High-resolution human-centric image generation offers significant potential across various industries, such as entertainment, media, and fashion. Diffusion models for text-to-image generation have significantly improved the quality of human image synthesis. However, when scaling to higher resolutions (2K, 4K, and above), they often encounter issues such as object repetition and structural distortion, which appear especially unnatural in human images. To address these challenges, we propose HRHuman, a tuning-free framework for Higher-Resolution Human Image Generation. By leveraging an open-source large human vision model that incorporates rich template knowledge as prior, we first introduce the Prompt Discretization scheme to discretize user-input text prompts, mapping them to image elements and human body parts. Additionally, we implement a Template-guided Prompt Filtering mechanism to align these discretized prompts with regional image semantics, ensuring fine-grained prompt guidance. Extensive experiments demonstrate that HRHuman achieves state-of-the-art performance in human-centeric higher-resolution image generation, significantly addressing both issues of object repetition and structural distortion.

We introduce a weighted and unconstrained variant of the well-known minimum$k$union problem: Given a bipartite graph$\mathcal {G}(U,V,E)$with weights for all nodes in$V$, find a set$S\subseteq V$such that the ratio between the total weight of the nodes in$S$and the number of theirdistinctadjacent nodes in$U$is maximized. Our problem, which we termHeavy Nodes in a Small Neighborhood(HNSN), finds applications in marketing, team formation, and money laundering detection. For example, in the latter application,$S$represents bank account holders who obtain illicit money from some peers of a criminal and route it through their accounts to a target account belonging to the criminal. We prove thatHNSNcan be solved exactly in polynomial time via linear programming. We also develop several algorithms offering different effectiveness/efficiency trade-offs: an exact algorithm, based on node contraction, graph decomposition, and linear programming, as well as three peeling algorithms. The first peeling algorithm is a near-linear time approximation algorithm with a tight approximation ratio, the second is an iterative algorithm that converges to an optimal solution in a very small number of iterations in practice, and the third is a near-linear time greedy heuristic. In addition, we formalize a money laundering scenario involving multiple target accounts and show how our algorithms can be extended to deal with it. Our experiments on real and synthetic datasets show that our algorithms find (near-)optimal solutions, outperforming a natural baseline, and that they can detect money laundering more effectively and efficiently than two state-of-the-art methods.

Time series are ubiquitous in domains ranging from medicine to marketing and finance. Frequent Pattern Mining (FPM) from a time series has thus received much attention. This general problem has been studied under different matching relations determining whether two time series match or not. Recently, it has been studied under the order-preserving (OP) matching relation stating that a match occurs when two time series have the same relative order (i.e., ranks) on their elements. Thus, a frequent OP pattern captures a trend shared by sufficiently many parts of the input time series. Here, we propose exact, highly scalable algorithms for FPM in the OP setting. Our algorithms employ an OP suffix tree (OPST) as an index to store and query time series efficiently. Unfortunately, there are no practical algorithms for OPST construction. Thus, we first propose a novel and practical$\mathcal{O}(n\sigma\log\sigma)$-time and$\mathcal{O}(n)$- space algorithm for constructing the OPST of a length-n time series over an alphabet of size$\sigma$. We also propose an alternative faster OPST construction algorithm running in$\mathcal{O}(n\log\sigma)$time using$\mathcal{O}(n)$space; this algorithm is mainly of theoretical interest. Then, we propose an exact$\mathcal{O}(n)$-time and$\mathcal{O}(n)$-space algorithm for mining all maximal frequent OP patterns, given an OPST. This significantly improves on the state of the art, which takes$\Omega(n^{3})$time in the worst case. We also formalize the notion of closed frequent OP patterns and propose an exact$\mathcal{O}(n)$-time and$\mathcal{O}(n)$-space algorithm for mining all closed frequent OP patterns, given an OPST. We conducted experiments using real-world, multi-million letter time series showing that our$\mathcal{O}(n\sigma\log\sigma)$- time OPST construction algorithm runs in$\mathcal{O}(n)$time on these datasets despite the$\mathcal{O}(n\sigma\log\sigma)$bound; that our frequent pattern mining algorithms are up to orders of magnitude faster than the state of the art and natural Apriori-like baselines; and that OP pattern-based clustering is effective.

Face retouching aims to remove facial blemishes, such as pigmentation and acne, and still retain fine-grain texture details. Nevertheless, existing methods just remove the blemishes but focus little on realism of the intermediate process, limiting their use more to beautifying facial images on social media rather than being effective tools for simulating changes in facial pigmentation and ance. Motivated by this limitation, we propose our Controllable and Gradual Face Retouching (CGFR). Our CGFR is based on physical modelling, adopting Sum-of-Gaussians to approximate skin subsurface scattering in a decomposed melanin and haemoglobin color space. Our CGFR offers a user-friendly control over the facial blemishes, achieving realistic and gradual blemishes retouching. Experimental results based on actual clinical data shows that CGFR can realistically simulate the blemishes’ gradual recovering process.

Explainable recommender system has recently drawn increasing attention due to its capability of providing justification to recommendation. Rather than focusing on certain topics or specific item features, the explanation generated by existing works are too general without the guidance of aspects. However, such information is not given in the practical scenario. To address this issue, we propose a novel Explainable recommender system with BERT-guided explanation generator, named ExBERT to generate reliable explanation with finer granularity. More specifically, a multi-head self-attention based encoder is employed to incorporate pseudo user and item profiles into semantic representation. Moreover, we propose a novel matched explanation prediction task with discriminative ability to enable personalization of the generated sentence. Extensive experiments conducted on two real-world explainable recommendation datasets significantly outperform the state-of-the-art in generation.