We present a comprehensive solution to learn and improve text-to-image models from human preference feedback. To begin with, we build ImageReward -- the first general-purpose reward model effectively encode preferences. Its training is based on our systematic annotation pipeline including rating ranking, which collects 137k expert comparisons date. In evaluation, outperforms existing scoring metrics, making it promising automatic metric for evaluating synthesis. On top of it, propose Reward...

Generating human mobility trajectories is of great importance to solve the lack large-scale trajectory data in numerous applications, which caused by privacy concerns. However, existing generation methods still require real-world centrally collected as training data, where there exists an inescapable risk leakage. To overcome this limitation, paper, we propose PateGail, a privacy-preserving imitation learning model generate trajectories, utilizes powerful generative adversary simulate...

Synthesized human trajectories are instrumental for a large number of applications. However, existing trajectory synthesizing models limited in either modeling variable-length with continuous temporal distribution or incorporating multi-dimensional context information. In this paper, we propose novel probabilistic model based on the variational point process to synthesize trajectories. This combines classical neural inference framework, leading its strong ability distribution, variable...

This paper proposes an effective material delivery algorithm to address the challenges associated with Unmanned Aerial Vehicle (UAV) transportation and delivery, which include complex route planning, low detection precision, hardware limitations. novel approach integrates Whale-Swarm Hybrid Algorithm (WSHA) APCR-YOLOv8 model enhance efficiency accuracy. For path placement paths are transformed into a Generalized Traveling Salesman Problem (GTSP) be able compute solutions. The Whale...

Visual Teach and Repeat 3 (VT&R3), a generalization of stereo VT&R, achieves long-term autonomous path-following using topometric mapping localization from single rich sensor stream. In this paper, we improve the capabilities LiDAR implementation VT &R3 to reliably detect avoid obstacles in changing environments. Our architecture simplifies obstacle-perception problem that place-dependent change detection. We then extend behaviour generic sample-based motion planners better suit...

In recent years, time series forecasting has been widely used in various fields, especially financial markets. Stock trend become one of the most common and complex challenges faced by investors researchers. However, much current research relies primarily on single-granularity stock data for forecasting, with relatively few studies multi-granularity fewer spatial correlation data. This inherent limitation restricts comprehensive extraction valuable information. To address this challenge, we...

Universities, enterprises, and students are the key subjects in talent training system of Big Data.This study used text mining, interviews, questionnaires, other methods to analyze characteristics deficiencies Chinese universities Data talents, requirements enterprises on professional quality big data cognition ability talents.Furthermore, this theory plan-action-inspection-action cycle evaluate cultivation management China.The results showed that employees with rich background proficiency...

The noninvasive diagnosis of cholangiocarcinoma (CCA) is insufficiently accurate. Therefore, the discovery new prognostic markers vital for understanding CCA mechanism and related treatment. information on patients in Cancer Genome Atlas database was used weighted gene co-expression network analysis. Gene Ontology (GO) analysis Kyoto Encyclopedia Genes Genomes (KEGG) pathway were applied to analyze modules interest. By using receiver operating characteristic (ROC) Human Protein (HPA),...

Gene expression profiles are essential in identifying different cancer phenotypes. Clustering gene datasets can provide accurate identification of cancerous cell lines, but this task is challenging due to the small sample size and high dimensionality. Using $K$-means clustering algorithm we determine organisation solution space for a variety using energy landscape theory. The landscapes allow us understand performance, guide more effective use when varying common dataset properties; number...

为确定底泥和藻体在太湖湖泛形成过程中对致黑物形成的贡献,采用室内模拟系统,研究底泥、蓝藻以及底泥+蓝藻3种处理,模拟湖水在不同厌氧程度下湖泛特征参数(黑度、铁及硫形态)的变化,分析不同处理以及受不同聚藻程度影响区(八房港、焦山)底泥对湖泛的诱发作用及致黑物供给潜力.结果表明,各处理组诱发太湖湖泛发生的难易顺序为:底泥+蓝藻处理组> 底泥处理组> 蓝藻处理组.底泥+蓝藻处理组中Fe²⁺浓度为蓝藻处理组的11~94倍,其平均浓度为后者的33倍,而底泥+蓝藻处理组中还原性硫浓度为其他处理组的2~56倍.研究还发现,聚藻区底泥较非聚藻区更易发生湖泛,这是由于聚藻区底泥富集了更高浓度的铁、硫等还原性物质,但厌氧处理与非厌氧处理在诱发湖泛发生的风险差异不明显.以上结果证实,厌氧环境下低价铁硫供应潜力的差异是决定湖泛发生的主要物质来源,加强对聚藻区底泥及蓝藻的控制是有效防控太湖湖泛发生的主要措施之一.;A laboratory simulation system was established to identify the contribution of algal...

Zero-Shot Sketch-Based Image Retrieval (ZS-SBIR) is a challenging cross-modal retrieval task. In prior arts, the conducted by sorting distance between query sketch and each image in gallery. However, domain gap zero-shot setting make neural networks hard to generalize. This paper tackles challenges from new perspective: utilizing gallery features. We propose Cluster-then-Retrieve (ClusterRetri) method that performs clustering on images uses cluster centroids as proxies for retrieval....

With increasing number of young target customers uses social media and network, their point view about specific brands on internet become more important than past. Brand image customer relationship management a new focus for brand. Using case study, this paper proposes the between brand level messages distribution, recommendations to maximize impact distribution get in touch with customers. This examines whether high levels messaging always have positive perception hypothesis message...

Sampling from the posterior is a key technical problem in Bayesian statistics. Rigorous guarantees are difficult to obtain for Markov Chain Monte Carlo algorithms of common use. In this paper, we study an alternative class based on diffusion processes. The constructed such way that, at its final time, it approximates target distribution. stochastic differential equation that defines process discretized (using Euler scheme) provide efficient sampling algorithm. Our construction notion...

A substantial body of empirical work documents the lack robustness in deep learning models to adversarial examples. Recent theoretical proved that examples are ubiquitous two-layers networks with sub-exponential width and ReLU or smooth activations, multi-layer width. We present a result same type, no restriction on for general locally Lipschitz continuous activations. More precisely, given neural network $f(,\cdot,;\mathbf{\theta})$ random weights $\mathbf{\theta}$, feature vector...

Research on social movements has been continuously robust and of significant theoretical importance in the field studies. This paper argues that there is no definitive criterion for categorizing into a binary standard either success or failure. Social emerge to instigate societal ideological transformations. By analysing various historical worldwide, study finds when reach extreme positions unleash devastating forces upon society, they often provoke counter-movements seek restore...

Automatic identification of seismic phases is one the important tasks in earthquake rapid reporting and early warning. The STA/LTA method currently most widely used automatic pickup method. However, accuracy stability its results heavily depend on selection feature functions, time window lengths, trigger thresholds, so it to some extent impossible achieve P-wave arrival time. We present an improved algorithm for automatically picking up times. In this scheme, weight factor K introduced...

We analyze the statistical properties of generalized cross-validation (GCV) and leave-one-out (LOOCV) applied to early-stopped gradient descent (GD) in high-dimensional least squares regression. prove that GCV is generically inconsistent as an estimator prediction risk GD, even for a well-specified linear model with isotropic features. In contrast, we show LOOCV converges uniformly along GD trajectory risk. Our theory requires only mild assumptions on data distribution does not require...

Diffusion models benefit from instillation of task-specific information into the score function to steer sample generation towards desired properties. Such is coined as guidance. For example, in text-to-image synthesis, text input encoded guidance generate semantically aligned images. Proper inputs are closely tied performance diffusion models. A common observation that strong promotes a tight alignment information, while reducing diversity generated samples. In this paper, we provide first...

Download This Paper Open PDF in Browser Add to My Library Share: Permalink Using these links will ensure access this page indefinitely Copy URL DOI

Diffusion models play a pivotal role in contemporary generative modeling, claiming state-of-the-art performance across various domains. Despite their superior sample quality, mainstream diffusion-based stochastic samplers like DDPM often require large number of score function evaluations, incurring considerably higher computational cost compared to single-step generators adversarial networks. While several acceleration methods have been proposed practice, the theoretical foundations for...