This guide explains statistical distances for evaluating generative models.
problem Evaluating the quality of samples from generative models.
method Four statistical distances: SW, C2ST, MMD, FID.
result Different distances can yield varying results on similar data.
We explore a new way to evaluate generative models using insights from evaluation of competitive games between human players. We show experimentally that tournaments between generators and discriminators provide an effective way to evaluate generative models. We introduce two methods for summarizing tournament outcomes…
RENAL test evaluates generative models for time series data.
problem Evaluating generative models for time series data is challenging.
method RENAL test uses recurrent neural networks to transform time series into conditionally independent data pairs for goodness-of-fit testing.
result RENAL test outperforms existing methods in evaluating generative models for time series data.
InferPy is a Python package for probabilistic modeling with deep neural networks. It defines a user-friendly API that trades-off model complexity with ease of use, unlike other libraries whose focus is on dealing with very general probabilistic models at the cost of having a more complex API. In particular, this packag…
In this paper we propose a novel method for detecting adversarial examples by training a binary classifier with both origin data and saliency data. In the case of image classification model, saliency simply explain how the model make decisions by identifying significant pixels for prediction. A model shows wrong classi…
This work introduces a novel method to evaluate generative model novelty.
problem Evaluating the novelty of generative models compared to a reference model.
method Spectral approach to differential clustering and Kernel-based Entropic Novelty (KEN) score.
result The KEN score effectively detects novel modes and compares generative models.
Paper addresses inconsistency between offline and online LTR performance.
problem Inconsistency between offline and online LTR performance in E-commerce.
method Proposes an evaluator-generator framework to maximize evaluator score using reinforcement learning.
result Significant improvement in Conversion Rate (CR) over existing models.
The ability to represent and compare machine learning models is crucial in order to quantify subtle model changes, evaluate generative models, and gather insights on neural network architectures. Existing techniques for comparing data distributions focus on global data properties such as mean and covariance; in that se…
Counterfactuals help compare different scenarios in CI and XAI.
problem Comparing how counterfactuals are used in causal inference and explainable AI.
method Formal definition and comparison of counterfactuals in CI and XAI.
result Identified opportunities for cross-fertilization between CI and XAI.
The true distribution parameterizations of commonly used image datasets are inaccessible. Rather than designing metrics for feature spaces with unknown characteristics, we propose to measure GAN performance by evaluating on explicitly parameterized, synthetic data distributions. As a case study, we examine the performa…
Deep reinforcement learning algorithms have shown an impressive ability to learn complex control policies in high-dimensional tasks. However, despite the ever-increasing performance on popular benchmarks, policies learned by deep reinforcement learning algorithms can struggle to generalize when evaluated in remarkably …
Evaluating generative adversarial networks (GANs) is inherently challenging. In this paper, we revisit several representative sample-based evaluation metrics for GANs, and address the problem of how to evaluate the evaluation metrics. We start with a few necessary conditions for metrics to produce meaningful scores, su…
New benchmarks measure image generation models' ability to generalize beyond training data.
problem Trivially memorizing training data yields better scores than state-of-the-art models on current benchmarks.
method Developed neural network divergences (NNDs) as evaluation metrics requiring large samples.
result Implemented and validated a black-box metric that measures diversity, sample quality, and generalization.
The paper introduces new metrics for evaluating generative models of behavior.
problem Lack of quantitative evaluation criteria for unsupervised behavior discovery.
method Proposed and investigated several metrics for generative models of behavior.
result The proposed metrics correspond with biologists' intuitions and allow for model evaluation and bias understanding.
We study proper losses for discrete generative models without knowing the target distribution.
problem Evaluating generative models in the discrete setting without direct access to the target distribution.
method Define and construct black-box proper losses using statistical estimation theory.
result Black-box proper losses must be of polynomial form and involve more samples than the polynomial degree.
INT benchmark tests theorem proving agents' ability to generalize to unseen theorems.
problem Evaluating theorem proving agents' ability to generalize to unseen theorems.
method INT benchmark based on a theorem generation and proof procedure with adjustable knobs for measuring 6 types of generalization.
result MCTS can help agents prove new theorems.
Optimizes crowdsourced preference-based subjective evaluation with online learning.
problem Large-scale evaluation of generative media using crowdsourcing due to combinatorial explosion.
method Automatic optimization of pair combination selections and evaluation volumes with online learning.
result Optimizes evaluation by reducing pair combinations and allocating optimal evaluation volumes.
Recent advances in generative modeling have led to an increased interest in the study of statistical divergences as means of model comparison. Commonly used evaluation methods, such as the Frechet Inception Distance (FID), correlate well with the perceived quality of samples and are sensitive to mode dropping. However,…
Enhanced metrics improve generative model evaluation reliability.
problem Lack of reliable quality metrics for generative models.
method Introduce Clipped Density and Clipped Coverage metrics.
result Metrics prevent out-of-distribution samples from biasing quality scores.
Generative adversarial networks (GANs) are a powerful approach to unsupervised learning. They have achieved state-of-the-art performance in the image domain. However, GANs are limited in two ways. They often learn distributions with low support---a phenomenon known as mode collapse---and they do not guarantee the exist…
Generative Policy-based Models aim to enable a coalition of systems, be they devices or services to adapt according to contextual changes such as environmental factors, user preferences and different tasks whilst adhering to various constraints and regulations as directed by a managing party or the collective vision of…
AI techniques explain synthetic tabular data weaknesses.
problem Challenges in evaluating synthetic tabular data quality.
method Apply explainable AI to a binary detection classifier.
result Reveals inconsistencies, unrealistic dependencies, or missing patterns in synthetic data.
This paper introduces a new method to train normalizing flows using precision-recall divergences.
problem Training generative models with mode dropping and low-quality samples.
method Introduces PR-divergences and proposes a novel generative model to minimize precision-recall trade-offs.
result Normalizing flows can be trained to achieve specific precision-recall trade-offs using PR-divergences.
ProEval efficiently estimates AI performance and discovers failures using pre-trained Gaussian Processes.
problem Resource-intensive evaluation of generative AI models.
method ProEval uses pre-trained Gaussian Processes and Bayesian quadrature to estimate performance and discover failures.
result ProEval requires significantly fewer samples to achieve accurate performance estimates and reveals more diverse failure cases.
We define general linguistic intelligence as the ability to reuse previously acquired knowledge about a language's lexicon, syntax, semantics, and pragmatic conventions to adapt to new tasks quickly. Using this definition, we analyze state-of-the-art natural language understanding models and conduct an extensive empiri…
In this paper, we propose Rogue-Gym, a simple and classic style roguelike game built for evaluating generalization in reinforcement learning (RL). Combined with the recent progress of deep neural networks, RL has successfully trained human-level agents without human knowledge in many games such as those for Atari 2600.…
This paper solves robust utility maximization with unknown claim dependencies.
problem Investor optimizes utility in the presence of an intractable contingent claim.
method Quantile optimization approach, transforming dynamic problem into static concave optimization.
result Optimal payoffs depend on ambiguity attitude, market conditions, and claim characteristics.
This research evaluates generalization measures in deep learning.
problem Understanding why deep learning models generalize well despite small training error.
method Empirical evaluation of generalization bounds and measures.
result Generalization measures should be evaluated using distributional robustness.
Proposes Topology Distance for evaluating GANs.
problem Challenges in evaluating GANs' goodness.
method Builds Vietoris-Rips complex on image features and defines TD based on latent manifold comparisons.
result Demonstrates TD's superiority over existing metrics.
Efficiently evaluate generative models at the prompt level using tensor factorization.
problem Fine-grained evaluations of generative models are costly and often misaligned with human judgment.
method Tensor factorization model that merges cheap autorater data with a small set of human gold-standard labels.
result The method provides accurate and tight confidence intervals for model performance.
KBB algorithm reduces sample complexity for policy evaluation in general state spaces.
problem Policy evaluation in large state spaces with high sample complexity.
method Alternates between fitting Bellman residual and estimating value function via adaptive feature set growth.
result Super-linear convergence rates demonstrated, with reductions in sample complexity.
CDRE estimates density ratios in streaming data without historical samples.
problem Online learning with shifting data distributions.
method Iterative estimation of density ratios between initial and current distributions.
result CDRE outperforms standard DRE in estimating divergences between distributions.
Stochastic-sampling-based Generative Neural Networks, such as Restricted Boltzmann Machines and Generative Adversarial Networks, are now used for applications such as denoising, image occlusion removal, pattern completion, and motion synthesis. In scenarios which involve performing such inference tasks with these model…
A new metric evaluates generative models by comparing real and generated samples.
problem Evaluating the quality of generative models.
method Relative Density Ratio (RDR) function, optimization on variational form of φ-divergence.
result The RDR function provides a clear, interpretable, and numerically stable evaluation metric.
Generative model for TPPs using signatures and distributional discrepancies.
problem Limitations of signature methods for TPPs and lack of global sequence-level loss in neural models.
method Introduce interarrival embedding to lift jump paths to continuous paths of bounded variation, enabling signature methods for discrete event sequences. Develop sigTPP, a signature-based generative model trained on path-level loss.
result sigTPP achieves the best average rank across multiple metrics and outperforms or is within a standard error of the strongest baseline in 64% of dataset-metric pairs.
New metric evaluates generative models across domains, diagnosing fidelity, diversity, and generalization.
problem Evaluating generative models in diverse domains with limited metrics.
method Introduces a 3D evaluation metric (α-Precision, β-Recall, Authenticity) for domain-agnostic diagnostics. result Unified metric characterizes fidelity, diversity, and generalization, diagnosing model performance.
Counterfactual policy evaluation improves autonomous driving policies' generalization.
problem Learnt policies often fail to generalize and handle novel situations.
method Introduces counterfactual policy evaluation using counterfactual worlds.
result Significantly decreases collision-rate while maintaining high success-rate.
The paper introduces a novel method for training neural network Stein critics with staged L2-regularization.
problem Learning to differentiate model distributions from observed data in high-dimensional settings.
method Developed a novel staging procedure for L2 regularization over training time, leveraging the advantages of highly-regularized training at early times. result Theoretical guarantees and empirical validation show that the method improves the approximation of the training dynamic by the kernel optimization, leading to faster convergence and better performance.
Generative models are known to be difficult to assess. Recent works, especially on generative adversarial networks (GANs), produce good visual samples of varied categories of images. However, the validation of their quality is still difficult to define and there is no existing agreement on the best evaluation process. …
Study evaluates two-sample tests for validating generative models in high dimensions.
problem Validating the performance and efficiency of non-parametric two-sample tests for high-dimensional generative models.
method Proposes and evaluates the sliced Wasserstein distance, mean of Kolmogorov-Smirnov statistics, and novel sliced Kolmogorov-Smirnov statistic.
result One-dimensional-based tests provide comparable sensitivity to other multivariate metrics but with lower computational cost.
This dissertation tackles challenges in reliable machine learning measurement.
problem Challenges in reproducibility, scalability, and uncertainty quantification in machine learning.
method Develops criteria for meaningful metrics and methodologies for scalable, reliable measurement.
result Provides methods for evaluating generative-AI systems and quantifying memorization.
TRADES generates realistic market simulations for financial modeling.
problem Generating realistic and responsive market simulations for financial tasks.
method TRADES uses a transformer-based denoising diffusion probabilistic engine to generate time series order flows conditioned on market state.
result TRADES improves market simulation metrics by 3.27-3.48 over state-of-the-art (SoTA) methods.
Study flaws in generative model evaluation metrics, especially for diffusion models.
problem Flaws in existing metrics for evaluating generative models, particularly for diffusion models.
method Systematic study of generative models, human perception experiments, and analysis of feature extractors.
result State-of-the-art perceptual realism of diffusion models is not reflected in commonly reported metrics.