIn this paper, we investigate the impact of remittances on control of corruption in Nigeria for a period of 1986–2016.Design/methodology/approachThe study uses ARDL modeling framework, dynamic OLS estimation, variance decomposition and impulse response analysis to examine the relationship between the two variables.FindingsThe study finds that remittances significantly improve the control of corruption in Nigeria. We further examine the robustness test of the results using dynamic OLS estimation, variance decomposition and impulse response analysis. Our results remain significant and consistent to the earlier one reported in ARDL framework which supports the extant literature.Practical implicationsOur study suggests that international remittances can be used, through the cross-border transfer of norms and practices, to significantly impact the socioeconomic progresses of a country by reducing corruption.Originality/valueThe existing studies on the relationship between corruption and remittances document conflicting results. In addition, study on corruption - remittances nexus that specifically focuses on any African country is largely absent despite the fact that most of the countries in the region are recognized as highly corrupt. This paper provides insights on how remittances can be used as part of tool kits to control corruption in African nation.

GCN-LASE: Towards Adequately Incorporating Link Attributes in Graph Convolutional Networks

Graph Convolutional Networks (GCNs) have proved to be a most powerful architecture in aggregating local neighborhood information for individual graph nodes. Low-rank proximities and node features are successfully leveraged in existing GCNs, however, attributes that graph links may carry are commonly ignored, as almost all of these models simplify graph links into binary or scalar values describing node connectedness. In our paper instead, links are reverted to hypostatic relationships between entities with descriptional attributes. We propose GCN-LASE (GCN with Link Attributes and Sampling Estimation), a novel GCN model taking both node and link attributes as inputs. To adequately captures the interactions between link and node attributes, their tensor product is used as neighbor features, based on which we define several graph kernels and further develop according architectures for LASE. Besides, to accelerate the training process, the sum of features in entire neighborhoods are estimated through Monte Carlo method, with novel sampling strategies designed for LASE to minimize the estimation variance. Our experiments show that LASE outperforms strong baselines over various graph datasets, and further experiments corroborate the informativeness of link attributes and our model's ability of adequately leveraging them.