Watermarking is the process of embedding information into an image that can survive under distortions, while requiring encoded to have little or no perceptual difference with original image. Recently, deep learning-based methods achieved impressive results in both visual quality and message payload a wide variety distortions. However, these all require differentiable models for distortions at training time, may generalize poorly unknown This undesirable since types applied watermarked images...

Significance Randomized controlled trials are central to the scientific process, but they can be costly. For example, a clinical trial may assign patients treatments that detrimental them. Adaptive experimental designs, such as multiarmed bandit algorithms, reduce costs by increasing probability of assigning promising over course experiment. However, because observations collected these methods dependent and their distribution is nonstationary, statistical inference challenging. We propose...

Watch-time prediction remains to be a key factor in reinforcing user engagement via video recommendations. It has become increasingly important given the ever-growing popularity of online videos. However, watch time not only depends on match between and but is often mislead by duration itself. With goal improving time, recommendation always biased towards videos with long duration. Models trained this imbalanced data face risk bias amplification, which misguides platforms over-recommend...

The wide popularity of short videos on social media poses new opportunities and challenges to optimize recommender systems the video-sharing platforms. Users sequentially interact with system provide complex multi-faceted responses, including WatchTime various types interactions multiple videos. On one hand, platforms aim at optimizing users' cumulative (main goal) in long term, which can be effectively optimized by Reinforcement Learning. other also need satisfy constraint accommodating...

This paper proposes a spatial-Radon domain computed tomography (CT) image reconstruction model based on data-driven tight frames (SRD-DDTF). The proposed SRD-DDTF combines the idea of joint and Radon inpainting Dong, Li, Shen [J. Sci. Comput., 54 (2013), pp. 333--349] that for denoising [J.-F. Cai, H. Ji, Z. Shen, G.-B. Ye, Appl. Comput. Harmon. Anal., 37 (2014), p. 89--105]. It is different from existing models in both CT its corresponding high quality projection are reconstructed...

Classification models play a critical role in data-driven decision-making applications such as medical diagnosis, user profiling, recommendation systems, and default detection. Traditional performance metrics, accuracy, focus on overall error rates but fail to account for the confidence of incorrect predictions, thereby overlooking risk confident misjudgments. This is particularly significant cost-sensitive safety-critical domains like diagnosis autonomous driving, where overconfident false...

Recommender systems are essential for content-sharing platforms by curating personalized content. To evaluate updates of recommender targeting content creators, frequently engage in creator-side randomized experiments to estimate treatment effect, defined as the difference outcomes when a new (vs. status quo) algorithm is deployed on platform. We show that standard difference-in-means estimator can lead biased effect estimate. This bias arises because interference, which occurs treated and...

In a wide variety of applications, including healthcare, bidding in first price auctions, digital recommendations, and online education, it can be beneficial to learn policy that assigns treatments individuals based on their characteristics. The growing policy-learning literature focuses settings which policies are learned from historical data the treatment assignment rule is fixed throughout data-collection period. However, adaptive collection becoming more common practice two primary...

Most existing recommender systems focus primarily on matching users (content consumers) to content which maximizes user satisfaction the platform. It is increasingly obvious, however, that providers have a critical influence through creation, largely determining pool available for recommendation. A natural question thus arises: can we design recommenders taking into account long-term utility of both and providers? By doing so, hope sustain more diverse satisfaction. Understanding full impact...

The Robust Satisficing (RS) model is an emerging approach to robust optimization, offering streamlined procedures and generalization across various applications. However, the statistical theory of RS remains unexplored in literature. This paper fills gap by comprehensively analyzing theoretical properties model. Notably, structure offers a more straightforward path deriving guarantees compared seminal Distributionally Optimization (DRO), resulting richer set results. In particular, we...

We consider the problem of identifying frames in a cardiac ultrasound video associated with left ventricular chamber end-systolic (ES, contraction) and end-diastolic (ED, expansion) phases cycle. Our procedure involves simple application non-negative matrix factorization (NMF) to series from single patient. Rank-2 NMF is performed compute two end-members. The end members are shown be close representations actual heart morphology at each phase function. Moreover, entire time can represented...

Long-term engagement is preferred over immediate in sequential recommendation as it directly affects product operational metrics such daily active users (DAUs) and dwell time. Meanwhile, reinforcement learning (RL) widely regarded a promising framework for optimizing long-term recommendation. However, due to expensive online interactions, very difficult RL algorithms perform state-action value estimation, exploration feature extraction when engagement. In this paper, we propose ResAct which...

It has become increasingly common for data to be collected adaptively, example using contextual bandits. Historical of this type can used evaluate other treatment assignment policies guide future innovation or experiments. However, policy evaluation is challenging if the target differs from one collect data, and popular estimators, including doubly robust (DR) plagued by bias, excessive variance, both. In particular, when pattern in looks little like generated evaluated, importance weights...

Adaptive experiment designs can dramatically improve statistical efficiency in randomized trials, but they also complicate inference. For example, it is now well known that the sample mean biased adaptive trials. Inferential challenges are exacerbated when our parameter of interest differs from trial was designed to target, such as we interested estimating value a sub-optimal treatment after running determine optimal using stochastic bandit design. In this context, typical estimators use...

The wide popularity of short videos on social media poses new opportunities and challenges to optimize recommender systems the video-sharing platforms. Users provide complex multi-faceted responses towards recommendations, including watch time various types interactions with videos. As a result, established recommendation algorithms that concern single objective are not adequate meet this demand optimizing comprehensive user experiences. In paper, we formulate problem video as constrained...

Ranking algorithms are fundamental to various online platforms across e-commerce sites content streaming services. Our research addresses the challenge of adaptively ranking items from a candidate pool for heterogeneous users, key component in personalizing user experience. We develop response model that considers diverse preferences and varying effects item positions, aiming optimize overall satisfaction with ranked list. frame this problem within contextual bandits framework, each list as...

Distributionally robust policy learning aims to find a that performs well under the worst-case distributional shift, and yet most existing methods for consider joint distribution of covariate outcome. The joint-modeling strategy can be unnecessarily conservative when we have more information on source shifts. This paper studiesa nuanced problem -- concept drift, only conditional relationship between outcome changes. To this end, first provide doubly-robust estimator evaluating average reward...

In many applications, e.g. in healthcare and e-commerce, the goal of a contextual bandit may be to learn an optimal treatment assignment policy at end experiment. That is, minimize simple regret. However, this objective remains understudied. We propose new family computationally efficient algorithms for stochastic setting, where tuning parameter determines weight placed on cumulative regret minimization (where we establish near-optimal minimax guarantees) versus state-of-the-art guarantees)....

Watermarking is the process of embedding information into an image that can survive under distortions, while requiring encoded to have little or no perceptual difference from original image. Recently, deep learning-based methods achieved impressive results in both visual quality and message payload a wide variety distortions. However, these all require differentiable models for distortions at training time, may generalize poorly unknown This undesirable since types applied watermarked images...

Most existing recommender systems focus primarily on matching users to content which maximizes user satisfaction the platform. It is increasingly obvious, however, that providers have a critical influence through creation, largely determining pool available for recommendation. A natural question thus arises: can we design recommenders taking into account long-term utility of both and providers? By doing so, hope sustain more diverse satisfaction. Understanding full impact recommendations...