It is increasingly common in digital environments to use A/B tests to compare the performance of recommendation algorithms. However, such experiments often violate the stable unit treatment value assumption (SUTVA), particularly SUTVA's ''no hidden treatments'' assumption, due to the shared data between algorithms being compared. This results in a novel form of bias, which we term ''symbiosis bias,'' where the performance of each algorithm is influenced by the training data generated by its competitor. In this paper, we investigate three experimental designs--cluster-randomized, data-diverted, and user-corpus co-diverted experiments--aimed at mitigating symbiosis bias. We present a theoretical model of symbiosis bias and simulate the impact of each design in dynamic recommendation environments. Our results show that while each design reduces symbiosis bias to some extent, they also introduce new challenges, such as reduced training data in data-diverted experiments. We further validate the existence of symbiosis bias using data from a large-scale A/B test conducted on a global recommender system, demonstrating that symbiosis bias affects treatment effect estimates in the field. Our findings provide actionable insights for researchers and practitioners seeking to design experiments that accurately capture algorithmic performance without bias in treatment effect estimates introduced by shared data.

As a secondary generation method, video recording will cause irreversible damage to the watermark within the video, which has always been challenging in video forensics. Although many video watermarking methods are reported in the literature, these methods, however, still cannot well resist camera recording. This has motivated the authors in this paper to introduce a new video watermarking method to resist camera recording. For the proposed method, two watermarks, i.e., copyright watermark and synchronization watermark, are embedded into the well-selected frequency domain coefficients. The synchronization watermark is used to ensure that the copyright watermark can be successfully extracted at the decoder side. To extract the copyright watermark without manual assistance, a neural network based segmentation model is applied to identify the watermarked video-playing region in the camera-recorded video. Meanwhile, automatic perspective correction is performed on the watermarked video-playing region so that the watermark information can be extracted accurately. The experiments show that the watermark data can be embedded into the raw video successfully and extracted from the camera-recorded video accurately by applying the proposed method. And, the proposed method significantly outperforms related works in terms of robustness in different scenarios, which has verified the superiority and applicability of the proposed method.

To address challenges in large-scale deployments and real-time wave observations using Global Navigation Satellite System (GNSS) buoys, we assessed the reliability and precision of the single-frequency L1 time-differenced carrier phase (TDCP) method for determining significant wave heights (SWHs) and average wave periods. Utilizing simulated dynamic velocity measurements derived from International GNSS Service static observations, this study demonstrates that the L1 TDCP method exhibits superior performance compared to dual-frequency TDCP, postprocessed kinematic (PPK), and Doppler algorithms. Notably, L1 TDCP achieves approximately five times greater precision than the Doppler method in vertical direction assessments. In the offshore waters of Qingdao, China, we deployed two GNSS buoys for wave observation experiments, complemented by a nearby accelerometer-equipped wave buoy (WB) for external validation. By integrating the velocity measurements obtained from L1 TDCP, we derived the sea surface displacement, which, after applying a high-pass filter, isolated the wave components. The experimental results confirmed a strong correlation between L1 TDCP and PPK methods in capturing SWH and wave periods. This study found a negligible difference between L1 TDCP and PPK in SWH (0 ± 7.1 mm, 99% correlation) and in average wave period (0 ± 0.13 s, 97% correlation). The wave observations from two independent GNSS buoys at the same location also exhibited remarkable consistency, with a difference in SWH of 0.0 ± 2.5 cm and in average wave period of 0.0 ± 0.1 s. Compared to the independent WB, the L1 TDCP method demonstrated a difference in SWH of$- 1.9\,\,\pm \,\,5.8\,\,\text {cm}$with a 93% correlation and in average period of 0 ± 0.1 s with a 94% correlation. Further, our analysis into GNSS data sampling frequencies highlighted the efficacy of 1-Hz GNSS data in wave inversion though certain spectral limitations were noted. Implementing the L1 TDCP method at 1 Hz offers a substantial reduction in associated hardware costs, storage requirements, and computational demands. Notably, in contrast to PPK, it negates the need for delayed precise ephemeris, facilitating real-time computations. This study highlights the potential of L1 TDCP for broader, real-time GNSS buoy wave observations, harmonizing with the United Nations Decade of Ocean Science’s aspirations for an augmented Global Ocean Observing System.

The topography of the seafloor is highly correlated with the local gravity through intrinsically nonlinear relationships across a particular wavelength band. The purpose of this study is to compare a fully connected deep neural network (FC-DNN) and a convolutional neural network (CNN) with the gravity-geologic method (GGM) to determine whether deep learning can provide superior predictions of bathymetry. We include the short-wavelength gravity and geological models as training parameters, and assess the performance of different models and parameter combinations using various inputs. Compared with the CNN method, the FC-DNN with the short-wavelength gravity as an input reduces the standard deviation of bathymetry differences from 118.6 m to about 73.5 m. The FC-DNN with short-wavelength gravity reduces the standard deviation of bathymetry differences by up to 13.3% compared with the conventional GGM. Furthermore, we demonstrate that the addition of geological information alongside the short-wavelength gravity does not significantly enhance the accuracy. Power spectral density analysis suggests that the FC-DNN is superior for predicting wavelengths shorter than 6 km.

The satellite altimetry missions could measure high-accuracy sea surface heights (SSHs) that can be used to recover the marine gravity field. Traditional methods for estimating the marine gravity field from SSHs all rely on approximate physical correlations between SSHs and gravity, which may neglect nature’s complex nonlinearity. This work presents a new deep network-based method to recover the gravity anomaly. This new method uses a multi-channel convolutional neural network (MCCNN) architecture to capture the nolinear features between ship-borne gravity and a group of input parameters including deflections of the vertical (DOVs), submarine topography and the geo-locations. To validate the gravity, ship-borne gravity anomalies on the two independent cruises were not used in the deep learning process. For comparison, we also estimated the gravity using the traditional inverse Vening Meinesz (IVM) method. Our results indicate that the MCCNN method can derive high-quality marine gravity anomalies. The assessments using 1 mGal-accuracy ship-borne gravity anomalies show that the average accuracy for gravity from MCCNN method is higher than 3 mGal and this method achieves 0.05-0.50 mGal improvement over benchmark methods IVM. Assessed by marine gravity anomaly models with the accuracy of 1-2 mGal, the MCCNN method has shown to improve the accuracy of gravity by at least 4%. Comparisons with the IVM results show that improvements of the MCCNN method were mainly in wavelengths between 8 km and 100 km due to the using of bathymetry. The results show that our deep learning method maintains good performance and is promising for gravity recovering.