From Instructions to Constraints: Language Model Alignment with Automatic Constraint Verification

User alignment is crucial for adapting general-purpose language models (LMs) to downstream tasks, but human annotations are often not available for all types of instructions, especially those with customized constraints. We observe that user instructions typically contain constraints. While assessing response quality in terms of the whole instruction is often costly, efficiently evaluating the satisfaction rate of constraints is feasible. We investigate common constraints in NLP tasks, categorize them into three classes based on the types of their arguments, and propose a unified framework, ACT (Aligning to ConsTraints), to automatically produce supervision signals for user alignment with constraints. Specifically, ACT uses constraint verifiers, which are typically easy to implement in practice, to compute constraint satisfaction rate (CSR) of each response. It samples multiple responses for each prompt and collect preference labels based on their CSR automatically. Subsequently, ACT adapts the LM to the target task through a ranking-based learning process. Experiments on fine-grained entity typing, abstractive summarization, and temporal question answering show that ACT is able to enhance LMs' capability to adhere to different classes of constraints, thereby improving task performance. Further experiments show that the constraint-following capabilities are transferable.

Game-theoretic Counterfactual Explanation for Graph Neural Networks

Graph Neural Networks (GNNs) have been a powerful tool for node classification tasks in complex networks. However, their decision-making processes remain a black-box to users, making it challenging to understand the reasoning behind their predictions. Counterfactual explanations (CFE) have shown promise in enhancing the interpretability of machine learning models. Prior approaches to compute CFE for GNNS often are learning-based approaches that require training additional graphs. In this paper, we propose a semivalue-based, non-learning approach to generate CFE for node classification tasks, eliminating the need for any additional training. Our results reveals that computing Banzhaf values requires lower sample complexity in identifying the counterfactual explanations compared to other popular methods such as computing Shapley values. Our empirical evidence indicates computing Banzhaf values can achieve up to a fourfold speed up compared to Shapley values. We also design a thresholding method for computing Banzhaf values and show theoretical and empirical results on its robustness in noisy environments, making it superior to Shapley values. Furthermore, the thresholded Banzhaf values are shown to enhance efficiency without compromising the quality (i.e., fidelity) in the explanations in three popular graph datasets.

A deep learning pipeline for cross-sectional and longitudinal multiview data integration

Biomedical research now commonly integrates diverse data types or views from the same individuals to better understand the pathobiology of complex diseases, but the challenge lies in meaningfully integrating these diverse views. Existing methods often require the same type of data from all views (cross-sectional data only or longitudinal data only) or do not consider any class outcome in the integration method, presenting limitations. To overcome these limitations, we have developed a pipeline that harnesses the power of statistical and deep learning methods to integrate cross-sectional and longitudinal data from multiple sources. Additionally, it identifies key variables contributing to the association between views and the separation among classes, providing deeper biological insights. This pipeline includes variable selection/ranking using linear and nonlinear methods, feature extraction using functional principal component analysis and Euler characteristics, and joint integration and classification using dense feed-forward networks and recurrent neural networks. We applied this pipeline to cross-sectional and longitudinal multi-omics data (metagenomics, transcriptomics, and metabolomics) from an inflammatory bowel disease (IBD) study and we identified microbial pathways, metabolites, and genes that discriminate by IBD status, providing information on the etiology of IBD. We conducted simulations to compare the two feature extraction methods. The proposed pipeline is available from the following GitHub repository: https://github.com/lasandrall/DeepIDA-GRU.

How Many and Which Training Points Would Need to be Removed to Flip this Prediction?

We consider the problem of identifying a minimal subset of training data $\mathcal{S}_t$ such that if the instances comprising $\mathcal{S}_t$ had been removed prior to training, the categorization of a given test point $x_t$ would have been different. Identifying such a set may be of interest for a few reasons. First, the cardinality of $\mathcal{S}_t$ provides a measure of robustness (if $|\mathcal{S}_t|$ is small for $x_t$, we might be less confident in the corresponding prediction), which we show is correlated with but complementary to predicted probabilities. Second, interrogation of $\mathcal{S}_t$ may provide a novel mechanism for contesting a particular model prediction: If one can make the case that the points in $\mathcal{S}_t$ are wrongly labeled or irrelevant, this may argue for overturning the associated prediction. Identifying $\mathcal{S}_t$ via brute-force is intractable. We propose comparatively fast approximation methods to find $\mathcal{S}_t$ based on influence functions, and find that -- for simple convex text classification models -- these approaches can often successfully identify relatively small sets of training examples which, if removed, would flip the prediction.

Influence Functions for Sequence Tagging Models

Many language tasks (e.g., Named Entity Recognition, Part-of-Speech tagging, and Semantic Role Labeling) are naturally framed as sequence tagging problems. However, there has been comparatively little work on interpretability methods for sequence tagging models. In this paper, we extend influence functions - which aim to trace predictions back to the training points that informed them - to sequence tagging tasks. We define the influence of a training instance segment as the effect that perturbing the labels within this segment has on a test segment level prediction. We provide an efficient approximation to compute this, and show that it tracks with the true segment influence, measured empirically. We show the practical utility of segment influence by using the method to identify systematic annotation errors in two named entity recognition corpora. Code to reproduce our results is available at https://github.com/successar/Segment_Influence_Functions.

Modular Self-Supervision for Document-Level Relation Extraction

Extracting relations across large text spans has been relatively underexplored in NLP, but it is particularly important for high-value domains such as biomedicine, where obtaining high recall of the latest findings is crucial for practical applications. Compared to conventional information extraction confined to short text spans, document-level relation extraction faces additional challenges in both inference and learning. Given longer text spans, state-of-the-art neural architectures are less effective and task-specific self-supervision such as distant supervision becomes very noisy. In this paper, we propose decomposing document-level relation extraction into relation detection and argument resolution, taking inspiration from Davidsonian semantics. This enables us to incorporate explicit discourse modeling and leverage modular self-supervision for each sub-problem, which is less noise-prone and can be further refined end-to-end via variational EM. We conduct a thorough evaluation in biomedical machine reading for precision oncology, where cross-paragraph relation mentions are prevalent. Our method outperforms prior state of the art, such as multi-scale learning and graph neural networks, by over 20 absolute F1 points. The gain is particularly pronounced among the most challenging relation instances whose arguments never co-occur in a paragraph.

Combining Feature and Instance Attribution to Detect Artifacts

Training the large deep neural networks that dominate NLP requires large datasets. Many of these are collected automatically or via crowdsourcing, and may exhibit systematic biases or annotation artifacts. By the latter, we mean correlations between inputs and outputs that are spurious, insofar as they do not represent a generally held causal relationship between features and classes; models that exploit such correlations may appear to perform a given task well, but fail on out of sample data. In this paper we propose methods to facilitate identification of training data artifacts, using new hybrid approaches that combine saliency maps (which highlight important input features) with instance attribution methods (which retrieve training samples influential to a given prediction). We show that this proposed training-feature attribution approach can be used to uncover artifacts in training data, and use it to identify previously unreported artifacts in a few standard NLP datasets. We execute a small user study to evaluate whether these methods are useful to NLP researchers in practice, with promising results. We make code for all methods and experiments in this paper available.

Does BERT Pretrained on Clinical Notes Reveal Sensitive Data?

Large Transformers pretrained over clinical notes from Electronic Health Records (EHR) have afforded substantial gains in performance on predictive clinical tasks. The cost of training such models (and the necessity of data access to do so) coupled with their utility motivates parameter sharing, i.e., the release of pretrained models such as ClinicalBERT. While most efforts have used deidentified EHR, many researchers have access to large sets of sensitive, non-deidentified EHR with which they might train a BERT model (or similar). Would it be safe to release the weights of such a model if they did? In this work, we design a battery of approaches intended to recover Personal Health Information (PHI) from a trained BERT. Specifically, we attempt to recover patient names and conditions with which they are associated. We find that simple probing methods are not able to meaningfully extract sensitive information from BERT trained over the MIMIC-III corpus of EHR. However, more sophisticated "attacks" may succeed in doing so: To facilitate such research, we make our experimental setup and baseline probing models available at https://github.com/elehman16/exposing_patient_data_release