Proposes a falsification framework to test algorithmic discriminant validity.
problem Unintended model behavior in predictive algorithms.
method Falsification framework based on statistical tests comparing prediction losses across outcomes.
result Establishes discriminant validity for some outcomes but not others.
Proposes a test to ensure predictive algorithms predict intended outcomes better than unintended ones.
problem Unintended model behavior leading to prediction of unintended outcomes.
method Falsification framework using nonparametric hypothesis testing to compare prediction losses across outcomes.
result Establishes discriminant validity with respect to gender but not race in an admissions setting.
VerifAI toolkit improves neural network-based aircraft taxiing system safety.
problem Improving safety of autonomous aircraft taxiing systems using neural networks.
method Unified approach to formal analysis and retraining of AI systems, including falsification, debugging, and retraining.
result Improved neural network performance and reduced failure cases in aircraft taxiing system.
Algorithm detects unmeasured confounding in observational data.
problem Estimating treatment effects in observational studies with untestable conditions.
method Two-stage procedure that detects dependencies between causal mechanisms.
result Algorithm efficiently detects confounding on simulated and semi-synthetic data.
Audit financial machine learning workflows to detect spurious predictability.
problem Spurious predictability in financial machine learning models.
method Falsification audit testing predictive workflows against synthetic environments.
result Many apparent financial predictions are artifacts, not genuine.
Study finds no statistically significant trading edge in MNQ futures signals from OHLCV data.
problem Testing intraday momentum signals from OHLCV data in MNQ futures under realistic execution constraints.
method 947 trading days of five-minute data, 14 signal families evaluated, strict institutional criteria applied.
result No signal satisfies all criteria simultaneously, gross edge insufficient to overcome costs.
This work examines fundamental limits in model falsification without assuming specific distributions.
problem Establishing lower bounds on model class risk in distribution-free settings.
method Model-agnostic fundamental hardness result for constructing lower bounds on test error.
result No positive lower bound on model class risk is possible in certain settings.
Tests validity of DML estimators without assumptions.
problem Validating DML estimators without making assumptions.
method Develops tests to falsify assumptions for DML estimators.
result Falsifies assumptions for DML estimators with non-trivial power.
This paper considers the problem of detection in distributed networks in the presence of data falsification (Byzantine) attacks. Detection approaches considered in the paper are based on fully distributed consensus algorithms, where all of the nodes exchange information only with their neighbors in the absence of a fus…
A new method predicts causal relationships without joint data.
problem Falsifying causal discovery algorithms without ground truth.
method Leave-One-Variable-Out (LOVO) prediction for causal graphs.
result LOVO method correlates prediction error with causal algorithm accuracy.
Efficiently generates models resistant to falsification.
problem Creating models that cannot be disproven by tests.
method Exploits connections between high-dimensional multicalibration and expected variational inequality problems to develop an efficient algorithm.
result First to efficiently produce online outcome indistinguishable generative models resistant to infinite classes of tests.
Survey of algorithms for testing AI-driven CPS safety.
problem Testing AI-driven CPS for safety in complex environments.
method Survey of applied algorithms for safety validation.
result Survey of existing tools and techniques for safety validation.
A new metric assesses causal graphs using node permutations to detect inconsistencies.
problem Quantifying the goodness of causal graphs and distinguishing them from random graphs.
method Constructing a baseline through node permutations and comparing inconsistencies.
result The proposed metric can distinguish between true and wrong causal graphs.
The standard taxonomy of predictive uncertainty is inconsistent with standard measures.
problem Uncertainty taxonomy and measure inconsistency
method Proof of inconsistency
result Uncertainty is not reducible to data collection
New method falsifies causal discovery results without ground truth.
problem Evaluation of causal discovery algorithms without ground truth data.
method Detects incompatibilities between causal graphs learned on different subsets of variables.
result Detection of incompatibilities can falsify wrongly inferred causal relations.
We information-theoretically reformulate two measures of capacity from statistical learning theory: empirical VC-entropy and empirical Rademacher complexity. We show these capacity measures count the number of hypotheses about a dataset that a learning algorithm falsifies when it finds the classifier in its repertoire …
GE finds failures in autonomous systems without domain heuristics.
problem Finding failures in autonomous systems without domain-specific heuristics.
method Adaptive stress testing using go-explore (GE) algorithm.
result GE finds failures in scenarios other RL techniques cannot solve.
The paper argues that machine learning is a falsificationist process.
problem The role of falsification in machine learning is underexplored.
method The paper presents a falsificationist account of artificial neural networks, emphasizing empirical risk minimization and implicit regularization.
result Artificial neural networks can be seen as a falsificationist process, rejecting inadequate prediction rules.
Study cyber-attacks on RL algorithms, focusing on cost signal manipulation.
problem Adversarial manipulation of cost signals in reinforcement learning.
method Quantitative analysis of TD(λ) and Q-learning algorithms under manipulation. result Bound on approximation error for TD(λ) and convergence properties for Q-learning under stealthy attacks. There are (at least) three approaches to quantifying information. The first, algorithmic information or Kolmogorov complexity, takes events as strings and, given a universal Turing machine, quantifies the information content of a string as the length of the shortest program producing it. The second, Shannon information…
New method tightens bounds on causation probabilities using independent datasets.
problem Challenging point identification of causation probabilities without strong assumptions.
method Imposes counterfactual consistency between SCMs constructed from independent datasets and uses conditional mutual information.
result Significantly tighter bounds on causation probabilities are established.
New methods improve robust decision-making under uncertainty in off-policy evaluation.
problem Statistical uncertainty and causal considerations in off-policy evaluation.
method Marginal Ratio (MR) estimator, Conformal Off-Policy Prediction (COPP), causal bounds.
result Improved robustness and uncertainty quantification in off-policy decision-making.
Faster Tsetlin Machines use clause indexing to speed inference and learning.
problem Overfitting and slow inference in Tsetlin Machines.
method Introduced a look-up table that indexes clauses based on feature falsification, enabling faster evaluation of clauses.
result Up to 15 times faster classification and three times faster learning on MNIST and Fashion-MNIST.
Causal Bayesian networks interpret actions as interventions to connect models to real-world outcomes.
problem Connecting causal model predictions to real-world outcomes.
method Formal framework to interpret actions as interventions and prove impossibility results.
result No non-circular interpretation exists that satisfies natural desiderata without violating some.
Many methods for automated software test generation, including some that explicitly use machine learning (and some that use ML more broadly conceived) derive new tests from existing tests (often referred to as seeds). Often, the seed tests from which new tests are derived are manually constructed, or at least simpler t…
A family of maximum mean discrepancy (MMD) kernel two-sample tests is introduced. Members of the test family are called Block-tests or B-tests, since the test statistic is an average over MMDs computed on subsets of the samples. The choice of block size allows control over the tradeoff between test power and computatio…
USP test improves on Pearson's chi-squared and G-test for independence.
problem Deficiencies in Pearson's chi-squared and G-test for independence. method USP test based on U-statistic estimator of population dependence measure. result USP test controls size, handles small cell counts, and detects minimal violations of independence.
E-C2ST uses E-values for high-dimensional data two-sample tests.
problem Statistical testing for high-dimensional data.
method Combines split likelihood ratio tests and predictive independence tests, using E-values for anytime-valid sequential tests.
result E-C2ST achieves enhanced statistical power by partitioning datasets into multiple batches.
This paper provides a comprehensive survey of Machine Learning Testing (ML testing) research. It covers 144 papers on testing properties (e.g., correctness, robustness, and fairness), testing components (e.g., the data, learning program, and framework), testing workflow (e.g., test generation and test evaluation), and …
Cheap permutation tests speed up distribution testing without sacrificing accuracy.
problem Efficiently testing distribution differences and independence.
method Group datapoints into bins and permute only these bins, using stored sufficient statistics.
result Cheap permutation tests maintain the accuracy and optimality of standard tests but are significantly faster.
Paper proposes a chi-square test for distance correlation.
problem Testing distance correlation is computationally expensive.
method Proposes a chi-square test for distance correlation, non-parametric, fast, applicable to various metrics.
result Chi-square test exhibits similar power to permutation test and can be valid and universally consistent for testing independence.
DRIFT uses RL to automate functional software testing efficiently.
problem Efficient and reliable automated software testing.
method DRIFT employs Q-learning with Graph Neural Networks on symbolic UI representations.
result DRIFT can robustly test software functionalities in a fully automated manner.
Optimizes two-sample tests for non-Euclidean domains using spectral regularization.
problem Optimizing two-sample tests for non-Euclidean domains.
method Spectral regularization of MMD test to achieve minimax optimality.
result Proposes a spectral regularization method that improves test optimality.
Robust test for distributions under Hellinger distance, simpler than optimal tests.
problem Testing and estimating distributions robustly under Hellinger distance.
method Simple robust hypothesis test with optimal sample complexity, robust to Hellinger distance perturbations.
result Empirically demonstrated robustness and power of the test on canonical distributions.
A new method for kernel tests without data splitting increases power.
problem Lack of power in kernel-based tests due to data splitting.
method Selective inference framework to learn hyperparameters and test on full sample.
result Empirically larger test power without data splitting, regardless of split proportion.
Framework for online hypothesis testing across various data types.
problem Testing various nonparametric hypotheses in data streams.
method Unified framework using operators on data distributions, leveraging ML models.
result Efficient, adaptive, and error-controlled sequential tests.
Model-X test detects conditional independence in streaming data.
problem Detecting conditional independence in data streams with arbitrary dependency.
method Sequential testing inspired by model-X and testing by betting.
result Significantly reduces type-I error rate and enhances data efficiency.
Simple methods combine statistical tests for out-of-distribution detection.
problem Detecting data points not following the training distribution.
method Combining classical parametric tests (Rao's score test) and a typicality test.
result Combining Fisher's method of test statistics improves out-of-distribution detection accuracy.
Unified score and distance-based GoF tests for model adequacy.
problem Difficulty in extending score-based GoF tests to nonparametric alternatives.
method Introducing semiparametric kernelized Stein discrepancy (SKSD) test.
result SKSD test is computationally efficient and universally consistent.
AutoML simplifies two-sample tests for detecting distribution shifts.
problem Detecting distribution shifts between datasets.
method Uses mean discrepancy of a witness function with squared loss minimization.
result AutoML simplifies and improves two-sample testing performance.
New methods for CI testing under model misspecification.
problem Challenges in CI testing with misspecified models.
method Proposes new approximations and upper bounds for testing errors of regression-based CI tests.
result Introduces the Rao-Blackwellized Predictor Test (RBPT) robust against misspecified inductive biases.
New KCM tests improve specification testing via RKHS.
problem Improving specification tests for econometric models.
method Kernel conditional moment (KCM) tests based on RKHS.
result KCM tests have better finite-sample performance than existing tests.
Discusses MultiFIT for multivariate dependence, comparing it to HSIC tests.
problem Comparing Multiscale Fisher's Independence Test (MultiFIT) to HSIC tests for multivariate dependence.
method Compares MultiFIT to HSIC tests, highlighting exact level control and performance limitations.
result Observes performance limitations of MultiFIT in terms of test power.
The repeated community-wide reuse of test sets in popular benchmark problems raises doubts about the credibility of reported test-error rates. Verifying whether a learned model is overfitted to a test set is challenging as independent test sets drawn from the same data distribution are usually unavailable, while other …
Develops hypothesis tests for conditional distributions using learning-theoretic bounds.
problem Testing differences in conditional distributions and functionals.
method Transforming learning-theoretic bounds into hypothesis tests for conditional expectations.
result Establishes comprehensive foundation for conditional testing, including theoretical guarantees and practical implementations.
Develops tests for conditional symmetry under group actions.
problem Testing conditional symmetry in distributions under group actions.
method Nonparametric randomization tests with kernel methods and asymptotic consistency.
result Tests achieve finite-sample Type I error control and power.
Paper proposes a new method for more accurate group testing of infected patients.
problem Identifying infected patients efficiently with reduced tests and corrected errors.
method Adaptive design of pools based on Bayesian posterior prediction using belief propagation algorithm.
result The proposed method results in more accurate identification of infected patients.
BERET improves binary expansion test for multivariate independence.
problem Testing independence of random vectors in arbitrary dimensions.
method Ensemble approach using sum of squared symmetry statistics and distance correlation.
result Improves power while preserving interpretability.