搜索结果：generator

The boundary between real and diffusion-generated time series is becoming increasingly difficult to draw, yet detection in this domain remains underexplored, especially when the generator is unknown. We compare white-box detection, which requires access to the generator, against black-box detection, which operates on the raw signal alone. The white-box approach, a reconstruction-based detector adapted from the image domain, works well in in-distribution but breaks down under generator shift: reconstruction-based detection in images succeeds because large generic generators provide a near-universal reconstruction prior, and no analogous generator exists for time series. In contrast, a simple off-the-shelf classifier used as a black-box detector performs remarkably well, achieving an average F1 of 79.2, a 22.1% relative improvement over the white-box approach, and a TPR@1%FPR of 57.2. Diffusion-generated time series detection is therefore not a direct transfer of the image domain problem. This work provides the first systematic exploration of white-box and black-box detection for diffusion-generated time series. We close by identifying several open and promising directions.

In this paper, we continue the study of the generator graph of a group. In 2023, Tacbobo [9] defined the generator graph of a nontrivial group to be the graph whose vertices are the elements of the group, with two vertices being adjacent if at least one of them is a generator of the group. Building on the properties established in [9], we prove that the diameter of the generator graph of a cyclic group is at most $2$. Furthermore, we present explicit formulas for some topological indices of the generator graph of a cyclic group with $n \ge 2$ elements and whose set of generators is $S$, expressed in terms of $n$ and $|S|$. Lastly, we determine the metric dimension of the generator graph of a nontrivial cyclic group as a function of its order $n$.

This is mostly an expository note about an example communicated to the author by Aise Johan de Jong. In a triangulated category $T$ an object $G$ is said to be a classical generator when the smallest triangulated subcategory containing $G$ coincides with the whole $T$, and it is said to be a generator when the orthogonal complement to $G$ in $T$ is zero, i.e., when any non-zero object of $T$ admits a non-zero map from a shift of $G$. Any classical generator is a generator, but not vice versa. We discuss a simple algebro-geometric example of a non-classical generator in the derived category of coherent sheaves on any smooth proper curve of genus $g \geq 2$. We also overview what is known and what is not known, in general, about generators and classical generators on curves.

We define a bounded local generator class (BLGC) for deterministic state evolution on graph-indexed systems. The construction consists of finite-range generators operating on bounded local state under deterministic composition. Each update acts only on a bounded-radius neighborhood and applies a bounded local transformation with projection onto a compact state domain. Under the BLGC constraints, per-step operator work remains independent of total system size M. Specifically, incremental update cost satisfies $W_t = O(1)$ with respect to $M \to \infty$ for fixed interaction radius $r$. The framework admits a Hilbert-space embedding in $\ell^2(V)\otimes \mathbb{R}^d$ and yields bounded operators under composition on admissible subspaces. The result establishes a structural decoupling between global state capacity and incremental computational work. The claims apply specifically to the bounded local generator class defined in this paper.

We introduce Generator Matching, a modality-agnostic framework for generative modeling using arbitrary Markov processes. Generators characterize the infinitesimal evolution of a Markov process, which we leverage for generative modeling in a similar vein to flow matching: we construct conditional generators which generate single data points, then learn to approximate the marginal generator which generates the full data distribution. We show that Generator Matching unifies various generative modeling methods, including diffusion models, flow matching and discrete diffusion models. Furthermore, it expands the design space to new and unexplored Markov processes such as jump processes. Finally, Generator Matching enables the construction of superpositions of Markov generative models and enables the construction of multimodal models in a rigorous manner. We empirically validate our method on image and multimodal generation, e.g. showing that superposition with a jump process improves performance.

This paper is devoted to a general solvability of a multi-dimensional backward stochastic differential equation (BSDE) of a diagonally quadratic generator $g(t,y,z)$, by relaxing the assumptions of \citet{HuTang2016SPA} on the generator and terminal value. More precisely, the generator $g(t,y,z)$ can have more general growth and continuity in $y$ in the local solution; while in the global solution, the generator $g(t,y,z)$ can have a skew sub-quadratic but in addition "strictly and diagonally" quadratic growth in the second unknown variable $z$, or the terminal value can be unbounded but the generator $g(t,y,z)$ is "diagonally dependent" on the second unknown variable $z$ (i.e., the $i$-th component $g^i$ of the generator $g$ only depends on the $i$-th row $z^i$ of the variable $z$ for each $i=1,\cdots,n$ ). Three new results are established on the local and global solutions when the terminal value is bounded and the generator $g$ is subject to some general assumptions. When the terminal value is unbounded but is of exponential moments of arbitrary order, an existence and uniqueness result is given under the assumptions that the generator $g(t,y,z)$ is Lipschitz continuous in the f

Generate-then-rank is a widely used mechanism for text generation, where a generator produces multiple text candidates and a ranker chooses the best one among the text candidates. However, existing methods usually train the generator and the ranker individually, neglecting the mutual feedback that could further enhance the generation quality. To tackle this limitation, we propose JGR, a novel joint training algorithm that integrates the generator and the ranker in a single framework. JGR optimizes the generator with a hybrid objective that combines data likelihood and ranker reward, and trains the ranker with a contrastive loss that compares the generator outputs. By iteratively updating the generator and the ranker, JGR can effectively harmonize their learning and enhance their quality jointly. We evaluate JGR on various text generation tasks and demonstrate that it surpasses existing methods on four public datasets across three common generation scenarios. Our code and models are publicly available at https://github.com/microsoft/ProphetNet/tree/master/JGR.

We introduce Optimal Eye Surgeon (OES), a framework for pruning and training deep image generator networks. Typically, untrained deep convolutional networks, which include image sampling operations, serve as effective image priors (Ulyanov et al., 2018). However, they tend to overfit to noise in image restoration tasks due to being overparameterized. OES addresses this by adaptively pruning networks at random initialization to a level of underparameterization. This process effectively captures low-frequency image components even without training, by just masking. When trained to fit noisy images, these pruned subnetworks, which we term Sparse-DIP, resist overfitting to noise. This benefit arises from underparameterization and the regularization effect of masking, constraining them in the manifold of image priors. We demonstrate that subnetworks pruned through OES surpass other leading pruning methods, such as the Lottery Ticket Hypothesis, which is known to be suboptimal for image recovery tasks (Wu et al., 2023). Our extensive experiments demonstrate the transferability of OES-masks and the characteristics of sparse-subnetworks for image generation. Code is available at https://gi

Pseudo-random number generators (PRNGs) play an important role to ensure the security and confidentiality of image cryptographic algorithms. Their primary function is to generate a sequence of numbers that possesses unpredictability and randomness, which is crucial for the algorithms to work effectively and provide the desired level of security. However, traditional PRNGs frequently encounter limitations like insufficient randomness, predictability, and vulnerability to cryptanalysis attacks. To overcome these limitations, we propose a novel method namely an elliptic curve genetic algorithm (ECGA) for the construction of an image-dependent pseudo-random number generator (IDPRNG) that merges elliptic curves (ECs) and a multi-objective genetic algorithm (MOGA). The ECGA consists of two primary stages. First, we generate an EC-based initial sequence of random numbers using pixels of a plain-image and parameters of an EC, that depart from traditional methods of population initialization. In our proposed approach, the image itself serves as the seed for the initial population in the genetic algorithm optimization, taking into account the image-dependent nature of cryptographic applicati

Let $\mathcal{A}$ be an essentially small abelian category. We prove that if $\mathcal{A}$ admits a generator $M$ with ${\rm End}_{\mathcal{A}}(M)$ right artinian, then $\mathcal{A}$ admits a projective generator. If $\mathcal{A}$ is further assumed to be Grothendieck, then this implies that $\mathcal{A}$ is equivalent to a module category. When $\mathcal{A}$ is Hom-finite over a field $k$, the existence of a generator is the same as the existence of a projective generator, and in case there is such a generator, $\mathcal{A}$ has to be equivalent to the category of finite dimensional right modules over a finite dimensional $k$-algebra. We also show that when $\mathcal{A}$ is a length category, then there is a one-to-one correspondence between exact abelian extension closed subcategories of $\mathcal{A}$ and collections of Hom-orthogonal Schur objects in $\mathcal{A}$.

Pancreatic ductal adenocarcinoma (PDAC) presents a critical global health challenge, and early detection is crucial for improving the 5-year survival rate. Recent medical imaging and computational algorithm advances offer potential solutions for early diagnosis. Deep learning, particularly in the form of convolutional neural networks (CNNs), has demonstrated success in medical image analysis tasks, including classification and segmentation. However, the limited availability of clinical data for training purposes continues to provide a significant obstacle. Data augmentation, generative adversarial networks (GANs), and cross-validation are potential techniques to address this limitation and improve model performance, but effective solutions are still rare for 3D PDAC, where contrast is especially poor owing to the high heterogeneity in both tumor and background tissues. In this study, we developed a new GAN-based model, named 3DGAUnet, for generating realistic 3D CT images of PDAC tumors and pancreatic tissue, which can generate the interslice connection data that the existing 2D CT image synthesis models lack. Our innovation is to develop a 3D U-Net architecture for the generator t

We introduce a new task named Story Ending Generation (SEG), whic-h aims at generating a coherent story ending from a sequence of story plot. Wepropose a framework consisting of a Generator and a Reward Manager for thistask. The Generator follows the pointer-generator network with coverage mech-anism to deal with out-of-vocabulary (OOV) and repetitive words. Moreover, amixed loss method is introduced to enable the Generator to produce story endingsof high semantic relevance with story plots. In the Reward Manager, the rewardis computed to fine-tune the Generator with policy-gradient reinforcement learn-ing (PGRL). We conduct experiments on the recently-introduced ROCStoriesCorpus. We evaluate our model in both automatic evaluation and human evalua-tion. Experimental results show that our model exceeds the sequence-to-sequencebaseline model by 15.75% and 13.57% in terms of CIDEr and consistency scorerespectively.

The rising use of machine learning in various fields requires robust methods to create synthetic tabular data. Data should preserve key characteristics while addressing data scarcity challenges. Current approaches based on Generative Adversarial Networks, such as the state-of-the-art CTGAN model, struggle with the complex structures inherent in tabular data. These data often contain both continuous and discrete features with non-Gaussian distributions. Therefore, we propose a novel Variational Autoencoder (VAE)-based model that addresses these limitations. Inspired by the TVAE model, our approach incorporates a Bayesian Gaussian Mixture model (BGM) within the VAE architecture. This avoids the limitations imposed by assuming a strictly Gaussian latent space, allowing for a more accurate representation of the underlying data distribution during data generation. Furthermore, our model offers enhanced flexibility by allowing the use of various differentiable distributions for individual features, making it possible to handle both continuous and discrete data types. We thoroughly validate our model on three real-world datasets with mixed data types, including two medically relevant ones

We compute the generator rank of a subhomgeneous C*-algebra in terms of the covering dimension of the pieces of its primitive ideal space corresponding to irreducible representations of a fixed dimension. We deduce that every Z-stable ASH-algebra has generator rank one, which means that a generic element in such an algebra is a generator. This leads to a strong solution of the generator problem for classifiable, simple, nuclear C*-algebras: a generic element in each such algebra is a generator. Examples of Villadsen show that this is not the case for all separable, simple, nuclear C*-algebras.

In this article, we show that, if $S\in \mathcal{B}(H)$ is irreducible and essential unitary, then $\{S,SS^*\}$ is a hyperrigid generator for the unital $C^*$-algebra $\mathcal{T}$ generated by $\{S,SS^*\}$. We prove that, if $T$ is an operator in $\mathcal{B}(H)$ that generates an unital $C^*$-algebra $\mathcal{A}$ then $\{T,T^*T,TT^*\}$ is a hyperrigid generator for $\mathcal{A}$. As a corollary it follows that, if $T\in \mathcal{B}(H)$ is normal then $\{T,TT^*\}$ is hyperrigid generator for the unital $C^*$-algebra generated by $T$ and if $T\in \mathcal{B}(H)$ is unitary then $\{T\}$ is hyperrigid generator for the $C^*$-algebra generated by $T$. We show that if $V\in \mathcal{B}(H)$ is an isometry (not unitary) that generates the $C^*$-algebra $\mathcal{A}$ then the minimal generating set $\{V\}$ is not hyperrigid for $\mathcal{A}$.

Conditional Generative Adversarial Networks are known to be difficult to train, especially when the conditions are continuous and high-dimensional. To partially alleviate this difficulty, we propose a simple generator regularization term on the GAN generator loss in the form of Lipschitz penalty. Thus, when the generator is fed with neighboring conditions in the continuous space, the regularization term will leverage the neighbor information and push the generator to generate samples that have similar conditional distributions for each neighboring condition. We analyze the effect of the proposed regularization term and demonstrate its robust performance on a range of synthetic and real-world tasks.

Quasi-cyclic (QC) codes form an important generalization of cyclic codes. It is well know that QC codes of length $s\ell$ with index $s$ over the finite field $\mathbb{F}$ are $\mathbb{F}[y]$-submodules of the ring $\frac{\mathbb{F}[x,y]}{< x^s-1,y^{\ell}-1 >}$. The aim of the present paper, is to study QC codes of length $s\ell$ with index $s$ over the finite field $\mathbb{F}$ and find generator polynomials and generator matrix for these codes. To achieve this aim, we apply a novel method to find generator polynomials for $\mathbb{F}[y]$-submodules of $\frac{\mathbb{F}[x,y]}{< x^s-1,y^{\ell}-1 >}$. These polynomials will be applied to obtain generator matrix for corresponding QC codes.

The Alternating Step(r,s) Generator, ASG(r,s), is a clock-controlled sequence generator which is recently proposed by A. Kanso. It consists of three registers of length l, m and n bits. The first register controls the clocking of the two others. The two other registers are clocked r times (or not clocked) (resp. s times or not clocked) depending on the clock-control bit in the first register. The special case r=s=1 is the original and well known Alternating Step Generator. Kanso claims there is no efficient attack against the ASG(r,s) since r and s are kept secret. In this paper, we present an Alternating Step Generator, ASG, model for the ASG(r,s) and also we present a new and efficient algebraic attack on ASG(r,s) using 3(m+n) bits of the output sequence to find the secret key with O((m^2+n^2)*2^{l+1}+ (2^{m-1})*m^3 + (2^{n-1})*n^3) computational complexity. We show that this system is no more secure than the original ASG, in contrast to the claim of the ASG(r,s)'s constructor.