A new approach to sensitivity analysis without the Sobol decomposition.
problem Traditional sensitivity indices like Sobol indices have limitations.
method Introducing sensitivity measures that generalize existing indices and define interaction effects.
result Sensitivity measures can create new indices and define interaction effects.
Enhances sensitivity analysis for correlated inputs.
problem Estimating sensitivity indices in models with correlated inputs.
method Proposes an extension of Sobol' estimator using a linear correlation model.
result Improves accuracy in variance-based sensitivity analysis.
Proposes a framework to incorporate global sensitivity into local surrogate models.
problem Narrowing focus to local scale in surrogate modeling leads to re-learning global trends.
method Integrates global sensitivity analysis into local surrogate models through input warping.
result Local models become equally sensitive to all input directions, focusing on local dynamics.
Fairness is becoming a rising concern w.r.t. machine learning model performance. Especially for sensitive fields such as criminal justice and loan decision, eliminating the prediction discrimination towards a certain group of population (characterized by sensitive features like race and gender) is important for enhanci…
We develop a method for quantile-based sensitivity analysis in models with discontinuities.
problem Uncertainty in interpreting discontinuous models using traditional derivatives.
method Quantile-based derivatives for discontinuous models with discrete inputs.
result Derivatives of quantile-based outputs are well-defined and provide meaningful insights.
Package provides sensitivity analysis for neural networks.
problem Understanding how input variables affect neural network predictions.
method Partial derivatives method to calculate sensitivities.
result Allows evaluation of variable importance and input-output relationships.
FANNet analyzes noise tolerance and training bias in neural networks.
problem Low noise tolerance and input sensitivity in neural networks lead to failures on unseen inputs.
method Formal analysis using model checking under different noise ranges.
result Noise tolerance of ±11% for the trained network, sensitive input nodes identified, and biasness confirmed. This paper explores tradeoffs between invariance and sensitivity in adversarial examples.
problem Understanding the limitations of existing adversarial defenses.
method Study of invariance-based adversarial examples and their impact on model accuracy.
result Adversarial defenses against sensitivity-based attacks can harm invariance-based attacks, necessitating new approaches.
The Renormalisation Group (RG) provides a framework in which it is possible to assess whether a deep-learning network is sensitive to small changes in the input data and hence prone to error, or susceptible to adversarial attack. Distinct classification outputs are associated with different RG fixed points and sensitiv…
Generalizes Hoeffding's decomposition for dependent inputs under mild conditions.
problem Performing global sensitivity analysis on black-box models with dependent inputs.
method Proposes a novel framework based on probability theory, functional analysis, and combinatorics to handle dependencies.
result Any square-integrable, real-valued function of random elements with mild dependence assumptions can be uniquely additively decomposed.
We present a novel algorithm (Principal Sensitivity Analysis; PSA) to analyze the knowledge of the classifier obtained from supervised machine learning techniques. In particular, we define principal sensitivity map (PSM) as the direction on the input space to which the trained classifier is most sensitive, and use anal…
Differentially private graph learning via bounded sensitivity PPR.
problem Protecting user data in graph learning algorithms.
method Proposes a sensitivity-bounded personalized PageRank (PPR) algorithm.
result Achieves similar accuracy to non-private algorithms with large degrees.
The paper develops methods to analyze sensitivity in stochastic models using surrogate models.
problem Quantifying the impact of input variability on stochastic simulators with randomness.
method The authors propose using generalized lambda models to emulate response distributions of stochastic simulators and estimate sensitivity indices.
result The proposed method can estimate sensitivity indices even with strong heteroskedasticity and small signal-to-noise ratio.
This research investigates reliable local explanations for machine listening models.
problem Generating reliable local explanations for machine listening models.
method Investigates the sensitivity of SoundLIME explanations to input perturbations and proposes a novel method for identifying suitable content types.
result SoundLIME explanations are sensitive to the content in occluded input regions, and the average magnitude of input mel-spectrogram bins is the most suitable content type for temporal explanations.
A new method reduces both input and output dimensions for better goal-oriented analysis.
problem Simultaneous reduction of input and output dimensions for more accurate analysis.
method Coupled input-output dimension reduction, optimizing gradient-based bounds.
result Determine most informative sensors and influential parameters efficiently.
Reverse sensitivity analysis for risk models under various stresses.
problem Understanding model changes under output stress.
method Deriving the closest stressed distribution and model parameters.
result Numerically efficient method for calculating stressed model.
Transformers are less sensitive to input perturbations compared to other models.
problem Understanding the inductive biases of transformers and distinguishing them from other architectures.
method Identified token-wise sensitivity as a metric to explain transformers' inductive biases across different data modalities.
result Transformers have lower sensitivity than MLPs, CNNs, ConvMixers, and LSTMs, across vision and language tasks.
We derive a novel sensitivity analysis of input variables for predictive epistemic and aleatoric uncertainty. We use Bayesian neural networks with latent variables as a model class and illustrate the usefulness of our sensitivity analysis on real-world datasets. Our method increases the interpretability of complex blac…
Saliency methods aim to explain the predictions of deep neural networks. These methods lack reliability when the explanation is sensitive to factors that do not contribute to the model prediction. We use a simple and common pre-processing step ---adding a constant shift to the input data--- to show that a transformatio…
Adjoint SA speeds up bioprocess parameter learning.
problem Challenges in digital twin development for biomanufacturing.
method Adjoint sensitivity analysis on multi-scale enzymatic reaction networks.
result Resilient sensitivities reveal bioprocess regulatory mechanisms.
New method calculates sensitivity of system failure probability.
problem Difficulty in computing sensitivity of failure probability.
method Monte Carlo strategy using response gradient and kernel smoothing.
result Single Monte Carlo run for sensitivity estimates.
Embeddings leak sensitive information about input data, which can be recovered or inferred.
problem Information leakage in embedding models.
method Developed three classes of attacks to study information leakage.
result Embeddings leak sensitive information about input data, which can be partially recovered or inferred.
Given a loss function F:X→R+ that can be written as the sum of losses over a large set of inputs a1,…,an, it is often desirable to approximate F by subsampling the input points. Strong theoretical guarantees require taking into account the importance of each point, measured by how …
Efficiently identifies key input variables for expensive functions using active learning.
problem Efficiently identify key input variables for expensive, black-box functions.
method Proposes novel active learning acquisition functions targeting derivative-based global sensitivity measures (DGSMs) under Gaussian process surrogate models.
result Active learning substantially enhances sample efficiency of DGSM estimation, especially with limited evaluation budgets.
Proposes ICE-based metric for better understanding interactions in black-box models.
problem Misleading global sensitivity metrics in black-box models due to interaction effects.
method Individual Conditional Expectation (ICE) curves to compute feature importance and interactions.
result ICE-based metric provides richer insights into feature importance and interactions.
Neural networks are vulnerable to small adversarial perturbations. Existing literature largely focused on understanding and mitigating the vulnerability of learned models. In this paper, we demonstrate an intriguing phenomenon about the most popular robust training method in the literature, adversarial training: Advers…
Noise injection improves inference privacy in DNN models.
problem Malicious servers can infer sensitive attributes from input data.
method Adaptive Noise Injection (ANI) using a lightweight DNN on the client.
result Significant improvement in privacy (up to 48.5% degradation in sensitive-task accuracy with <1% degradation in primary accuracy).
Global Sensitivity Analysis improves feature importance ranking in Random Forests.
problem Improving feature importance ranking in Random Forests.
method Applying Global Sensitivity Analysis to Random Forests for feature ranking.
result Our method provides a novel way to rank features based on their importance.
Convolutional neural networks (CNNs) have recently emerged as a popular building block for natural language processing (NLP). Despite their success, most existing CNN models employed in NLP share the same learned (and static) set of filters for all input sentences. In this paper, we consider an approach of using a smal…
This research reverses feature visualization in neural networks to optimize for specific feature objectives.
problem The invertibility of feature visualization in neural networks is not well understood.
method The approach involves optimizing for the feature objective that generates the input used in feature visualization, using the gradient of a specific objective function.
result A closed-form solution is found to minimize the gradient, providing an alternative view on network sensitivity.
Study shows resampling can drastically alter PCA results.
problem Stability and sensitivity of PCA under data resampling.
method Analyzed resampling sensitivity of high-dimensional PCA.
result PCA's principal components become asymptotically orthogonal when resampling is significant.
Deep Convolutional Networks (DCNs) have been shown to be sensitive to Universal Adversarial Perturbations (UAPs): input-agnostic perturbations that fool a model on large portions of a dataset. These UAPs exhibit interesting visual patterns, but this phenomena is, as yet, poorly understood. Our work shows that visually …
Recently, it has been shown that many functions on sets can be represented by sum decompositions. These decompositons easily lend themselves to neural approximations, extending the applicability of neural nets to set-valued inputs---Deep Set learning. This work investigates a core component of Deep Set architecture: ag…
Study adapts AI research methods to analyze image augmentation impacts on neural network operations.
problem Understanding how image augmentation affects neural network performance and sensitivity.
method Adapted treatment-control paradigm, uses variance decomposition, Sobol indices, and Shapley values for sensitivity analysis.
result Visualizes and quantifies sensitivity to different image augmentation parameters.
SSRCA simplifies ABM sensitivity analysis using machine learning.
problem Hardness of performing sensitivity analysis for complex ABMs.
method Machine learning pipeline (Simulate, Summarize, Reduce, Cluster, Analyze) for ABMs.
result SSRCA identifies sensitive parameters and common output patterns for ABMs.
Linking output sensitivity to deep learning generalization.
problem Understanding and comparing the generalization properties of deep neural networks.
method Linking the loss function to output sensitivity and analyzing its relation to bias-variance decomposition.
result Output sensitivity is a strong metric for comparing generalization performance of deep networks.
We investigate the generalizability of deep learning based on the sensitivity to input perturbation. We hypothesize that the high sensitivity to the perturbation of data degrades the performance on it. To reduce the sensitivity to perturbation, we propose a simple and effective regularization method, referred to as spe…
Emulator speeds up landslide run-out modeling sensitivity analysis.
problem Computational challenges in assessing landslide run-out model sensitivity.
method Gaussian process emulation integrated into r.avaflow.
result Strong interactions detected between friction coefficients and release volume.
Recently, machine learning algorithms have successfully entered large-scale real-world industrial applications (e.g. search engines and email spam filters). Here, the CPU cost during test time must be budgeted and accounted for. In this paper, we address the challenge of balancing the test-time cost and the classifier …
Fairness is a critical trait in decision making. As machine-learning models are increasingly being used in sensitive application domains (e.g. education and employment) for decision making, it is crucial that the decisions computed by such models are free of unintended bias. But how can we automatically validate the fa…
Sobol method applied to probabilistic networks for sensitivity analysis.
problem Measuring influence of probabilistic network nodes on a quantity of interest.
method Transforms global sensitivity analysis into marginalization inference exploiting network structure.
result Efficient computation of sensitivity indices for complex networks.
Two ANOVA-based algorithms boost random Fourier feature models for function approximation.
problem Approximating high-dimensional functions with low-order interactions.
method Utilizes ANOVA decomposition to learn low-order functions and index sets of important variables.
result Significantly reduces approximation error compared to existing methods.
RISE learns decisions with sensitive variables, improving worst-case outcomes.
problem Uncertainty and bias in decisions due to delayed sensitive variable data.
method Incorporates sensitive variables offline but not at deployment, using quantile or infimum optimization.
result Improves worst-case outcomes for individuals affected by unavailable sensitive variables.
Paper uses PCE to quantify ML model and input uncertainties.
problem Accurately quantify and propagate combined uncertainties in ML predictions.
method Polynomial Chaos Expansion (PCE) for joint input and model uncertainty.
result Efficient and accurate calculation of output variability and sensitivity.
Global sensitivity analysis with variance-based measures suffers from several theoretical and practical limitations, since they focus only on the variance of the output and handle multivariate variables in a limited way. In this paper, we introduce a new class of sensitivity indices based on dependence measures which o…
In practice it is often found that large over-parameterized neural networks generalize better than their smaller counterparts, an observation that appears to conflict with classical notions of function complexity, which typically favor smaller models. In this work, we investigate this tension between complexity and gen…
Work proposes a new framework to improve uncertainty estimation in deep Bayesian models.
problem Traditional training procedures underestimate uncertainty in NLMs, leading to unreliable predictions.
method Introduces a novel training framework that captures useful predictive uncertainties for out-of-distribution inputs.
result Demonstrates that traditional methods for NLMs significantly underestimate uncertainty and propose a new framework to address this issue.
We study the problem of data release with privacy, where data is made available with privacy guarantees while keeping the usability of the data as high as possible --- this is important in health-care and other domains with sensitive data. In particular, we propose a method of masking the private data with privacy guar…