Freddy Lecue

TMLR Journal 2025 Journal Article

Interpretable LLM-based Table Question Answering

• Giang Nguyen
• Ivan Brugere
• Shubham Sharma
• Sanjay Kariyappa
• Anh Totti Nguyen
• Freddy Lecue

Interpretability in Table Question Answering (Table QA) is critical, especially in high-stakes domains like finance and healthcare. While recent Table QA approaches based on Large Language Models (LLMs) achieve high accuracy, they often produce ambiguous explanations of how answers are derived. We propose Plan-of-SQLs (POS), a new Table QA method that makes the model's decision-making process interpretable. POS decomposes a question into a sequence of atomic steps, each directly translated into an executable SQL command on the table, thereby ensuring that every intermediate result is transparent. Through extensive experiments, we show that: First, POS generates the highest-quality explanations among compared methods, which markedly improves the users' ability to simulate and verify the model’s decisions. Second, when evaluated on standard Table QA benchmarks (TabFact, WikiTQ, and FeTaQA), POS achieves QA accuracy that is competitive to existing methods, while also offering greater efficiency—requiring significantly fewer LLM calls and table database queries (up to 25x fewer)—and more robust performance on large-sized tables. Finally, we observe high agreement (up to 90.59% in forward simulation) between LLMs and human users when making decisions based on the same explanations, suggesting that LLMs could serve as an effective proxy for humans in evaluating Table QA explanations.

NeurIPS Conference 2025 Conference Paper

The Unseen Threat: Residual Knowledge in Machine Unlearning under Perturbed Samples

• Hsiang Hsu
• Pradeep Niroula
• Zichang He
• Ivan Brugere
• Freddy Lecue
• Richard Chen

Machine unlearning offers a practical alternative to avoid full model re-training by approximately removing the influence of specific user data. While existing methods certify unlearning via statistical indistinguishability from re-trained models, these guarantees do not naturally extend to model outputs when inputs are adversarially perturbed. In particular, slight perturbations of forget samples may still be correctly recognized by the unlearned model---even when a re-trained model fails to do so---revealing a novel privacy risk: information about the forget samples may persist in their local neighborhood. In this work, we formalize this vulnerability as residual knowledge and show that it is inevitable in high-dimensional settings. To mitigate this risk, we propose a fine-tuning strategy, named RURK, that penalizes the model’s ability to re-recognize perturbed forget samples. Experiments on vision benchmarks with deep neural networks demonstrate that residual knowledge is prevalent across existing unlearning methods and that our approach effectively prevents residual knowledge.

IJCAI Conference 2024 Conference Paper

Are Logistic Models Really Interpretable?

• Danial Dervovic
• Freddy Lecue
• Nicolas Marchesotti
• Daniele Magazzeni

The demand for open and trustworthy AI models points towards widespread publishing of model weights. Consumers of these model weights must be able to act accordingly with the information provided. That said, one of the simplest AI classification models, Logistic Regression (LR), has an unwieldy interpretation of its model weights, with greater difficulties when extending LR to generalised additive models. In this work, we show via a User Study that skilled participants are unable to reliably reproduce the action of small LR models given the trained parameters. As an antidote to this, we define Linearised Additive Models (LAMs), an optimal piecewise linear approximation that augments any trained additive model equipped with a sigmoid link function, requiring no retraining. We argue that LAMs are more interpretable than logistic models -- survey participants are shown to solve model reasoning tasks with LAMs much more accurately than with LR given the same information. Furthermore, we show that LAMs do not suffer from large performance penalties in terms of ROC-AUC and calibration with respect to their logistic counterparts on a broad suite of public financial modelling data.

NeurIPS Conference 2024 Conference Paper

Fair Wasserstein Coresets

• Zikai Xiong
• Niccolò Dalmasso
• Shubham Sharma
• Freddy Lecue
• Daniele Magazzeni
• Vamsi K. Potluru
• Tucker Balch
• Manuela Veloso

Data distillation and coresets have emerged as popular approaches to generate a smaller representative set of samples for downstream learning tasks to handle large-scale datasets. At the same time, machine learning is being increasingly applied to decision-making processes at a societal level, making it imperative for modelers to address inherent biases towards subgroups present in the data. While current approaches focus on creating fair synthetic representative samples by optimizing local properties relative to the original samples, their impact on downstream learning processes has yet to be explored. In this work, we present fair Wasserstein coresets ($\texttt{FWC}$), a novel coreset approach which generates fair synthetic representative samples along with sample-level weights to be used in downstream learning tasks. $\texttt{FWC}$ uses an efficient majority minimization algorithm to minimize the Wasserstein distance between the original dataset and the weighted synthetic samples while enforcing demographic parity. We show that an unconstrained version of $\texttt{FWC}$ is equivalent to Lloyd's algorithm for k-medians and k-means clustering. Experiments conducted on both synthetic and real datasets show that $\texttt{FWC}$: (i) achieves a competitive fairness-performance tradeoff in downstream models compared to existing approaches, (ii) improves downstream fairness when added to the existing training data and (iii) can be used to reduce biases in predictions from large language models (GPT-3. 5 and GPT-4).

NeurIPS Conference 2024 Conference Paper

RashomonGB: Analyzing the Rashomon Effect and Mitigating Predictive Multiplicity in Gradient Boosting

• Hsiang Hsu
• Ivan Brugere
• Shubham Sharma
• Freddy Lecue
• Chun-Fu Chen

The Rashomon effect is a mixed blessing in responsible machine learning. It enhances the prospects of finding models that perform well in accuracy while adhering to ethical standards, such as fairness or interpretability. Conversely, it poses a risk to the credibility of machine decisions through predictive multiplicity. While recent studies have explored the Rashomon effect across various machine learning algorithms, its impact on gradient boosting---an algorithm widely applied to tabular datasets---remains unclear. This paper addresses this gap by systematically analyzing the Rashomon effect and predictive multiplicity in gradient boosting algorithms. We provide rigorous theoretical derivations to examine the Rashomon effect in the context of gradient boosting and offer an information-theoretic characterization of the Rashomon set. Additionally, we introduce a novel inference technique called RashomonGB to efficiently inspect the Rashomon effect in practice. On more than 20 datasets, our empirical results show that RashomonGB outperforms existing baselines in terms of improving the estimation of predictive multiplicity metrics and model selection with group fairness constraints. Lastly, we propose a framework to mitigate predictive multiplicity in gradient boosting and empirically demonstrate its effectiveness.

NeurIPS Conference 2024 Conference Paper

Sequential Harmful Shift Detection Without Labels

• Salim I. Amoukou
• Tom Bewley
• Saumitra Mishra
• Freddy Lecue
• Daniele Magazzeni
• Manuela Veloso

We introduce a novel approach for detecting distribution shifts that negatively impact the performance of machine learning models in continuous production environments, which requires no access to ground truth data labels. It builds upon the work of Podkopaev and Ramdas [2022], who address scenarios where labels are available for tracking model errors over time. Our solution extends this framework to work in the absence of labels, by employing a proxy for the true error. This proxy is derived using the predictions of a trained error estimator. Experiments show that our method has high power and false alarm control under various distribution shifts, including covariate and label shifts and natural shifts over geography and time.

NeurIPS Conference 2023 Conference Paper

Comparing Apples to Oranges: Learning Similarity Functions for Data Produced by Different Distributions

• Leonidas Tsepenekas
• Ivan Brugere
• Freddy Lecue
• Daniele Magazzeni

Similarity functions measure how comparable pairs of elements are, and play a key role in a wide variety of applications, e. g. , notions of Individual Fairness abiding by the seminal paradigm of Dwork et al. , as well as Clustering problems. However, access to an accurate similarity function should not always be considered guaranteed, and this point was even raised by Dwork et al. For instance, it is reasonable to assume that when the elements to be compared are produced by different distributions, or in other words belong to different ``demographic'' groups, knowledge of their true similarity might be very difficult to obtain. In this work, we present an efficient sampling framework that learns these across-groups similarity functions, using only a limited amount of experts' feedback. We show analytical results with rigorous theoretical bounds, and empirically validate our algorithms via a large suite of experiments.

AAAI Conference 2022 Short Paper

A Simplified Benchmark for Ambiguous Explanations of Knowledge Graph Link Prediction Using Relational Graph Convolutional Networks (Student Abstract)

• Nicholas Halliwell
• Fabien Gandon
• Freddy Lecue

Relational Graph Convolutional Networks (RGCNs) are commonly used on Knowledge Graphs (KGs) to perform black box link prediction. Several algorithms have been proposed to explain their predictions. Evaluating performance of explanation methods for link prediction is difficult without ground truth explanations. Furthermore, there can be multiple explanations for a given prediction in a KG. No dataset exists where observations have multiple ground truth explanations to compare against. Additionally, no standard scoring metrics exist to compare predicted explanations against multiple ground truth explanations. We propose and evaluate a method, including a dataset, to benchmark explanation methods on the task of explainable link prediction using RGCNs.

AAAI Conference 2022 Conference Paper

FisheyeHDK: Hyperbolic Deformable Kernel Learning for Ultra-Wide Field-of-View Image Recognition

• Ola Ahmad
• Freddy Lecue

Conventional convolution neural networks (CNNs) trained on narrow Field-of-View (FoV) images are the state-of-the art approaches for object recognition tasks. Some methods proposed the adaptation of CNNs to ultra-wide FoV images by learning deformable kernels. However, they are limited by the Euclidean geometry and their accuracy degrades under strong distortions caused by fisheye projections. In this work, we demonstrate that learning the shape of convolution kernels in non-Euclidean spaces is better than existing deformable kernel methods. In particular, we propose a new approach that learns deformable kernel parameters (positions) in hyperbolic space. FisheyeHDK is a hybrid CNN architecture combining hyperbolic and Euclidean convolution layers for positions and features learning. First, we provide intuition of hyperbolic space for wide FoV images. Using synthetic distortion profiles, we demonstrate the effectiveness of our approach. We select two datasets - Cityscapes and BDD100K 2020 - of perspective images which we transform to fisheye equivalents at different scaling factors (analogue to focal lengths). Finally, we provide an experiment on data collected by a real fisheye camera. Validations and experiments show that our approach improves existing deformable kernel methods for CNN adaptation on fisheye images.

AAMAS Conference 2022 Conference Paper

Interpretable Preference-based Reinforcement Learning with Tree-Structured Reward Functions

• Tom Bewley
• Freddy Lecue

The potential of reinforcement learning (RL) to deliver aligned and performant agents is partially bottlenecked by the reward engineering problem. One alternative to heuristic trial-and-error is preference-based RL (PbRL), where a reward function is inferred from sparse human feedback. However, prior PbRL methods lack interpretability of the learned reward structure, which hampers the ability to assess robustness and alignment. We propose an online, active preference learning algorithm that constructs reward functions with the intrinsically interpretable, compositional structure of a tree. Using both synthetic and human-provided feedback, we demonstrate sample-efficient learning of tree-structured reward functions in several environments, then harness the enhanced interpretability to explore and debug for alignment.

AAAI Conference 2019 Conference Paper

Human-in-the-Loop Feature Selection

• Alvaro H. C. Correia
• Freddy Lecue

Feature selection is a crucial step in the conception of Machine Learning models, which is often performed via datadriven approaches that overlook the possibility of tapping into the human decision-making of the model’s designers and users. We present a human-in-the-loop framework that interacts with domain experts by collecting their feedback regarding the variables (of few samples) they evaluate as the most relevant for the task at hand. Such information can be modeled via Reinforcement Learning to derive a per-example feature selection method that tries to minimize the model’s loss function by focusing on the most pertinent variables from a human perspective. We report results on a proof-of-concept image classification dataset and on a real-world risk classification task in which the model successfully incorporated feedback from experts to improve its accuracy.

KR Conference 2018 Conference Paper

Transfer Learning Explanation with Ontologies

• Jiaoyan Chen
• Freddy Lecue
• Jeff Z Pan
• Ian Horrocks
• Huajun Chen

IJCAI Conference 2017 Conference Paper

Learning from Ontology Streams with Semantic Concept Drift

• Jiaoyan Chen
• Freddy Lecue
• Jeff Z. Pan
• Huajun Chen

Data stream learning has been largely studied for extracting knowledge structures from continuous and rapid data records. In the semantic Web, data is interpreted in ontologies and its ordered sequence is represented as an ontology stream. Our work exploits the semantics of such streams to tackle the problem of concept drift i. e. , unexpected changes in data distribution, causing most of models to be less accurate as time passes. To this end we revisited (i) semantic inference in the context of supervised stream learning, and (ii) models with semantic embeddings. The experiments show accurate prediction with data from Dublin and Beijing.

AAAI Conference 2015 Conference Paper

Consistent Knowledge Discovery from Evolving Ontologies

• Freddy Lecue
• Jeff Pan

Deductive reasoning and inductive learning are the most common approaches for deriving knowledge. In real world applications when data is dynamic and incomplete, especially those exposed by sensors, reasoning is limited by dynamics of data while learning is biased by data incompleteness. Therefore discovering consistent knowledge from incomplete and dynamic data is a challenging open problem. In our approach the semantics of data is captured through ontologies to empower learning (mining) with (Description Logics) reasoning. Consistent knowledge discovery is achieved by applying generic, significative, representative association semantic rules. The experiments have shown scalable, accurate and consistent knowledge discovery with data from Dublin.

AAAI Conference 2015 Conference Paper

Minimizing User Involvement for Accurate Ontology Matching Problems

• Anika Schumann
• Freddy Lecue

Many various types of sensors coming from different complex devices collect data from a city. Their underlying data representation follows specific manufacturer specifications that have possibly incomplete descriptions (in ontology) alignments. This paper addresses the problem of determining accurate and complete matching of ontologies given some common descriptions and their pre-determined high level alignments. In this context the problem of ontology matching consists of automatically determining all matching given the latter alignments, and manually verifying the matching results. Especially for applications where it is crucial that ontologies are matched correctly the latter can turn into a very time-consuming task for the user. This paper tackles this challenge and addresses the problem of computing the minimum number of user inputs needed to verify all matchings. We show how to represent this problem as a reasoning problem over a bipartite graph and how to encode it over pseudo Boolean constraints. Experiments show that our approach can be successfully applied to real-world data sets.

IJCAI Conference 2015 Conference Paper

Scalable Maintenance of Knowledge Discovery in an Ontology Stream

• Freddy Lecue

In dynamic settings where data is exposed by streams, knowledge discovery aims at learning associations of data across streams. In the semantic Web, streams expose their meaning through evolutive versions of ontologies. Such settings pose challenges of scalability for discovering (a posteriori) knowledge. In our work, the semantics, identifying knowledge similarity and rarity in streams, together with incremental, approximate maintenance, control scalability while preserving accuracy of streams associations (as semantic rules) discovery.

AAAI Conference 2014 Conference Paper

Towards Scalable Exploration of Diagnoses in an Ontology Stream

• Freddy Lecue

Diagnosis, or the process of identifying the nature and cause of an anomaly in an ontology, has been largely studied by the Semantic Web community. In the context of ontology stream, diagnosis results are not captured by a unique fixed ontology but numerous time-evolving ontologies. Thus any anomaly can be diagnosed by a large number of different explanations depending on the version and evolution of the ontology. We address the problems of identifying, representing, exploiting and exploring the evolution of diagnoses representations. Our approach consists in a graph-based representation, which aims at (i) efficiently organizing and linking time-evolving diagnoses and (ii) being used for scalable exploration. The experiments have shown scalable diagnoses exploration in the context of real and live data from Dublin City.

AAAI Conference 2012 Conference Paper

Diagnosing Changes in An Ontology Stream: A DL Reasoning Approach

• Freddy Lecue

Recently, ontology stream reasoning has been introduced as a multidisciplinary approach, merging synergies from Artificial Intelligence, Database and World-Wide-Web to reason on semantics-augmented data streams, thus a way to answering questions on real time events. However existing approaches do not consider stream change diagnosis i. e. , identification of the nature and cause of changes, where explaining the logical connection of knowledge and inferring insight on timechanging events are the main challenges. We exploit the Description Logics (DL)-based semantics of streams to tackle these challenges. Based on an analysis of stream behavior through change and inconsistency over DL axioms, we tackled change diagnosis by determining and constructing a comprehensive view on potential causes of inconsistencies. We report a large-scale evaluation of our approach in the context of live stream data from Dublin City Council.

AAAI Conference 2011 Conference Paper

Personalizing Your Web Services with Constructive DL Reasoning Join

• Freddy Lecue

Nowadays web users have clearly expressed their wishes to receive and interact with personalized services directly. However, existing approaches, largely syntactic content-based, fail to provide robust, accurate and useful personalized services to its users. Towards such an issue, the semantic web provides technologies to annotate and match services’ descriptions with users’ features, interests and preferences, thus allowing for more efficient access to services and more generally information. The aim of our work, part of service personalization, is on automated instantiation of services which is crucial for advanced usability i. e. , how to prepare and present services ready to be executed while limiting useless interactions with users? We introduce the constructive Description Logics reasoning join and couple it with concept abduction to i) identify useful parts of users profiles that satisfy services requirements and ii) compute the description required by a service to be executed but not provided by users profiles.

AAAI Conference 2007 Conference Paper

Making the Difference in Semantic Web Service Composition

• Freddy Lecue

Automation of Web service composition is one of the most interesting challenges facing the Semantic Web today. In this paper we propose a mean of performing automated Web service composition by exploiting semantic matchmaking between Web service parameters (i. e. , outputs and inputs) to enable their connection and interaction. The key idea is that the matchmaking enables, at run time, finding semantic compatibilities among independently defined Web service descriptions. To this end, our approach extends existing methods in order to explain misconnections between Web services. From this we generate Web service compositions that realize the goal, satisfying and optimizing the semantic connections between Web services. Moreover a process of relaxing the hard constraints is introduced in case the composition process failed. Our system is implemented and interacting with Web services dedicated on a Telecom scenario. The preliminary evaluation results showed high efficiency and effectiveness of the proposed approach.