Text-to-image diffusion models such as Stable Diffusion have recently attracted the interest of many researchers, and inverting the diffusion process can play an important role in better understanding the generative process and how to engineer prompts in order to obtain the desired images. To this end, we introduce the new task of predicting the text prompt given an image generated by a generative diffusion model. We combine a series of white-box and black-box models (with and without access to the weights of the diffusion network) to deal with the proposed task. We propose a novel learning framework comprising of a joint prompt regression and multi-label vocabulary classification objective that generates improved prompts. To further improve our method, we employ a curriculum learning procedure that promotes the learning of image-prompt pairs with lower labeling noise (i.e. that are better aligned), and an unsupervised domain-adaptive kernel learning method that uses the similarities between samples in the source and target domains as extra features. We conduct experiments on the DiffusionDB data set, predicting text prompts from images generated by Stable Diffusion. Our novel learning framework produces excellent results on the aforementioned task, yielding the highest gains when applied on the white-box model. In addition, we make an interesting discovery: training a diffusion model on the prompt generation task can make the model generate images that are much better aligned with the input prompts, when the model is directly reused for text-to-image generation.
相關內容
Spiking Neural P systems are a class of membrane computing models inspired directly by biological neurons. Besides the theoretical progress made in this new computational model, there are also numerous applications of P systems in fields like formal verification, artificial intelligence, or cryptography. Motivated by all the use cases of SN P systems, in this paper, we present a new privacy-preserving protocol that enables a client to compute a linear function using an SN P system hosted on a remote server. Our protocol allows the client to use the server to evaluate functions of the form t_1k + t_2 without revealing t_1, t_2 or k and without the server knowing the result. We also present an SN P system to implement any linear function over natural numbers and some security considerations of our protocol in the honest-but-curious security model.
It is a long-standing desire of industry and research to automate the software development and testing process as much as possible. In this process, requirements engineering (RE) plays a fundamental role for all other steps that build on it. Model-based design and testing methods have been developed to handle the growing complexity and variability of software systems. However, major effort is still required to create specification models from a large set of functional requirements provided in natural language. Numerous approaches based on natural language processing (NLP) have been proposed in the literature to generate requirements models using mainly syntactic properties. Recent advances in NLP show that semantic quantities can also be identified and used to provide better assistance in the requirements formalization process. In this work, we present and discuss principal ideas and state-of-the-art methodologies from the field of NLP in order to guide the readers on how to create a set of rules and methods for the semi-automated formalization of requirements according to their specific use case and needs. We discuss two different approaches in detail and highlight the iterative development of rule sets. The requirements models are represented in a human- and machine-readable format in the form of pseudocode. The presented methods are demonstrated on two industrial use cases from the automotive and railway domains. It shows that using current pre-trained NLP models requires less effort to create a set of rules and can be easily adapted to specific use cases and domains. In addition, findings and shortcomings of this research area are highlighted and an outlook on possible future developments is given.
Many systems and services rely on timing assumptions for performance and availability to perform critical aspects of their operation, such as various timeouts for failure detectors or optimizations to concurrency control mechanisms. Many such assumptions rely on the ability of different components to communicate on time -- a delay in communication may trigger the failure detector or cause the system to enter a less-optimized execution mode. Unfortunately, these timing assumptions are often set with little regard to actual communication guarantees of the underlying infrastructure -- in particular, the variability of communication delays between processes in different nodes/servers. The higher communication variability holds especially true for systems deployed in the public cloud since the cloud is a utility shared by many users and organizations, making it prone to higher performance variance due to noisy neighbor syndrome. In this work, we present Cloud Latency Tester (CLT), a simple tool that can help measure the variability of communication delays between nodes to help engineers set proper values for their timing assumptions. We also provide our observational analysis of running CLT in three major cloud providers and share the lessons we learned.
Hand-crafted image quality metrics, such as PSNR and SSIM, are commonly used to evaluate model privacy risk under reconstruction attacks. Under these metrics, reconstructed images that are determined to resemble the original one generally indicate more privacy leakage. Images determined as overall dissimilar, on the other hand, indicate higher robustness against attack. However, there is no guarantee that these metrics well reflect human opinions, which, as a judgement for model privacy leakage, are more trustworthy. In this paper, we comprehensively study the faithfulness of these hand-crafted metrics to human perception of privacy information from the reconstructed images. On 5 datasets ranging from natural images, faces, to fine-grained classes, we use 4 existing attack methods to reconstruct images from many different classification models and, for each reconstructed image, we ask multiple human annotators to assess whether this image is recognizable. Our studies reveal that the hand-crafted metrics only have a weak correlation with the human evaluation of privacy leakage and that even these metrics themselves often contradict each other. These observations suggest risks of current metrics in the community. To address this potential risk, we propose a learning-based measure called SemSim to evaluate the Semantic Similarity between the original and reconstructed images. SemSim is trained with a standard triplet loss, using an original image as an anchor, one of its recognizable reconstructed images as a positive sample, and an unrecognizable one as a negative. By training on human annotations, SemSim exhibits a greater reflection of privacy leakage on the semantic level. We show that SemSim has a significantly higher correlation with human judgment compared with existing metrics. Moreover, this strong correlation generalizes to unseen datasets, models and attack methods.
Multimodal Large Language Models (MLLMs) that integrate text and other modalities (especially vision) have achieved unprecedented performance in various multimodal tasks. However, due to the unsolved adversarial robustness problem of vision models, MLLMs can have more severe safety and security risks by introducing the vision inputs. In this work, we study the adversarial robustness of Google's Bard, a competitive chatbot to ChatGPT that released its multimodal capability recently, to better understand the vulnerabilities of commercial MLLMs. By attacking white-box surrogate vision encoders or MLLMs, the generated adversarial examples can mislead Bard to output wrong image descriptions with a 22% success rate based solely on the transferability. We show that the adversarial examples can also attack other MLLMs, e.g., a 26% attack success rate against Bing Chat and a 86% attack success rate against ERNIE bot. Moreover, we identify two defense mechanisms of Bard, including face detection and toxicity detection of images. We design corresponding attacks to evade these defenses, demonstrating that the current defenses of Bard are also vulnerable. We hope this work can deepen our understanding on the robustness of MLLMs and facilitate future research on defenses. Our code is available at //github.com/thu-ml/Attack-Bard.
Signal processing and Information theory are two disparate fields used for characterizing signals for various scientific and engineering applications. Spectral/Fourier analysis, a technique employed in signal processing, helps estimation of power at different frequency components present in the signal. Characterizing a time-series based on its average amount of information (Shannon entropy) is useful for estimating its complexity and compressibility (eg., for communication applications). Information theory doesn't deal with spectral content while signal processing doesn't directly consider the information content or compressibility of the signal. In this work, we attempt to bring the fields of signal processing and information theory together by using a lossless data compression algorithm to estimate the amount of information or `compressibility' of time series at different scales. To this end, we employ the Effort-to-Compress (ETC) algorithm to obtain what we call as a Compression Spectrum. This new tool for signal analysis is demonstrated on synthetically generated periodic signals, a sinusoid, chaotic signals (weak and strong chaos) and uniform random noise. The Compression Spectrum is applied on heart interbeat intervals (RR) obtained from real-world normal young and elderly subjects. The compression spectrum of healthy young RR tachograms in the log-log scale shows behaviour similar to $1/f$ noise whereas the healthy old RR tachograms show a different behaviour. We envisage exciting possibilities and future applications of the Compression Spectrum.
Deep learning techniques have received much attention in the area of image denoising. However, there are substantial differences in the various types of deep learning methods dealing with image denoising. Specifically, discriminative learning based on deep learning can ably address the issue of Gaussian noise. Optimization models based on deep learning are effective in estimating the real noise. However, there has thus far been little related research to summarize the different deep learning techniques for image denoising. In this paper, we offer a comparative study of deep techniques in image denoising. We first classify the deep convolutional neural networks (CNNs) for additive white noisy images; the deep CNNs for real noisy images; the deep CNNs for blind denoising and the deep CNNs for hybrid noisy images, which represents the combination of noisy, blurred and low-resolution images. Then, we analyze the motivations and principles of the different types of deep learning methods. Next, we compare the state-of-the-art methods on public denoising datasets in terms of quantitative and qualitative analysis. Finally, we point out some potential challenges and directions of future research.
Compared with cheap addition operation, multiplication operation is of much higher computation complexity. The widely-used convolutions in deep neural networks are exactly cross-correlation to measure the similarity between input feature and convolution filters, which involves massive multiplications between float values. In this paper, we present adder networks (AdderNets) to trade these massive multiplications in deep neural networks, especially convolutional neural networks (CNNs), for much cheaper additions to reduce computation costs. In AdderNets, we take the $\ell_1$-norm distance between filters and input feature as the output response. The influence of this new similarity measure on the optimization of neural network have been thoroughly analyzed. To achieve a better performance, we develop a special back-propagation approach for AdderNets by investigating the full-precision gradient. We then propose an adaptive learning rate strategy to enhance the training procedure of AdderNets according to the magnitude of each neuron's gradient. As a result, the proposed AdderNets can achieve 74.9% Top-1 accuracy 91.7% Top-5 accuracy using ResNet-50 on the ImageNet dataset without any multiplication in convolution layer.
InteractE: Improving Convolution-based Knowledge Graph Embeddings by Increasing Feature Interactions
Most existing knowledge graphs suffer from incompleteness, which can be alleviated by inferring missing links based on known facts. One popular way to accomplish this is to generate low-dimensional embeddings of entities and relations, and use these to make inferences. ConvE, a recently proposed approach, applies convolutional filters on 2D reshapings of entity and relation embeddings in order to capture rich interactions between their components. However, the number of interactions that ConvE can capture is limited. In this paper, we analyze how increasing the number of these interactions affects link prediction performance, and utilize our observations to propose InteractE. InteractE is based on three key ideas -- feature permutation, a novel feature reshaping, and circular convolution. Through extensive experiments, we find that InteractE outperforms state-of-the-art convolutional link prediction baselines on FB15k-237. Further, InteractE achieves an MRR score that is 9%, 7.5%, and 23% better than ConvE on the FB15k-237, WN18RR and YAGO3-10 datasets respectively. The results validate our central hypothesis -- that increasing feature interaction is beneficial to link prediction performance. We make the source code of InteractE available to encourage reproducible research.
The design of deep graph models still remains to be investigated and the crucial part is how to explore and exploit the knowledge from different hops of neighbors in an efficient way. In this paper, we propose a novel RNN-like deep graph neural network architecture by incorporating AdaBoost into the computation of network; and the proposed graph convolutional network called AdaGCN~(AdaBoosting Graph Convolutional Network) has the ability to efficiently extract knowledge from high-order neighbors and integrate knowledge from different hops of neighbors into the network in an AdaBoost way. We also present the architectural difference between AdaGCN and existing graph convolutional methods to show the benefits of our proposal. Finally, extensive experiments demonstrate the state-of-the-art prediction performance and the computational advantage of our approach AdaGCN.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191