Improved scalability and interpretability in training data attribution.
problem Identifying which training data drives specific behaviors, especially unintended ones.
method Leveraging interpretable structures within the model to attribute model behavior to semantic directions, not individual test examples.
result Simple probe-based attribution methods are first-order approximations of Concept Influence that achieve comparable performance while being over an order-of-magnitude faster.
New method predicts model output distributions to improve data attribution.
problem Traditional training data attribution ignores randomness in model training.
method Distributional Training Data Attribution (d-TDA) using influence functions.
result Influence functions are effective and emerge naturally from d-TDA.
TRAK traces model predictions to training data efficiently.
problem Inefficiency in data attribution methods for large-scale models.
method TRAK: a new data attribution method that is both effective and computationally tractable.
result TRAK matches the performance of methods requiring thousands of models with just a handful.
In-Run Data Shapley offers efficient data attribution for large-scale models.
problem Existing data attribution methods are computationally intensive and cannot target specific models.
method In-Run Data Shapley, which efficiently attributes data contributions to a specific model without re-training.
result In-Run Data Shapley achieves significant efficiency, enabling data attribution for pretraining models.
Nonparametric method measures influence of training images on diffusion model outputs.
problem Quantifying influence of individual training examples on diffusion model outputs.
method Patch-level similarity between generated and training images, using optimal score function.
result Strong attribution performance, matching gradient-based approaches and outperforming baselines.
This work addresses fairness in ML models by training and evaluating attribute classifiers under uncertain and incomplete data.
problem Challenges in fairness metrics due to uncertain and incomplete data.
method Developed a theoretical and empirical analysis to understand and improve bias estimation in the data-scarce regime.
result The test accuracy of the attribute classifier is not always correlated with its effectiveness in bias estimation.
Algorithm uncovers latent attribute graph from molecular data.
problem Learning latent representations and interpreting them for limited data.
method Perturbation experiments on latent codes of a generative autoencoder.
result Effective graphical model of latent codes and attributes.
GGDA simplifies DA for large models, speeding up attribution by up to 50x.
problem Computational intensity of existing DA methods limits their applicability to large-scale models.
method Generalized Group Data Attribution (GGDA) framework attributing to groups of training points.
result GGDA achieves up to 50x speedups over standard DA methods while maintaining effectiveness.
Proposes new attribution methods for trees with regularization.
problem Feature attribution for trees trained with regularization.
method Prediction Decomposition Attribution (PreDecomp) and TreeInner.
result TreeInner shows state-of-the-art feature selection performance.
New method reveals how training data influence diffusion model outputs.
problem Difficulty in assessing training data impact on diffusion model outputs.
method Use of ensembles trained on carefully engineered splits of training data to identify influential training examples.
result Demonstrated the viability of ensembles as generative models and validity of assessing influence.
MAGIC method optimally estimates model predictions changes.
problem Estimating how training data affects model predictions in large-scale settings.
method Combines classical methods and recent advances in metadifferentiation.
result MAGIC method nearly optimally estimates model predictions changes.
Paper tackles source attribution for LLM-generated texts.
problem Identify the data provider of synthetic texts generated by LLMs.
method Watermarking to embed information about data providers in synthetic texts.
result Proposes and evaluates a source attribution framework that achieves effective source attribution.
Bayesian approach scores influential training examples for model predictions.
problem Enhance interpretability and safety of machine learning models.
method Formulate TDA as a Bayesian information-theoretic problem, scoring subsets by information loss.
result Method aligns with classical influence scores while promoting diversity for subsets.
MAIN network learns attributes without unseen class attributes for faster, more adaptable ZSL.
problem Learning unseen categories without known attributes and handling continual learning.
method Meta-learning attribute self-interaction network with inverse regularization.
result Main network outperforms state-of-the-art ZSL methods without unseen class attributes.
Study shows data attribution methods are sensitive to hyperparameters, making tuning costly.
problem Hyperparameter sensitivity in data attribution methods makes tuning impractical.
method Theoretical analysis and lightweight procedure for selecting regularization value without retraining.
result Proposes a lightweight procedure for selecting regularization value without model retraining.
Study on protecting sensitive properties of datasets during analysis.
problem Ensuring privacy of sensitive properties in datasets.
method Proposes definitions and mechanisms for attribute privacy using the Pufferfish framework.
result Developed efficient and inefficient mechanisms for attribute privacy.
ASTRA improves TDA by more accurately approximating iHVP.
problem Improving insights into training data attribution.
method ASTRA uses EKFAC-preconditioner on Neumann series iterations to accurately approximate iHVP.
result Improving iHVP approximation significantly improves TDA performance.
We show new connections between adversarial learning and explainability for deep neural networks (DNNs). One form of explanation of the output of a neural network model in terms of its input features, is a vector of feature-attributions. Two desirable characteristics of an attribution-based explanation are: (1) $\texti…
"Overlearning" means that a model trained for a seemingly simple objective implicitly learns to recognize attributes and concepts that are (1) not part of the learning objective, and (2) sensitive from a privacy or bias perspective. For example, a binary gender classifier of facial images also learns to recognize races…
PASS protects private attributes by stochastically substituting data.
problem Protecting private attributes in ML services while maintaining data utility.
method PASS uses stochastic data substitution with a novel loss function derived from information theory.
result PASS effectively protects private attributes across various datasets.
GPT-GNN pre-trains GNNs on unlabeled graphs to improve downstream performance.
problem Training GNNs requires labeled data, which is expensive.
method Generative pre-training of GNNs on unlabeled data with self-supervision.
result GPT-GNN significantly outperforms state-of-the-art GNNs without pre-training.
Improved data attribution for high-dimensional models.
problem Inaccurate influence functions in high-dimensional settings.
method Rescaled influence functions (RIF) for more precise data attribution.
result RIFs offer significantly better predictions in practice.
New framework detects out-of-distribution data by considering intrinsic ID attributes in outliers.
problem Deploying reliable machine learning systems requires effective out-of-distribution detection.
method Structured multi-view-based out-of-distribution detection learning (MVOL) framework.
result MVOL effectively utilizes both auxiliary OOD datasets and wild datasets with noisy in-distribution data.
PSI models and infers feature attributions efficiently and accurately.
problem Modeling and inferring feature attributions in flexible predictive models.
method Probabilistic Shapley inference (PSI) framework using latent random variables and a masking-based neural network architecture.
result PSI learns feature attribution distributions centered at Shapley values, revealing meaningful uncertainty.
We are interested in attribute-guided face generation: given a low-res face input image, an attribute vector that can be extracted from a high-res image (attribute image), our new method generates a high-res face image for the low-res input that satisfies the given attributes. To address this problem, we condition the …
A new method for disentangled latent spaces in VAEs that can manipulate attributes.
problem Disentangled representation of attributes in latent spaces of VAEs.
method Attribute-based regularization loss to enforce monotonic relationships between attributes and latent codes.
result Manipulation of attributes in latent spaces post-training.
Method generates time-series attribution maps with identifiability guarantees.
problem Lack of identifiability guarantees in gradient-based attribution methods.
method Regularized contrastive learning algorithm trained on time-series data.
result Empirically shows robust approximation of zero vs. non-zero entries in the ground-truth attribution map.
Develops methods to measure and reduce fairness in datasets with limited protected attribute labels.
problem Measuring and reducing fairness in datasets with limited protected attribute labels.
method Proposes methods to estimate fairness metrics and train models to limit fairness violations using probabilistic protected attribute labels.
result Our methods provide tighter bounds on true disparity and effectively reduce fairness violations with lesser fairness-accuracy trade-offs.
Paper proposes a new method to measure model sensitivity using final model only.
problem Understanding model behavior using only the final trained model.
method Reframe TDA as measuring sensitivity, propose further training as gold standard, unify gradient-based methods.
result Gradient-based methods approximate further training but vary in quality.
Improved conversion rate prediction in online advertising using self-supervised pre-training.
problem Data sparsity and calibration issues in predicting conversions given clicks.
method Self-supervised pre-training on all conversion events to enrich CVR prediction model without compromising calibration.
result Improvements in offline training and online A/B tests, with full deployment to Yahoo native advertising system.
I consider how to influence CycleGAN, image-to-image translation, by using additional constraints from a neural network trained on art composition attributes. I show how I trained the the Art Composition Attributes Network (ACAN) by incorporating domain knowledge based on the rules of art evaluation and the result of a…
Infer-AVAE infers missing user attributes from incomplete data using a novel adversarial approach.
problem Incomplete user attributes in social networks.
method Infer-AVAE combines MLP and GNNs with adversarial training to infer missing attributes.
result Infer-AVAE outperforms baselines by 7.0% in accuracy on real-world datasets.
New bounds on machine learning data leakage identified.
problem Machine Learning models can leak sensitive information.
method Formalized attack setups, derived universal bounds, studied mutual information.
result Connected attack success rate to generalization gap and mutual information.
In predictive process analytics, current and historical process data in event logs is used to predict the future, e.g., to predict the next activity or how long a process will still require to complete. Recurrent neural networks (RNN) and its subclasses have been demonstrated to be well suited for creating prediction m…
In this paper we investigate the usage of adversarial perturbations for the purpose of privacy from human perception and model (machine) based detection. We employ adversarial perturbations for obfuscating certain variables in raw data while preserving the rest. Current adversarial perturbation methods are used for dat…
In principle, zero-shot learning makes it possible to train a recognition model simply by specifying the category's attributes. For example, with classifiers for generic attributes like \emph{striped} and \emph{four-legged}, one can construct a classifier for the zebra category by enumerating which properties it posses…
SHAP Distance assesses semantic fidelity of synthetic tabular data.
problem Semantic fidelity of synthetic tabular data is not well evaluated.
method SHAP Distance, defined as cosine distance between global SHAP attribution vectors.
result SHAP Distance detects semantic discrepancies overlooked by standard measures.
Ground-A-Video edits videos without training, preserving intended changes.
problem Complex multi-attribute video editing with omitted or wrong changes.
method Grounding-guided video-to-video translation with Cross-Frame Gated Attention.
result Zero-shot multi-attribute video editing with improved accuracy and frame consistency.
Recent research has demonstrated that feature attribution methods for deep networks can themselves be incorporated into training; these attribution priors optimize for a model whose attributions have certain desirable properties -- most frequently, that particular features are important or unimportant. These attributio…
Deep neural network algorithms are difficult to analyze because they lack structure allowing to understand the properties of underlying transforms and invariants. Multiscale hierarchical convolutional networks are structured deep convolutional networks where layers are indexed by progressively higher dimensional attrib…
CDLEEDS detects local changes in evolving data streams for accurate feature attributions.
problem Local feature attributions become obsolete in evolving data streams.
method CDLEEDS, a flexible framework for detecting local change and concept drift.
result CDLEEDS reliably detects both local and global concept drift.
Study uncovers bias in image classification models using attribution maps.
problem Data bias in image classification models.
method Created an artificial dataset with known bias, trained CNNs, and used attribution maps to inspect decisions.
result Different attribution map techniques highlight bias better than others, and metrics support bias identification.
In this paper we analyze a budgeted learning setting, in which the learner can only choose and observe a small subset of the attributes of each training example. We develop efficient algorithms for ridge and lasso linear regression, which utilize the geometry of the data by a novel data-dependent sampling scheme. When …
RoSHAP stabilizes feature attribution in machine learning models.
problem Stochastic variation in feature attribution measures.
method Modeling feature attribution score distribution and estimating it through bootstrap resampling and kernel density estimation.
result RoSHAP provides stable feature rankings and improves model performance.
Develops new methods to evaluate data influence in SAM for improved model training.
problem Challenges in mislabeled noisy data and privacy concerns in SAM.
method Two innovative data valuation methods based on influence functions (IF) for SAM.
result Demonstrates effectiveness in identifying mislabeled data and enhancing interpretability.
Paper creates fair synthetic data ensuring equal predictions across sensitive attributes.
problem Ensuring fair predictions across sensitive attributes in synthetic data.
method Equalizing target probability distributions across sensitive attributes in synthetic data generation.
result Synthetic data provides strong fair predictions, equal across all thresholds.
New method for efficient sketching of gradients and Hessians.
problem Memory constraints in training machine learning models.
method A novel framework for scalable gradient and HVP sketching tailored for modern hardware.
result Theoretical guarantees and practical applications in training data attribution and Hessian spectrum analysis.
Text-based analysis methods allow to reveal privacy relevant author attributes such as gender, age and identify of the text's author. Such methods can compromise the privacy of an anonymous author even when the author tries to remove privacy sensitive content. In this paper, we propose an automatic method, called Adver…