Feature attribution methods identify which features of an input most influence a model's output. Most widely-used feature attribution methods (such as SHAP, LIME, and Grad-CAM) are "class-dependent" methods in that they generate a feature attribution vector as a function of class. In this work, we demonstrate that class-dependent methods can "leak" information about the selected class, making that class appear more likely than it is. Thus, an end user runs the risk of drawing false conclusions when interpreting an explanation generated by a class-dependent method. In contrast, we introduce "distribution-aware" methods, which favor explanations that keep the label's distribution close to its distribution given all features of the input. We introduce SHAP-KL and FastSHAP-KL, two baseline distribution-aware methods that compute Shapley values. Finally, we perform a comprehensive evaluation of seven class-dependent and three distribution-aware methods on three clinical datasets of different high-dimensional data types: images, biosignals, and text.
Deep Neural Networks (DNNs) are prone to learn shortcut patterns that damage the generalization of the DNN during deployment. Shortcut Learning is concerning, particularly when the DNNs are applied to safety-critical domains. This paper aims to better understand shortcut learning through the lens of the learning dynamics of the internal neurons during the training process. More specifically, we make the following observations: (1) While previous works treat shortcuts as synonymous with spurious correlations, we emphasize that not all spurious correlations are shortcuts. We show that shortcuts are only those spurious features that are "easier" than the core features. (2) We build upon this premise and use instance difficulty methods (like Prediction Depth) to quantify "easy" and to identify this behavior during the training phase. (3) We empirically show that shortcut learning can be detected by observing the learning dynamics of the DNN's early layers, irrespective of the network architecture used. In other words, easy features learned by the initial layers of a DNN early during the training are potential shortcuts. We verify our claims on simulated and real medical imaging data and justify the empirical success of our hypothesis by showing the theoretical connections between Prediction Depth and information-theoretic concepts like V-usable information. Lastly, our experiments show the insufficiency of monitoring only accuracy plots during training (as is common in machine learning pipelines), and we highlight the need for monitoring early training dynamics using example difficulty metrics.
Diffusion-based generative models (DBGMs) perturb data to a target noise distribution and reverse this inference diffusion process to generate samples. The choice of inference diffusion affects both likelihoods and sample quality. For example, extending the inference process with auxiliary variables leads to improved sample quality. While there are many such multivariate diffusions to explore, each new one requires significant model-specific analysis, hindering rapid prototyping and evaluation. In this work, we study Multivariate Diffusion Models (MDMs). For any number of auxiliary variables, we provide a recipe for maximizing a lower-bound on the MDMs likelihood without requiring any model-specific analysis. We then demonstrate how to parameterize the diffusion for a specified target noise distribution; these two points together enable optimizing the inference diffusion process. Optimizing the diffusion expands easy experimentation from just a few well-known processes to an automatic search over all linear diffusions. To demonstrate these ideas, we introduce two new specific diffusions as well as learn a diffusion process on the MNIST, CIFAR10, and ImageNet32 datasets. We show learned MDMs match or surpass bits-per-dims (BPDs) relative to fixed choices of diffusions for a given dataset and model architecture.
Methods which utilize the outputs or feature representations of predictive models have emerged as promising approaches for out-of-distribution (OOD) detection of image inputs. However, these methods struggle to detect OOD inputs that share nuisance values (e.g. background) with in-distribution inputs. The detection of shared-nuisance out-of-distribution (SN-OOD) inputs is particularly relevant in real-world applications, as anomalies and in-distribution inputs tend to be captured in the same settings during deployment. In this work, we provide a possible explanation for SN-OOD detection failures and propose nuisance-aware OOD detection to address them. Nuisance-aware OOD detection substitutes a classifier trained via empirical risk minimization and cross-entropy loss with one that 1. is trained under a distribution where the nuisance-label relationship is broken and 2. yields representations that are independent of the nuisance under this distribution, both marginally and conditioned on the label. We can train a classifier to achieve these objectives using Nuisance-Randomized Distillation (NuRD), an algorithm developed for OOD generalization under spurious correlations. Output- and feature-based nuisance-aware OOD detection perform substantially better than their original counterparts, succeeding even when detection based on domain generalization algorithms fails to improve performance.
Early detection of many life-threatening diseases (e.g., prostate and breast cancer) within at-risk population can improve clinical outcomes and reduce cost of care. While numerous disease-specific "screening" tests that are closer to Point-of-Care (POC) are in use for this task, their low specificity results in unnecessary biopsies, leading to avoidable patient trauma and wasteful healthcare spending. On the other hand, despite the high accuracy of Magnetic Resonance (MR) imaging in disease diagnosis, it is not used as a POC disease identification tool because of poor accessibility. The root cause of poor accessibility of MR stems from the requirement to reconstruct high-fidelity images, as it necessitates a lengthy and complex process of acquiring large quantities of high-quality k-space measurements. In this study we explore the feasibility of an ML-augmented MR pipeline that directly infers the disease sidestepping the image reconstruction process. We hypothesise that the disease classification task can be solved using a very small tailored subset of k-space data, compared to image reconstruction. Towards that end, we propose a method that performs two tasks: 1) identifies a subset of the k-space that maximizes disease identification accuracy, and 2) infers the disease directly using the identified k-space subset, bypassing the image reconstruction step. We validate our hypothesis by measuring the performance of the proposed system across multiple diseases and anatomies. We show that comparable performance to image-based classifiers, trained on images reconstructed with full k-space data, can be achieved using small quantities of data: 8% of the data for detecting multiple abnormalities in prostate and brain scans, and 5% of the data for knee abnormalities. To better understand the proposed approach and instigate future research, we provide an extensive analysis and release code.
There exist features that are related to the label in the same way across different settings for that task; these are semantic features or semantics. Features with varying relationships to the label are nuisances. For example, in detecting cows from natural images, the shape of the head is a semantic and because images of cows often have grass backgrounds but only in certain settings, the background is a nuisance. Relationships between a nuisance and the label are unstable across settings and, consequently, models that exploit nuisance-label relationships face performance degradation when these relationships change. Direct knowledge of a nuisance helps build models that are robust to such changes, but knowledge of a nuisance requires extra annotations beyond the label and the covariates. In this paper, we develop an alternative way to produce robust models by data augmentation. These data augmentations corrupt semantic information to produce models that identify and adjust for where nuisances drive predictions. We study semantic corruptions in powering different robust-modeling methods for multiple out-of distribution (OOD) tasks like classifying waterbirds, natural language inference, and detecting Cardiomegaly in chest X-rays.
Treatment decisions in cancer care are guided by treatment effect estimates from randomized controlled trials (RCTs). RCTs estimate the average effect of one treatment versus another in a certain population. However, treatments may not be equally effective for every patient in a population. Knowing the effectiveness of treatments tailored to specific patient and tumor characteristics would enable individualized treatment decisions. Getting tailored treatment effects by averaging outcomes in different patient subgroups in RCTs requires an unfeasible number of patients to have sufficient statistical power in all relevant subgroups for all possible treatments. The American Joint Committee on Cancer (AJCC) recommends that researchers develop outcome prediction models (OPMs) in an effort to individualize treatment decisions. OPMs sometimes called risk models or prognosis models, use patient and tumor characteristics to predict a patient outcome such as overall survival. The assumption is that the predictions are useful for treatment decisions using rules such as "prescribe chemotherapy only if the OPM predicts the patient has a high risk of recurrence". Recognizing the importance of reliable predictions, the AJCC published a checklist for OPMs to ensure dependable OPM prediction accuracy in the patient population for which the OPM was designed. However, accurate outcome predictions do not imply that these predictions yield good treatment decisions. In this perspective, we show that OPM rely on a fixed treatment policy which implies that OPM that were found to accurately predict outcomes in validation studies can still lead to patient harm when used to inform treatment decisions. We then give guidance on how to develop models that are useful for individualized treatment decisions and how to evaluate whether a model has value for decision-making.
Survival analysis, the art of time-to-event modeling, plays an important role in clinical treatment decisions. Recently, continuous time models built from neural ODEs have been proposed for survival analysis. However, the training of neural ODEs is slow due to the high computational complexity of neural ODE solvers. Here, we propose an efficient alternative for flexible continuous time models, called Survival Mixture Density Networks (Survival MDNs). Survival MDN applies an invertible positive function to the output of Mixture Density Networks (MDNs). While MDNs produce flexible real-valued distributions, the invertible positive function maps the model into the time-domain while preserving a tractable density. Using four datasets, we show that Survival MDN performs better than, or similarly to continuous and discrete time baselines on concordance, integrated Brier score and integrated binomial log-likelihood. Meanwhile, Survival MDNs are also faster than ODE-based models and circumvent binning issues in discrete models.
Conditional randomization tests (CRTs) assess whether a variable $x$ is predictive of another variable $y$, having observed covariates $z$. CRTs require fitting a large number of predictive models, which is often computationally intractable. Existing solutions to reduce the cost of CRTs typically split the dataset into a train and test portion, or rely on heuristics for interactions, both of which lead to a loss in power. We propose the decoupled independence test (DIET), an algorithm that avoids both of these issues by leveraging marginal independence statistics to test conditional independence relationships. DIET tests the marginal independence of two random variables: $F(x \mid z)$ and $F(y \mid z)$ where $F(\cdot \mid z)$ is a conditional cumulative distribution function (CDF). These variables are termed "information residuals." We give sufficient conditions for DIET to achieve finite sample type-1 error control and power greater than the type-1 error rate. We then prove that when using the mutual information between the information residuals as a test statistic, DIET yields the most powerful conditionally valid test. Finally, we show DIET achieves higher power than other tractable CRTs on several synthetic and real benchmarks.