A good metric, which promises a reliable comparison between solutions, is essential for any well-defined task. Unlike most vision tasks that have per-sample ground-truth, image synthesis tasks target generating unseen data and hence are usually evaluated through a distributional distance between one set of real samples and another set of generated samples. This study presents an empirical investigation into the evaluation of synthesis performance, with generative adversarial networks (GANs) as a representative of generative models. In particular, we make in-depth analyses of various factors, including how to represent a data point in the representation space, how to calculate a fair distance using selected samples, and how many instances to use from each set. Extensive experiments conducted on multiple datasets and settings reveal several important findings. Firstly, a group of models that include both CNN-based and ViT-based architectures serve as reliable and robust feature extractors for measurement evaluation. Secondly, Centered Kernel Alignment (CKA) provides a better comparison across various extractors and hierarchical layers in one model. Finally, CKA is more sample-efficient and enjoys better agreement with human judgment in characterizing the similarity between two internal data correlations. These findings contribute to the development of a new measurement system, which enables a consistent and reliable re-evaluation of current state-of-the-art generative models.
相關內容
The growth and progression of brain tumors is governed by patient-specific dynamics. Even when the tumor appears well-delineated in medical imaging scans, tumor cells typically already have infiltrated the surrounding brain tissue beyond the visible lesion boundaries. Quantifying and understanding these growth dynamics promises to reveal this otherwise hidden spread and is key to individualized therapies. Current treatment plans for brain tumors, such as radiotherapy, typically involve delineating a standard uniform margin around the visible tumor on imaging scans to target this invisible tumor growth. This "one size fits all" approach is derived from population studies and often fails to account for the nuances of individual patient conditions. Here, we present the framework GliODIL which infers the full spatial distribution of tumor cell concentration from available imaging data based on PDE-constrained optimization. The framework builds on the newly introduced method of Optimizing the Discrete Loss (ODIL), data are assimilated in the solution of the Partial Differential Equations (PDEs) by optimizing a cost function that combines the discrete form of the equations and data as penalty terms. By utilizing consistent and stable discrete approximations of the PDEs, employing a multigrid method, and leveraging automatic differentiation, we achieve computation times suitable for clinical application such as radiotherapy planning. Our method performs parameter estimation in a manner that is consistent with the PDEs. Through a harmonious blend of physics-based constraints and data-driven approaches, GliODIL improves the accuracy of estimating tumor cell distribution and, clinically highly relevant, also predicting tumor recurrences, outperforming all other studied benchmarks.
Quantum multiprover interactive proof systems with entanglement MIP* are much more powerful than their classical counterpart MIP (Babai et al. '91, Ji et al. '20): while MIP = NEXP, the quantum class MIP* is equal to RE, a class including the halting problem. This is because the provers in MIP* can share unbounded quantum entanglement. However, recent works of Qin and Yao '21 and '23 have shown that this advantage is significantly reduced if the provers' shared state contains noise. This paper attempts to exactly characterize the effect of noise on the computational power of quantum multiprover interactive proof systems. We investigate the quantum two-prover one-round interactive system MIP*[poly, O(1)], where the verifier sends polynomially many bits to the provers and the provers send back constantly many bits. We show noise completely destroys the computational advantage given by shared entanglement in this model. Specifically, we show that if the provers are allowed to share arbitrarily many EPR states, where each EPR state is affected by an arbitrarily small constant amount of noise, the resulting complexity class is contained in NEXP = MIP. This improves significantly on the previous best-known bound of NEEEXP (nondeterministic triply exponential time) by Qin and Yao '21. We also show that this collapse in power is due to the noise, rather than the O(1) answer size, by showing that allowing for noiseless EPR states gives the class the full power of RE = MIP*[poly, poly]. Along the way, we develop two technical tools of independent interest. First, we give a new, deterministic tester for the positivity of an exponentially large matrix, provided it has a low-degree Fourier decomposition in terms of Pauli matrices. Secondly, we develop a new invariance principle for smooth matrix functions having bounded third-order Fr\'echet derivatives or which are Lipschitz continous.
Optimization pipelines targeting polyhedral programs try to maximize the compute throughput. Traditional approaches favor reuse and temporal locality; while the communicated volume can be low, failure to optimize spatial locality may cause a low I/O performance. Memory allocation schemes using data partitioning such as data tiling can improve the spatial locality, but they are domain-specific and rarely applied by compilers when an existing allocation is supplied. In this paper, we propose to derive a partitioned memory allocation for tiled polyhedral programs using their data flow information. We extend the existing MARS partitioning to handle affine dependences, and determine which dependences can lead to a regular, simple control flow for communications. While this paper consists in a theoretical study, previous work on data partitioning in inter-node scenarios has shown performance improvements due to better bandwidth utilization.
The heteroscedastic probabilistic principal component analysis (PCA) technique, a variant of the classic PCA that considers data heterogeneity, is receiving more and more attention in the data science and signal processing communities. In this paper, to estimate the underlying low-dimensional linear subspace (simply called \emph{ground truth}) from available heterogeneous data samples, we consider the associated non-convex maximum-likelihood estimation problem, which involves maximizing a sum of heterogeneous quadratic forms over an orthogonality constraint (HQPOC). We propose a first-order method -- generalized power method (GPM) -- to tackle the problem and establish its \emph{estimation performance} guarantee. Specifically, we show that, given a suitable initialization, the distances between the iterates generated by GPM and the ground truth decrease at least geometrically to some threshold associated with the residual part of certain "population-residual decomposition". In establishing the estimation performance result, we prove a novel local error bound property of another closely related optimization problem, namely quadratic optimization with orthogonality constraint (QPOC), which is new and can be of independent interest. Numerical experiments are conducted to demonstrate the superior performance of GPM in both Gaussian noise and sub-Gaussian noise settings.
This paper gives a broad account of the various sequent-based proof formalisms in the proof-theoretic literature. We consider formalisms for various modal and tense logics, intuitionistic logic, conditional logics, and bunched logics. After providing an overview of the logics and proof formalisms under consideration, we show how these sequent-based formalisms can be placed in a hierarchy in terms of the underlying data structure of the sequents. We then discuss how this hierarchy can be traversed using translations. Translating proofs up this hierarchy is found to be relatively easy while translating proofs down the hierarchy is substantially more difficult. Finally, we inspect the prevalent distinction in structural proof theory between 'internal calculi' and 'external calculi'. It is observed that these classes resist a rigorous separation, and we critically assess the properties that (calculi from) these classes are purported to possess.
Quasiperiodic systems are important space-filling ordered structures, without decay and translational invariance. How to solve quasiperiodic systems accurately and efficiently is of great challenge. A useful approach, the projection method (PM) [J. Comput. Phys., 256: 428, 2014], has been proposed to compute quasiperiodic systems. Various studies have demonstrated that the PM is an accurate and efficient method to solve quasiperiodic systems. However, there is a lack of theoretical analysis of PM. In this paper, we present a rigorous convergence analysis of the PM by establishing a mathematical framework of quasiperiodic functions and their high-dimensional periodic functions. We also give a theoretical analysis of quasiperiodic spectral method (QSM) based on this framework. Results demonstrate that PM and QSM both have exponential decay, and the QSM (PM) is a generalization of the periodic Fourier spectral (pseudo-spectral) method. Then we analyze the computational complexity of PM and QSM in calculating quasiperiodic systems. The PM can use fast Fourier transform, while the QSM cannot. Moreover, we investigate the accuracy and efficiency of PM, QSM and periodic approximation method in solving the linear time-dependent quasiperiodic Schr\"{o}dinger equation.
The Knapsack Problem is a classic problem in combinatorial optimisation. Solving these problems may be computationally expensive. Recent years have seen a growing interest in the use of deep learning methods to approximate the solutions to such problems. A core problem is how to enforce or encourage constraint satisfaction in predicted solutions. A promising approach for predicting solutions to constrained optimisation problems is the Lagrangian Dual Framework which builds on the method of Lagrangian Relaxation. In this paper we develop neural network models to approximate Knapsack Problem solutions using the Lagrangian Dual Framework while improving constraint satisfaction. We explore the problems of output interpretation and model selection within this context. Experimental results show strong constraint satisfaction with a minor reduction of optimality as compared to a baseline neural network which does not explicitly model the constraints.
With the growth of large language models, now incorporating billions of parameters, the hardware prerequisites for their training and deployment have seen a corresponding increase. Although existing tools facilitate model parallelization and distributed training, deeper model interactions, crucial for interpretability and responsible AI techniques, still demand thorough knowledge of distributed computing. This often hinders contributions from researchers with machine learning expertise but limited distributed computing background. Addressing this challenge, we present FlexModel, a software package providing a streamlined interface for engaging with models distributed across multi-GPU and multi-node configurations. The library is compatible with existing model distribution libraries and encapsulates PyTorch models. It exposes user-registerable HookFunctions to facilitate straightforward interaction with distributed model internals, bridging the gap between distributed and single-device model paradigms. Primarily, FlexModel enhances accessibility by democratizing model interactions and promotes more inclusive research in the domain of large-scale neural networks. The package is found at //github.com/VectorInstitute/flex_model.
With the advance of the powerful heterogeneous, parallel and distributed computing systems and ever increasing immense amount of data, machine learning has become an indispensable part of cutting-edge technology, scientific research and consumer products. In this study, we present a review of modern machine and deep learning. We provide a high-level overview for the latest advanced machine learning algorithms, applications, and frameworks. Our discussion encompasses parallel distributed learning, deep learning as well as federated learning. As a result, our work serves as an introductory text to the vast field of modern machine learning.
Graph Neural Networks (GNNs) have been studied from the lens of expressive power and generalization. However, their optimization properties are less well understood. We take the first step towards analyzing GNN training by studying the gradient dynamics of GNNs. First, we analyze linearized GNNs and prove that despite the non-convexity of training, convergence to a global minimum at a linear rate is guaranteed under mild assumptions that we validate on real-world graphs. Second, we study what may affect the GNNs' training speed. Our results show that the training of GNNs is implicitly accelerated by skip connections, more depth, and/or a good label distribution. Empirical results confirm that our theoretical results for linearized GNNs align with the training behavior of nonlinear GNNs. Our results provide the first theoretical support for the success of GNNs with skip connections in terms of optimization, and suggest that deep GNNs with skip connections would be promising in practice.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191