Generative diffusion models have achieved spectacular performance in many areas of machine learning and generative modeling. While the fundamental ideas behind these models come from non-equilibrium physics, variational inference, and stochastic calculus, in this paper we show that many aspects of these models can be understood using the tools of equilibrium statistical mechanics. Using this reformulation, we show that generative diffusion models undergo second-order phase transitions corresponding to symmetry breaking phenomena. We show that these phase transitions are always in a mean-field universality class, as they are the result of a self-consistency condition in the generative dynamics. We argue that the critical instability arising from these phase transitions lies at the heart of their generative capabilities, which are characterized by a set of mean-field critical exponents. Finally, we show that the dynamic equation of the generative process can be interpreted as a stochastic adiabatic transformation that minimizes the free energy while keeping the system in thermal equilibrium.

Due to the high dimensional integration over latent variables, computing marginal likelihood and posterior distributions for the parameters of a general hierarchical model is a difficult task. The Markov Chain Monte Carlo (MCMC) algorithms are commonly used to approximate the posterior distributions. These algorithms, though effective, are computationally intensive and can be slow for large, complex models. As an alternative to the MCMC approach, the Laplace approximation (LA) has been successfully used to obtain fast and accurate approximations to the posterior mean and other derived quantities related to the posterior distribution. In the last couple of decades, LA has also been used to approximate the marginal likelihood function and the posterior distribution. In this paper, we show that the bias in the Laplace approximation to the marginal likelihood has substantial practical consequences.

With the rapid development of quantum communication technologies, controlled double-direction cyclic (CDDC) quantum communication has become an important research direction. However, how to choose an appropriate quantum state as a channel to achieve double-direction cyclic (DDC) quantum communication for multi-particle entangled states remains an unresolved challenge. This study aims to address this issue by constructing a suitable quantum channel and investigating the DDC quantum communication of two-particle states. Initially, we create a 25-particle entangled state using Hadamard and controlled-NOT (CNOT) gates, and provide its corresponding quantum circuit implementation. Based on this entangled state as a quantum channel, we propose two new four-party CDDC schemes, applied to quantum teleportation (QT) and remote state preparation (RSP), respectively. In both schemes, each communicating party can synchronously transmit two different arbitrary two-particle states to the other parties under supervisory control, achieving controlled quantum cyclic communication in both clockwise and counterclockwise directions. Additionally, the presented two schemes of four-party CDDC quantum communication are extended to situations where n>3 communicating parties. In each proposed scheme, we provide universal analytical formulas for the local operations of the sender, supervisor, and receiver, demonstrating that the success probability of each scheme can reach 100%. These schemes only require specific two-particle projective measurements, single-particle von Neumann measurements, and Pauli gate operations, all of which can be implemented with current technologies. We have also evaluated the inherent efficiency, security, and control capabilities of the proposed schemes. In comparison to earlier methods, the results demonstrate that our schemes perform exceptionally well. This study provides a theoretical foundation for bidirectional controlled quantum communication of multi-particle states, aiming to enhance security and capacity while meeting the diverse needs of future network scenarios.

Studying public sentiment during crises is crucial for understanding how opinions and sentiments shift, resulting in polarized societies. We study Weibo, the most popular microblogging site in China, using posts made during the outbreak of the COVID-19 crisis. The study period includes the pre-COVID-19 stage, the outbreak stage, and the early stage of epidemic prevention. We use Llama 3 8B, a large language model, to analyze users’ sentiments on the platform by classifying them into positive, negative, sarcastic, and neutral categories. Analyzing sentiment shifts on Weibo provides insights into how social events and government actions influence public opinion. This study contributes to understanding the dynamics of social sentiments during health crises, fulfilling a gap in sentiment analysis for Chinese platforms. By examining these dynamics, we aim to offer valuable perspectives on digital communication’s role in shaping society’s responses during unprecedented global challenges.

The objective of this study is to develop a mathematical model capable of correctly displaying the dynamics of an individual’s cognitive activity under conditions of external information influence of different modalities. The adequacy of the model is verified by comparing the results of numerical analysis and experimental data. The mathematical basis of such a model is the apparatus of self-oscillating quantum mechanics. To describe algorithms for the processes of transmission, processing, and generation of information, the theory of information images/representations is used. Methods. This article provides a brief description of the proposed theory. The cognitive activity of an individual is represented mathematically as a one-dimensional potential hole with finite walls of different sizes. The internal potential barrier in this model can model the boundary between consciousness and subconsciousness. The authors developed a parameterization of the systems under study taking into account the proposed theory. Next, a mathematical model was developed based on the apparatus of self-oscillatory quantum mechanics. The authors formulated an equation that describes the function of the state of an information image in the process of human cognitive activity. Computer modeling of various variations of information impact was carried out. The authors also conducted a specially designed experiment. Conclusions. The authors have identified characteristic patterns of such processes and shown the oscillating nature of changes in the state function of information images and the appearance of characteristic threshold effects. A comparison of the obtained model and experimental data showed the adequacy of the developed tool (the coincidence of the general dynamics and characteristic patterns was shown).

We investigate the quantum random walk in momentum space of a spinor kicked rotor with a non-Hermitian kicking potential. We find that the variance in momentum distributions transitions from quadratic to linear growth over time for the non-Hermitian case. Correspondingly, the momentum distributions are in the shape of Gaussian wavepackets, providing clear evidence of a classical random walk induced by the non-Hermitian-driven potential. Remarkably, the rate of the linear growth of the variance diverges as the non-Hermitian parameter approaches zero. In the Hermitian case, deviations from the quantum resonance condition dramatically suppress the quadratic growth of the variance, leading to dynamical localization of the quantum walk. Under such quantum non-resonance conditions, the classical random walk is significantly reduced by the non-Hermitian-driven potential. Interestingly, non-Hermiticity enhances quantum entanglement between internal degrees of freedom, while deviations from the quantum resonance condition reduce it. Possible applications of our findings are discussed.

There are complex graph structures and rich textual information on social networks. Text provides important information for various tasks, while graph structures offer multilevel context for the semantics of the text. Contemporary researchers tend to represent these kinds of data by text-attributed graphs (TAGs). Most TAG-based representation learning methods focus on designing frameworks that convey graph structures to large language models (LLMs) to generate semantic embeddings for downstream graph neural networks (GNNs). However, these methods only provide text attributes for nodes, which fails to capture the multilevel context and leads to the loss of valuable information. To tackle this issue, we introduce the Multilevel Context Learner (MCL) model, which leverages multilevel context on social networks to enhance LLMs’ semantic embedding capabilities. We model the social network as a multilevel context textual-edge graph (MC-TEG), effectively capturing both graph structure and semantic relationships. Our MCL model leverages the reasoning capabilities of LLMs to generate semantic embeddings by integrating these multilevel contexts. The tailored bidirectional dynamic graph attention layers are introduced to further distinguish the weight information. Experimental evaluations on six real social network datasets show that the MCL model consistently outperforms all baseline models. Specifically, the MCL model achieves prediction accuracies of 77.98%, 77.63%, 74.61%, 76.40%, 72.89%, and 73.40%, with absolute improvements of 9.04%, 9.19%, 11.05%, 7.24%, 6.11%, and 9.87% over the next best models. These results demonstrate the effectiveness of the proposed MCL model.

Steganography has been widely used in the field of image privacy protection. However, with the advancement of steganalysis techniques, deep learning-based models are now capable of accurately detecting modifications in stego-images, posing a significant threat to traditional steganography. To address this, we propose AGASI, a GAN-based approach for strengthening adversarial image steganography. This method employs an encoder as the generator in conjunction with a discriminator to form a generative adversarial network (GAN), thereby enhancing the robustness of stego-images against steganalysis tools. Additionally, the GAN framework reduces the gap between the original secret image and the extracted image, while the decoder effectively extracts the secret image from the stego-image, achieving the goal of image privacy protection. Experimental results demonstrate that the AGASI method not only ensures high-quality secret images but also effectively reduces the accuracy of neural network classifiers, inducing misclassifications and significantly increasing the embedding capacity of the steganography system. For instance, under PGD attack, the adversarial stego-images generated by the GAN, at higher disturbance levels, successfully maintain the quality of the secret image while achieving an 84.73% misclassification rate in neural network detection. Compared to images with the same visual quality, our method increased the misclassification rate by 23.31%.

Event temporal relation extraction is a crucial task in natural language processing, aimed at recognizing the temporal relations between event triggers in a text. Despite extensive efforts in this area, the existing methods face two main issues. Firstly, the previous models for event temporal relation extraction mainly rely on a classification framework, which fails to output the crucial contextual words necessary for predicting the temporal relations between two event triggers. Secondly, the prior research that formulated natural language processing tasks as text generation problems usually trained the generative models by maximum likelihood estimation. However, this approach encounters potential difficulties when the optimization objective is misaligned with the task performance metrics. To resolve these limitations, we introduce a reinforcement learning-based generative framework for event temporal relation extraction. Specifically, to output the important contextual words from the input sentence for temporal relation identification, we introduce dependency path generation as an auxiliary task to complement event temporal relation extraction. This task is solved alongside temporal relation prediction to enhance model performance. To achieve this, we reformulate the event temporal relation extraction task as a text generation problem, aiming to generate both event temporal relation labels and dependency path words based on the input sentence. To bridge the gap between the optimization objective and task performance metrics, we employ the REINFORCE algorithm to optimize our generative model, designing a novel reward function to simultaneously capture the accuracy of temporal prediction and the quality of generation. Lastly, to mitigate the high variance issue encountered when using the REINFORCE algorithm in multi-task generative model training, we propose a baseline policy gradient algorithm to improve the stability and efficiency of the training process. Experimental results on two widely used datasets, MATRES and TB-DENSE, show that our approach exhibits competitive performance.

Backdoor attacks remain a critical area of focus in machine learning research, with one prominent approach being the introduction of backdoor training injection mechanisms. These mechanisms embed backdoor triggers into the training process, enabling the model to recognize specific trigger inputs and produce predefined outputs post-training. In this paper, we identify a unifying pattern across existing backdoor injection methods in generative models and propose a novel backdoor training injection paradigm. This paradigm leverages a unified loss function design to facilitate backdoor injection across diverse generative models. We demonstrate the effectiveness and generalizability of this paradigm through experiments on generative adversarial networks (GANs) and Diffusion Models. Our experimental results on GANs confirm that the proposed method successfully embeds backdoor triggers, enhancing the model’s security and robustness. This work provides a new perspective and methodological framework for backdoor injection in generative models, making a significant contribution toward improving the safety and reliability of these models.