Publications
† students, * corresponding author
Preprints
2026
- SSRNPeter S. Lee†, K. Sudhir, and Tong Wang
Sequential content—such as episodic fiction, television series, and online courses—requires consumers to make sequential decisions about whether to continue by incurring access costs or to exit. These choices reflect forward-looking tradeoffs among engagement, delay, and payment, and depend jointly on platform policies and the content encountered along the progression. Yet incorporating high-dimensional content into dynamic choice models remains challenging. We address this by extending adversarial inverse reinforcement learning (AIRL) to accommodate unobserved heterogeneity, identifiable action-dependent utilities, and scalable representations of unstructured content within a structural dynamic framework. We apply the model to 15 months of detailed readership and payment data from a serialized fiction platform. The estimates reveal two dominant segments—Pay & Read and Wait & Read—that account for most consumption and nearly all purchases. Counterfactual analyses show that free access can increase downstream purchases within a user, delay policies can shift engagement toward paid consumption, and content-aware pricing improves monetization while preserving overall usage. More broadly, the paper provides a scalable approach for estimating dynamic choice models with high-dimensional state spaces, offering a tractable framework for studying sequential engagement in modern digital markets.
2025
- SSRNTong Wang, K. Sudhir, and Hengguang Zhou
- When AI Learns the Why, It Becomes Smarter—and More Responsible, Yale Insights, December 2025
Generative AI models are increasingly used to produce marketing content. Since off-the-shelf models are misaligned with desired marketing outcomes, they are fine-tuned using context experiments that identify what content is correlated with higher engagement. Yet optimizing only for what works risks overfitting, reward hacking, and poor generalization, yielding content that succeeds in-sample but fail in new contexts or drift toward clickbait. We propose a principled knowledge-alignment framework that moves beyond merely what works to why it works. In our approach, an LLM iteratively generates hypotheses about mechanisms (e.g., emotional language, narrative framing) to explain observed performance differences on a small set of data (abduction), then validates them on held-out data (induction). The optimized set of validated hypotheses form an interpretable, domain-specific knowledge base that regularizes fine-tuning via Direct Preference Optimization (DPO), constraining the model toward generalizable principles. Our LLM-based approach extends the tradition of theory-guided machine learning to domains where relevant knowledge is tacit and therefore hard to explicitly encode in models. Using a dataset of over 23,000 A/B-tested news headlines across 4,500+ articles, we show that our knowledge-guided framework outperforms supervised fine-tuning, DPO and multi-dimensional DPO in improving engagement (click-through), while avoiding clickbait and maintaining lexical diversity.
2024
- ArxivTong Wang and K Sudhir
As large language models increasingly mediate firm-customer interactions, firms face a tradeoff: the most capable models perform well but are costly and difficult to control at scale. Existing knowledge distillation methods address this challenge by training weaker, deployable models to imitate frontier outputs; however, such open-loop approaches are poorly suited to interactive, multi-turn settings where responses must be sequenced coherently across conversational states. We propose a shift in what knowledge is distilled—from output imitation to contextual guidance. We develop a framework in which a superior teacher model constructs a reusable library of strategic textual guidance for particular scenarios likely to be encountered by the student. When deployed, the student retrieves the context-specific guidance at inference time, enabling adaptive behavior without retraining. Using customer-service interactions, we show that this approach improves service quality and customer satisfaction relative to standard fine-tuning while maintaining alignment with firm policies. The results position inference-time textual guidance as a scalable and controllable approach to distillation for interactive AI agents in marketing settings.
- SSRNTong Wang*, Ronilo Ragodos†, Lu Feng, and Jeffrey Yu Hu
Post hoc explainers such as SHAP and LIME are used widely in business research to interpret complex machine learning models. Although they were designed to explain model predictions, there has been an increasing trend in which the explanations they generate are treated as evidence about underlying data relationships. Based on a systematic review of 181 studies, including 56 published in leading journals, we document that this explanation interpretation is widespread and examine its validity. We also evaluate how well post hoc explanations reliably recover the direction and relative importance of features in true data-generating process. We introduce two metrics—direction alignment and strength alignment—and assess SHAP and LIME using simulated data with known ground truth. Although explanations often appear reasonable on average, they exhibit substantial heterogeneity in their alignment and can fail to be aligned even when their predictive accuracy is high. High predictive performance is therefore necessary but insufficient for reliable explanation. We further show that feature correlation and the Rashomon effect (where many equally accurate models rely on different feature attributions) are key drivers of misalignment. Agreement in explanations across such models provides a practical diagnostic of reliability. Overall, our findings caution against using post hoc explainers for hypothesis validation and instead position them as exploratory tools that generate, rather than confirm, substantive insights.
Marketing and Management
2024
- POMTong Wang, Fujie Jin, Yu Jeffrey Hu, Lu Feng, and Yuan ChengProduction and Operations Management
Medical crowdfunding is a popular channel for people seeking financial assistance to cover their medical expenses, allowing them to collect donations from a large number of donors. However, a mismatch between the supply and demand of donations creates large heterogeneity in the fundraising outcomes across medical crowdfunding campaigns, and such uncertainty can impede the timely planning of treatment for patients. Providing early and accurate forecasts for medical crowdfunding performance can better inform fundraisers and assist them in optimizing timely interventions to improve fundraising outcomes. In this study, we propose a new approach that effectively combines time-varying features and time-invariant features in a deep learning model, to provide dynamic predictions of fundraising outcomes. When compared with a comprehensive set of baseline models, our model consistently demonstrates higher predictive accuracy while requiring a shorter observation window of data, thus achieving both accurate and early prediction objectives. We further conduct a temporal clustering analysis to analyze the heterogeneous patterns in how the time-varying features relate to fundraising outcomes. In addition, we perform simulation analyses to demonstrate that interventions from fundraisers can significantly improve the fundraising performance of disadvantaged cases that are predicted to receive the lowest donation amounts, particularly when the interventions are implemented early. These findings show that our deep learning prediction model and the actionable insights can provide timely feedback to fundraisers and promote equal access to resources for all. Our proposed approach is applicable to various contexts, enabling effective processing of diverse sources of data and facilitating early interventions.
2023
- JMRAndrea Webb Luangrath, Yixiang Xu†, and Tong WangJournal of Marketing Research
Brands and consumers alike have become creators and distributors of digital words, thus generating increasing interest in insights to be gained from text-based content. This work develops an algorithm to identify textual paralanguage, defined as nonverbal parts of speech expressed in online communication. The authors develop and validate a paralanguage classifier (called PARA) using social media data from Twitter, YouTube, and Instagram (N = 1,241,489 posts). Using auditory, tactile, and visual properties of text, PARA detects nonverbal communication cues, aspects of text often neglected by other word-based sentiment lexica. This work is the first to reveal the importance of textual paralanguage as a critical indicator of sentiment valence and intensity. The authors further demonstrate that automatically detected textual paralanguage can predict consumer engagement above and beyond existing text analytics tools. The algorithm is designed for researchers, scholars, and practitioners seeking to optimize marketing communications and offers a methodological advancement to quantify the importance of not only what is said verbally but how it is said nonverbally.
2021
- ISRTong Wang, Cheng He, Fujie Jin, and Yu Jeffrey HuInformation Systems Research
We develop a novel interpretable machine learning model, GANNM, and use newly available data to evaluate how different types of marketing campaigns and budget allocations influence malls’ customer traffic. We observe that the response curves that measure the impact of campaign budget on customer traffic differ for different categories of campaigns, with sales incentives or experience incentives, during peak periods, off-peak periods, or online promotion periods. Based on such accurate response curves from GANNM, the optimized budget allocation is estimated to yield a 11.2% increase in customer traffic compared with the original allocation. Our findings provide novel insights on managing mall campaigns. Mall managers should increase marketing spending to areas that were likely overlooked before and avoid over-crowding budget to campaigns during times with high levels of competition and are likely already over-marketed. We provide empirical evidence showing that the recent trend of employing novel approaches for enhancing customer experience in physical stores can effectively encourage customers to visit malls. Furthermore, we show that online promotions could also create opportunities for offline businesses—investing in campaigns in the major online promotion periods could significantly increase customer traffic for malls, given sufficient investment in the campaigns to raise customer awareness.
Machine Learning Methods
2026
- MLM. Cristina Molero Del Rio†, Boxuan Li†, Tong Wang, and Cynthia RudinMachine Learning
A sparse rule set (SRS) is a small predictive model that is a disjunctive normal form – an “OR of ANDs”. SRS models are understandable to human experts and robust to outliers. Constructing an SRS efficiently has always been one of the fundamental problems of interpretable machine learning. However, the practical challenge goes beyond simply optimizing for a single sparse SRS; the first interpretable model an algorithm finds often has flaws that need to be fixed. Thus, constructing a model for practical use involves human interaction with an algorithm. If we work within the Rashomon set paradigm, we would generate many models that a human can explore; this simplifies the interaction, but requires generating a large pool of models, which is computationally demanding. This work presents FastSRS, an efficient algorithm for generating a large quantity of high quality SRS models. Its key advantage is a set of theoretical bounds that efficiently reduces the size of the search space to enable fast computation. FastSRS produces models on the accuracy vs. sparsity frontier more consistently and efficiently than previous approaches, scales better to larger datasets, and sets a new state of the art for generating Rashomon sets for SRSs.
2025
- NeurIPSRose Gurung†, Ronilo Ragodos†, Chiyu Ma, Tong Wang, and Chaofan ChenAdvances in Neural Information Processing Systems
We present Prototypical Pair Network (ProtoPairNet), a novel interpretable architecture that combines deep learning with case-based reasoning to predict continuous targets. While prototype-based models have primarily addressed image classification with discrete outputs, extending these methods to continuous targets, such as regression, poses significant challenges. Existing architectures which rely heavily on one-to-one comparison with prototypes lack the directional information necessary for continuous predictions. Our method redefines the role of prototypes in such tasks by incorporating prototypical pairs into the reasoning process. Predictions are derived based on the input’s relative dissimilarities to these pairs, leveraging an intuitive geometric interpretation. Our method further reduces the complexity of the reasoning process by relying on the single most relevant pair of prototypes, rather than all prototypes in the model as was done in prior works. Our model is versatile enough to be used in both vision-based regression and continuous control in reinforcement learning. Our experiments demonstrate that ProtoPairNet achieves performance on par with its black-box counterparts across these tasks. Comprehensive analyses confirm the meaningfulness of prototypical pairs and the faithfulness of our model’s interpretations, and extensive user studies highlight our model’s improved interpretability over existing methods.
- IJOCTong Wang, Jingyi Yang, Yunyi Li, and Boxiang WangINFORMS Journal on Computing
We propose a new form of predictive models, Partially Interpretable Estimators (PIE), which jointly train an interpretable model and a black-box model to achieve partial model transparency while maintaining high predictive performance. Our design is motivated by prior research showing that interpretability does not require exposing all model details. Therefore, our objective is to explain the main components of the prediction, withholding complicated calculations that may not be necessary for users. PIE is designed to attribute a prediction to the contribution from individual features via a sparse linear additive model to achieve interpretability while complementing the prediction with a black-box model to boost the predictive performance. As such, the linear additive model captures the primary feature contributions, while the black-box component augments PIE’s predictive power by capturing the “nuances” of feature interactions as a refinement. Moreover, we include a sparsity constraint, allowing users to adjust the model to meet domain-specific needs of interpretability. To optimize predictive performance, we propose a coordinated training algorithm that jointly trains the two components of PIE. Experimental results show that PIE achieves accuracy comparable to state-of-the-art black-box models, with human assessments confirming that its interpretability is nearly equivalent to linear models.
2024
- NeurIPSYiyang Sun†, Tong Wang, and Cynthia RudinAdvances in Neural Information Processing Systems
Sparsity is a central aspect of interpretability in machine learning. Typically, sparsity is measured in terms of the size of a model globally, such as the number of variables it uses. However, this notion of sparsity is not particularly relevant for decision making; someone subjected to a decision does not care about variables that do not contribute to the decision. In this work, we dramatically expand a notion of decision sparsity called the Sparse Explanation Value (SEV) so that its explanations are more meaningful. SEV considers movement along a hypercube towards a reference point. By allowing flexibility in that reference and by considering how distances along the hypercube translate to distances in feature space, we can derive sparser and more meaningful explanations for various types of function classes. We present cluster-based SEV and its variant tree-based SEV, introduce a method that improves credibility of explanations, and propose algorithms that optimize decision sparsity in machine learning models.
- AISTATSYiyang Sun†, Zhi Chen, Vittorio Orlandi, Tong Wang, and Cynthia RudinInternational conference on artificial intelligence and statistics
Winner of Data Mining Best Paper Award, INFORMS, 2023.
Even if a model is not globally sparse, it is possible for decisions made from that model to be accurately and faithfully described by a small number of features. For instance, an application for a large loan might be denied to someone because they have no credit history, which overwhelms any evidence towards their creditworthiness. In this work, we introduce the Sparse Explanation Value (SEV), a new way of measuring sparsity in machine learning models. In the loan denial example above, the SEV is 1 because only one factor is needed to explain why the loan was denied. SEV is a measure of decision sparsity rather than overall model sparsity, and we are able to show that many machine learning models – even if they are not sparse – actually have low decision sparsity, as measured by SEV. SEV is defined using movements over a hypercube, allowing SEV to be defined consistently over various model classes, with movement restrictions reflecting real-world constraints. Our algorithms reduce SEV without sacrificing accuracy, providing sparse and completely faithful explanations, even without globally sparse models.
2023
- JMLRDat Hong†, Tong Wang*, and Stephen BaekJournal of Machine Learning Research
We propose a novel interpretable deep neural network for text classification, called ProtoryNet, based on a new concept of prototype trajectories. Motivated by the prototype theory in modern linguistics, ProtoryNet makes a prediction by finding the most similar prototype for each sentence in a text sequence and feeding an RNN backbone with the proximity of each sentence to the corresponding active prototype. The RNN backbone then captures the temporal pattern of the prototypes, which we refer to as prototype trajectories. Prototype trajectories enable intuitive and fine-grained interpretation of the reasoning process of the RNN model, in resemblance to how humans analyze texts. We also design a prototype pruning procedure to reduce the total number of prototypes used by the model for better interpretability. Experiments on multiple public datasets demonstrate that ProtoryNet achieves higher accuracy than the baseline prototype-based deep neural net and narrows the performance gap when compared to state-of-the-art black-box models. In addition, after prototype pruning, the resulting ProtoryNet models only need less than or around 20 prototypes for all datasets, which significantly benefits interpretability. Furthermore, we report survey results indicating that human users find ProtoryNet more intuitive and easier to understand compared to other prototype-based methods.
2022
- NeurIPSRonilo Ragodos†, Tong Wang*, Qihang Lin, and Xun ZhouAdvances in Neural Information Processing Systems
While deep reinforcement learning has proven to be successful in solving control tasks, the "black-box" nature of an agent has received increasing concerns. We propose a prototype-based post-hoc policy explainer, ProtoX, that explains a black-box agent by prototyping the agent’s behaviors into scenarios, each represented by a prototypical state. When learning prototypes, ProtoX considers both visual similarity and scenario similarity. The latter is unique to the reinforcement learning context since it explains why the same action is taken in visually different states. To teach ProtoX about visual similarity, we pre-train an encoder using contrastive learning via self-supervised learning to recognize states as similar if they occur close together in time and receive the same action from the black-box agent. We then add an isometry layer to allow ProtoX to adapt scenario similarity to the downstream task. ProtoX is trained via imitation learning using behavior cloning, and thus requires no access to the environment or agent. In addition to explanation fidelity, we design different prototype shaping terms in the objective function to encourage better interpretability. We conduct various experiments to test ProtoX. Results show that ProtoX achieved high fidelity to the original black-box agent while providing meaningful and understandable explanations.
- IJOCTong Wang and Cynthia RudinINFORMS Journal on Computing
A key question in causal inference analyses is how to find subgroups with elevated treatment effects. This paper takes a machine learning approach and introduces a generative model, causal rule sets (CRS), for interpretable subgroup discovery. A CRS model uses a small set of short decision rules to capture a subgroup in which the average treatment effect is elevated. We present a Bayesian framework for learning a causal rule set. The Bayesian model consists of a prior that favors simple models for better interpretability as well as avoiding overfitting and a Bayesian logistic regression that captures the likelihood of data, characterizing the relation between outcomes, attributes, and subgroup membership. The Bayesian model has tunable parameters that can characterize subgroups with various sizes, providing users with more flexible choices of models from the treatment-efficient frontier. We find maximum a posteriori models using iterative discrete Monte Carlo steps in the joint solution space of rules sets and parameters. To improve search efficiency, we provide theoretically grounded heuristics and bounding strategies to prune and confine the search space. Experiments show that the search algorithm can efficiently recover true underlying subgroups. We apply CRS on public and real-world data sets from domains in which interpretability is indispensable. We compare CRS with state-of-the-art rule-based subgroup discovery models. Results show that CRS achieves consistently competitive performance on data sets from various domains, represented by high treatment-efficient frontiers.
- SIGKDDDat Hong†, Alberto Maria Segre, and Tong Wang*ACM SIGKDD Conference on Knowledge Discovery and Data Mining
Recurrent neural networks (RNN) are widely used for handling sequence data. However, their black-box nature makes it difficult for users to interpret the decision-making process. We propose a new method to construct deterministic finite automata to explain RNN. In an automaton, states are abstracted from hidden states produced by the RNN, and the transitions represent input symbols. Thus, users can follow the paths of transitions, called patterns, to understand how a prediction is produced. Existing methods for extracting automata partition the hidden state space at the beginning of the extraction, which often leads to solutions that are either inaccurate or too large in size to comprehend. Unlike previous methods, our approach allows the automata states to be formed adaptively during the extraction. Instead of defining patterns on pre-determined clusters, our proposed model, AdaAX, identifies small sets of hidden states determined by patterns with finer granularity in data. Then these small sets are gradually merged to form states, allowing users to trade fidelity for lower complexity. Experiments show that our automata can achieve higher fidelity while being significantly smaller in size than baseline methods on synthetic and complex real datasets.
2021
- JMLRTong Wang and Qihang LinJournal of Machine Learning Research
Finalist, Best Paper Award at INFORMS Workshop on Data Mining & Decision Analytics, 2018.
Interpretable machine learning has become a strong competitor for black-box models. However, the possible loss of the predictive performance for gaining understandability is often inevitable, especially when it needs to satisfy users with diverse backgrounds or high standards for what is considered interpretable. This tension puts practitioners in a dilemma of choosing between high accuracy (black-box models) and interpretability (interpretable models). In this work, we propose a novel framework for building a Hybrid Predictive Model that integrates an interpretable model with any pre-trained black-box model to combine their strengths. The interpretable model substitutes the black-box model on a subset of data where the interpretable model is most competent, gaining transparency at a low cost of the predictive accuracy. We design a principled objective function that considers predictive accuracy, model interpretability, and model transparency (defined as the percentage of data processed by the interpretable substitute.) Under this framework, we propose two hybrid models, one substituting with association rules and the other with linear models, and design customized training algorithms for both models. We test the hybrid models on structured data and text data where interpretable models collaborate with various state-of-the-art black-box models. Results show that hybrid models obtain an efficient trade-off between transparency and predictive performance, characterized by pareto frontiers. Finally, we apply the proposed model on a real-world patients dataset for predicting cardiovascular disease and propose multi-model Pareto frontiers to assist model selection in real applications.
2020
- AISTATSDanqing Pan†, Tong Wang, and Satoshi HaraInternational conference on artificial intelligence and statistics
We present an interpretable companion model for any pre-trained black-box classifiers. The idea is that for any input, a user can decide to either receive a prediction from the black-box model, with high accuracy but no explanations, or employ a companion rule to obtain an interpretable prediction with slightly lower accuracy. The companion model is trained from data and the predictions of the black-box model, with the objective combining area under the transparency–accuracy curve and model complexity. Our model provides flexible choices for practitioners who face the dilemma of choosing between always using interpretable models and always using black-box models for a predictive task, so users can, for any given input, take a step back to resort to an interpretable prediction if they find the predictive performance satisfying, or stick to the black-box model if the rules are unsatisfying. To show the value of companion models, we design a human evaluation on more than a hundred people to investigate the tolerable accuracy loss to gain interpretability for humans.
- ICMLHassan Rafique†, Tong Wang*, Qihang Lin, and Arshia SinghaniInternational Conference on Machine Learning
Runner-up for Best Paper Award at INFORMS Workshop on Data Science, 2019.
We propose a novel type of hybrid model for multi-class classification, which utilizes competing linear models to collaborate with an existing black-box model, promoting transparency in the decision-making process. Our proposed hybrid model, Model-Agnostic Linear Competitors (MALC), brings together the interpretable power of linear models and the good predictive performance of the state-of-the-art black-box models. We formulate the training of a MALC model as a convex optimization problem, optimizing the predictive accuracy and transparency (defined as the percentage of data captured by the linear models) in the objective function. Experiments show that MALC offers more model flexibility for users to balance transparency and accuracy, in contrast to the currently available choice of either a pure black-box model or a pure interpretable model. The human evaluation also shows that more users are likely to choose MALC for this model flexibility compared with interpretable models and black-box models.
2019
- ICMLTong WangInternational Conference on Machine Learning
This work addresses the situation where a black-box model with good predictive performance is chosen over its interpretable competitors, and we show interpretability is still achievable in this case. Our solution is to find an interpretable substitute on a subset of data where the black-box model is overkill or nearly overkill while leaving the rest to the black-box. This transparency is obtained at minimal cost or no cost of the predictive performance. Under this framework, we develop a Hybrid Rule Sets (HyRS) model that uses decision rules to capture the subspace of data where the rules are as accurate or almost as accurate as the black-box provided. To train a HyRS, we devise an efficient search algorithm that iteratively finds the optimal model and exploits theoretically grounded strategies to reduce computation. Our framework is agnostic to the black-box during training. Experiments on structured and text data show that HyRS obtains an effective trade-off between transparency and interpretability.
2018
- NeurIPSTong WangAdvances in neural information processing systems
We present the Multi-value Rule Set (MRS) for interpretable classification with feature efficient presentations. Compared to rule sets built from single-value rules, MRS adopts a more generalized form of association rules that allows multiple values in a condition. Rules of this form are more concise than classical single-value rules in capturing and describing patterns in data. Our formulation also pursues a higher efficiency of feature utilization, which reduces possible cost in data collection and storage. We propose a Bayesian framework for formulating an MRS model and develop an efficient inference method for learning a maximum a posteriori, incorporating theoretically grounded bounds to iteratively reduce the search space and improve the search efficiency. Experiments on synthetic and real-world data demonstrate that MRS models have significantly smaller complexity and fewer features than baseline models while being competitive in predictive accuracy.
2017
- JMLRTong Wang, Cynthia Rudin, Finale Doshi-Velez, Yimin Liu, Erica Klampfl, and Perry MacNeilleJournal of Machine Learning Research
We present a machine learning algorithm for building classifiers that are comprised of a small number of short rules. These are restricted disjunctive normal form models. An example of a classifier of this form is as follows: If X satisfies (condition A AND condition B) OR (condition C) OR ..., then Y=1. Models of this form have the advantage of being interpretable to human experts since they produce a set of rules that concisely describe a specific class. We present two probabilistic models with prior parameters that the user can set to encourage the model to have a desired size and shape, to conform with a domain-specific definition of interpretability. We provide a scalable MAP inference approach and develop theoretical bounds to reduce computation by iteratively pruning the search space. We apply our method (Bayesian Rule Sets, BRS) to characterize and predict user behavior with respect to in-vehicle context-aware personalized recommender systems. Our method has a major advantage over classical associative classification methods and decision trees in that it does not greedily grow the model.
Machine Learning Applications
2025
- SIGSPATIALConPro-GAIL: Interpretable Policy Learning via Conceptual Prototyping for Human Spatiotemporal Decision UnderstandingRonilo Ragodos†, Xun Zhou, Tong Wang, Yajun Pan, and Jun LuoACM SIGSPATIAL International Conference on Advances in Geographic Information Systems
The problem of human spatiotemporal (ST) decision understanding, which consists of extracting faithful and interpretable decision strategies from human agents’ behavioral records in space and time, is important for many applications, such as improving taxi drivers’ route planning and efficiency. It is challenging because ST data are not as readily interpretable as images or text data, which leads to difficulties in constructing data-driven explanations. Existing research on this topic defines the problem as a Markov Decision Process (MDP) and uses imitation learning to extract a policy approximating the underlying human policy for post-hoc interpretation. However, such methods cannot provide direct interpretation through model training and may result in incomprehensible interpretations when using ST data. We address these limitations by designing ConPro-GAIL, a prototype-based interpretable GAIL model for intrinsically interpretable ST policy extraction. ConPro-GAIL learns and represents the optimal policy in terms of prototypical sets of concepts that correspond to general scenarios in the MDP. It explains a decision associated with an input state via inductive generalization from what occurred in the state’s most similar prototypes to the input state itself. Experiments and case studies on two taxi trajectory datasets show that ConPro-GAIL achieves better policy faithfulness than its black-box competitors and better interpretability than post-hoc explainers.
2023
- EJORXinwei Chen†, Tong Wang, Barrett W Thomas, and Marlin W UlmerEuropean Journal of Operational Research
The demand for same-day delivery (SDD) has increased rapidly in the last few years and has particularly boomed during the COVID-19 pandemic. The fast growth is not without its challenge. In 2016, due to low concentrations of memberships and far distance from the depot, certain minority neighborhoods were excluded from receiving Amazon’s SDD service, raising concerns about fairness. In this paper, we study the problem of offering fair SDD-service to customers. The service area is partitioned into different regions. Over the course of a day, customers request for SDD service, and the timing of requests and delivery locations are not known in advance. The dispatcher dynamically assigns vehicles to make deliveries to accepted customers before their delivery deadline. In addition to the overall service rate (utility), we maximize the minimal regional service rate across all regions (fairness). We model the problem as a multi-objective Markov decision process and develop a deep Q-learning solution approach. We introduce a novel transformation of learning from rates to actual services, which creates a stable and efficient learning process. Computational results demonstrate the effectiveness of our approach in alleviating unfairness both spatially and temporally in different customer geographies. We also show this effectiveness is valid with different depot locations, providing businesses with an opportunity to achieve better fairness from any location. Further, we consider the impact of ignoring fairness in service, and results show that our policies eventually outperform the utility-driven baseline when customers have a high expectation on service level.
2022
- DSSChaofan Chen, Kangcheng Lin, Cynthia Rudin, Yaron Shaposhnik, Sijia Wang, and Tong WangDecision Support Systems
This project is the winner of FICO Recognition Award for the FICO xML Challenge, for building an interpretable model that beats black-box models. See the blog on FICO website about this entry.
Lending decisions are usually made with proprietary models that provide minimally acceptable explanations to users. In a future world without such secrecy, what decision support tools would one want to use for justified lending decisions? This question is timely, since the economy has dramatically shifted due to a pandemic, and a massive number of new loans will be necessary in the short term. We propose a framework for such decisions, including a globally interpretable machine learning model, an interactive visualization of it, and several types of summaries and explanations for any given decision. The machine learning model is a two-layer additive risk model, which resembles a two-layer neural network, but is decomposable into subscales. In this model, each node in the first (hidden) layer represents a meaningful subscale model, and all of the nonlinearities are transparent. Our online visualization tool allows exploration of this model, showing precisely how it came to its conclusion. We provide three types of explanations that are simpler than, but consistent with, the global model: case-based reasoning explanations that use neighboring past cases, a set of features that were the most important for the model’s prediction, and summary-explanations that provide a customized sparse explanation for any particular lending decision made by the model. Our framework earned the FICO recognition award for the Explainable Machine Learning Challenge, which was the first public challenge in the domain of explainable machine learning.
- IJOCRonilo Ragodos† and Tong WangINFORMS Journal on Computing
In this study, we present an interpretable model, disjunctive rule list (DisRL) for regression. This research is motivated by the increasing need for model interpretability, especially in high-stakes decisions such as medicine, where decisions are made on or related to humans. DisRL is a generalized form of rule lists. A DisRL model consists of a list of disjunctive rules embedded in an if-else logic structure that stratifies the data space. Compared with traditional decision trees and other rule list models in the literature that stratify the feature space with single itemsets (an itemset is a conjunction of conditions), each disjunctive rule in DisRL uses a set of itemsets to collectively cover a subregion in the feature space. In addition, a DisRL model is constructed under a global objective that balances the predictive performance and model complexity. To train a DisRL model, we devise a hierarchical stochastic local search algorithm that exploits the properties of DisRL’s unique structure to improve search efficiency. The algorithm adopts the main structure of simulated annealing and customizes the proposing strategy for faster convergence. Meanwhile, the algorithm uses a prefix bound to locate a subset of the search area, effectively pruning the search space at each iteration. An ablation study shows the effectiveness of this strategy in pruning the search space. Experiments on public benchmark datasets demonstrate that DisRL outperforms baseline interpretable models, including decision trees and other rule-based regressors.
- Sci. Rep.Ronilo Ragodos†, Tong Wang*, George Wehby, Seth Weinberg, Deborah Dawson, Mary Mzrazita, and 2 more authorsScientific Reports
Children with orofacial clefting (OFC) present with a wide range of dental anomalies. Identifying these anomalies is vital to understand their etiology and to discern the complex phenotypic spectrum of OFC. Such anomalies are currently identified using intra-oral exams by dentists, a costly and time-consuming process. We claim that automating the process of anomaly detection using deep neural networks (DNNs) could increase efficiency and provide reliable anomaly detection while potentially increasing the speed of research discovery. This study characterizes the use of DNNs to identify dental anomalies by training a DNN model using intraoral photographs from the largest international cohort to date of children with nonsyndromic OFC and controls (OFC1). In this project, the intraoral images were submitted to a Convolutional Neural Network model to perform multi-label multi-class classification of 10 dental anomalies. The network predicts whether an individual exhibits any of the 10 anomalies and can do so significantly faster than a human rater can. For all but three anomalies, F1 scores suggest that our model performs competitively at anomaly detection when compared to a dentist with 8 years of clinical experience. In addition, we use saliency maps to provide a post-hoc interpretation for our model’s predictions. This enables dentists to examine and verify our model’s predictions.
2019
- SociusAnna Filippova, Connor Gilroy, Ridhi Kashyap, Antje Kirchner, Allison Morgan, Kivan Polimis, and 2 more authorsSocius
Survey data sets are often wider than they are long. This high ratio of variables to observations raises concerns about overfitting during prediction, making informed variable selection important. Recent applications in computer science have sought to incorporate human knowledge into machine-learning methods to address these problems. The authors implement such a “human-in-the-loop” approach in the Fragile Families Challenge. The authors use surveys to elicit knowledge from experts and laypeople about the importance of different variables to different outcomes. This strategy offers the option to subset the data before prediction or to incorporate human knowledge as scores in prediction models, or both together. The authors find that human intervention is not obviously helpful. Human-informed subsetting reduces predictive performance, and considered alone, approaches incorporating scores perform marginally worse than approaches that do not. However, incorporating human knowledge may still improve predictive performance, and future research should consider new ways of doing so.
2018
- Am. J. SurgVeerajalandhar Allareddy, Tong Wang, Sankeerth Rampa, Jennifer Caplin, Romesh Nalliah, Aditya Badheka, and 1 more authorAmerican Journal of Surgery
2015
- BigDataTong Wang, Cynthia Rudin, Daniel Wagner, and Rich SevieriBig Data
Second place winner of the INFORMS “Doing Good with Good OR” Award, 2015.
One of the most challenging problems facing crime analysts is that of identifying crime series, which are sets of crimes committed by the same individual or group. Detecting crime series can be an important step in predictive policing, as knowledge of a pattern can be of paramount importance toward finding the offenders or stopping the pattern. Currently, crime analysts detect crime series manually; our goal is to assist them by providing automated tools for discovering crime series from within a database of crimes. Our approach relies on a key hypothesis that each crime series possesses at least one core of crimes that are very similar to each other, which can be used to characterize the modus operandi (M.O.) of the criminal. Based on this assumption, as long as we find all of the cores in the database, we have found a piece of each crime series. We propose a subspace clustering method, where the subspace is the M.O. of the series. The method has three steps: We first construct a similarity graph to link crimes that are generally similar, second we find cores of crime using an integer linear programming approach, and third we construct the rest of the crime series by merging cores to form the full crime series. To judge whether a set of crimes is indeed a core, we consider both pattern-general similarity, which can be learned from past crime series, and pattern-specific similarity, which is specific to the M.O. of the series and cannot be learned. Our method can be used for general pattern detection beyond crime series detection, as cores exist for patterns in many domains.
2013
- ECML-PKDDTong Wang, Cynthia Rudin, Daniel Wagner, and Rich SevieriEuropean Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases
Ideas from this work were adopted by the New York Police Department in their algorithm Patternizr, which has been running live in New York City since 2016 (see this paper for details).
- Predictive Policing: Using Machine Learning to Detect Patterns of Crime, WIRED, 2013
- Cambridge police look at math to solve crimes, The Boston Globe, 2013
- Crime analysis, Wikipedia
Our goal is to automatically detect patterns of crime. Among a large set of crimes that happen every year in a major city, it is challenging, time-consuming, and labor-intensive for crime analysts to determine which ones may have been committed by the same individual(s). If automated, data-driven tools for crime pattern detection are made available to assist analysts, these tools could help police to better understand patterns of crime, leading to more precise attribution of past crimes, and the apprehension of suspects. To do this, we propose a pattern detection algorithm called Series Finder, that grows a pattern of discovered crimes from within a database, starting from a "seed" of a few crimes. Series Finder incorporates both the common characteristics of all patterns and the unique aspects of each specific pattern, and has had promising results on a decade’s worth of crime pattern data collected by the Crime Analysis Unit of the Cambridge Police Department.