Author name cluster

Freddy Lécué

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

12 papers

2 author rows

ICLR Conference 2025 Conference Paper

Interpreting Language Reward Models via Contrastive Explanations

Junqi Jiang
Tom Bewley
Saumitra Mishra
Freddy Lécué
Manuela Veloso

Reward models (RMs) are a crucial component in the alignment of large language models’ (LLMs) outputs with human values. RMs approximate human preferences over possible LLM responses to the same prompt by predicting and comparing reward scores. However, as they are typically modified versions of LLMs with scalar output heads, RMs are large black boxes whose predictions are not explainable. More transparent RMs would enable improved trust in the alignment of LLMs. In this work, we propose to use contrastive explanations to explain any binary response comparison made by an RM. Specifically, we generate a diverse set of new comparisons similar to the original one to characterise the RM’s local behaviour. The perturbed responses forming the new comparisons are generated to explicitly modify manually specified high-level evaluation attributes, on which analyses of RM behaviour are grounded. In quantitative experiments, we validate the effectiveness of our method for finding high-quality contrastive explanations. We then showcase the qualitative usefulness of our method for investigating global sensitivity of RMs to each evaluation attribute, and demonstrate how representative examples can be automatically extracted to explain and compare behaviours of different RMs. We see our method as a flexible framework for RM explanation, providing a basis for more interpretable and trustworthy LLM alignment.

Details

ICML Conference 2025 Conference Paper

Quantifying Prediction Consistency Under Fine-tuning Multiplicity in Tabular LLMs

Faisal Hamman
Pasan Dissanayake
Saumitra Mishra
Freddy Lécué
Sanghamitra Dutta

Fine-tuning LLMs on tabular classification tasks can lead to the phenomenon of fine-tuning multiplicity where equally well-performing models make conflicting predictions on the same input. Fine-tuning multiplicity can arise due to variations in the training process, e. g. , seed, weight initialization, minor changes to training data, etc. , raising concerns about the reliability of Tabular LLMs in high-stakes applications such as finance, hiring, education, healthcare. Our work formalizes this unique challenge of fine-tuning multiplicity in Tabular LLMs and proposes a novel measure to quantify the consistency of individual predictions without expensive model retraining. Our measure quantifies a prediction’s consistency by analyzing (sampling) the model’s local behavior around that input in the embedding space. Interestingly, we show that sampling in the local neighborhood can be leveraged to provide probabilistic guarantees on prediction consistency under a broad class of fine-tuned models, i. e. , inputs with sufficiently high local stability (as defined by our measure) also remain consistent across several fine-tuned models with high probability. We perform experiments on multiple real-world datasets to show that our local stability measure preemptively captures consistency under actual multiplicity across several fine-tuned models, outperforming competing measures.

Details

AAAI Conference 2024 Conference Paper

Knowledge-Aware Neuron Interpretation for Scene Classification

Yong Guan
Freddy Lécué
Jiaoyan Chen
Ru Li
Jeff Z. Pan

Although neural models have achieved remarkable performance, they still encounter doubts due to the intransparency. To this end, model prediction explanation is attracting more and more attentions. However, current methods rarely incorporate external knowledge and still suffer from three limitations: (1) Neglecting concept completeness. Merely selecting concepts may not sufficient for prediction. (2) Lacking concept fusion. Failure to merge semantically-equivalent concepts. (3) Difficult in manipulating model behavior. Lack of verification for explanation on original model. To address these issues, we propose a novel knowledge-aware neuron interpretation framework to explain model predictions for image scene classification. Specifically, for concept completeness, we present core concepts of a scene based on knowledge graph, ConceptNet, to gauge the completeness of concepts. Our method, incorporating complete concepts, effectively provides better prediction explanations compared to baselines. Furthermore, for concept fusion, we introduce a knowledge graph-based method known as Concept Filtering, which produces over 23% point gain on neuron behaviors for neuron interpretation. At last, we propose Model Manipulation, which aims to study whether the core concepts based on ConceptNet could be employed to manipulate model behavior. The results show that core concepts can effectively improve the performance of original model by over 26%.

PDF Details DOI

ICML Conference 2024 Conference Paper

Progressive Inference: Explaining Decoder-Only Sequence Classification Models Using Intermediate Predictions

Sanjay Kariyappa
Freddy Lécué
Saumitra Mishra
Christopher Pond
Daniele Magazzeni
Manuela Veloso

This paper proposes Progressive inference–a framework to explain the predictions of decoder-only transformer models trained to perform sequence classification tasks. Our work is based on the insight that the classification head of a decoder-only model can be used to make intermediate predictions by evaluating them at different points in the input sequence. Due to the masked attention mechanism used in decoder-only models, these intermediate predictions only depend on the tokens seen before the inference point, allowing us to obtain the model’s prediction on a masked input sub-sequence, with negligible computational overheads. We develop two methods to provide sub-sequence level attributions using this core insight. First, we propose Single Pass-Progressive Inference (SP-PI) to compute attributions by simply taking the difference between intermediate predictions. Second, we exploit a connection with Kernel SHAP to develop Multi Pass-Progressive Inference (MP-PI); this uses intermediate predictions from multiple masked versions of the input to compute higher-quality attributions that approximate SHAP values. We perform studies on several text classification datasets to demonstrate that our proposal provides better explanations compared to prior work, both in the single-pass and multi-pass settings.

Details

AAAI Conference 2024 Conference Paper

SHAP@k: Efficient and Probably Approximately Correct (PAC) Identification of Top-K Features

Sanjay Kariyappa
Leonidas Tsepenekas
Freddy Lécué
Daniele Magazzeni

The SHAP framework provides a principled method to explain the predictions of a model by computing feature importance. Motivated by applications in finance, we introduce the Top-k Identification Problem (TkIP) (and its ordered variant TkIP- O), where the objective is to identify the subset (or ordered subset for TkIP-O) of k features corresponding to the highest SHAP values with PAC guarantees. While any sampling-based method that estimates SHAP values (such as KernelSHAP and SamplingSHAP) can be trivially adapted to solve TkIP, doing so is highly sample inefficient. Instead, we leverage the connection between SHAP values and multi-armed bandits (MAB) to show that both TkIP and TkIP-O can be reduced to variants of problems in MAB literature. This reduction allows us to use insights from the MAB literature to develop sample-efficient variants of KernelSHAP and SamplingSHAP. We propose KernelSHAP@k and SamplingSHAP@k for solving TkIP; along with KernelSHAP-O and SamplingSHAP-O to solve the ordering problem in TkIP-O. We perform extensive experiments using several credit-related datasets to show that our methods offer significant improvements of up to 40× in sample efficiency and 39× in runtime.

PDF Details DOI

KR Conference 2020 Conference Paper

Ontology-guided Semantic Composition for Zero-shot Learning

Jiaoyan Chen
Freddy Lécué
Yuxia Geng
Jeff Z. Pan
Huajun Chen

Zero-shot learning (ZSL) is a popular research problem that aims at predicting for those classes that have never appeared in the training stage by utilizing the inter-class relationship with some side information. In this study, we propose to model the compositional and expressive semantics of class labels by an OWL (Web Ontology Language) ontology, and further develop a new ZSL framework with ontology embedding. The effectiveness has been verified by some primary experiments on animal image classification and visual question answering.

PDF Details DOI

IJCAI Conference 2019 Conference Paper

Augmenting Transfer Learning with Semantic Reasoning

Freddy Lécué
Jiaoyan Chen
Jeff Z. Pan
Huajun Chen

Transfer learning aims at building robust prediction models by transferring knowledge gained from one problem to another. In the semantic Web, learning tasks are enhanced with semantic representations. We exploit their semantics to augment transfer learning by dealing with when to transfer with semantic measurements and what to transfer with semantic embeddings. We further present a general framework that integrates the above measurements and embeddings with existing transfer learning algorithms for higher performance. It has demonstrated to be robust in two real-world applications: bus delay forecasting and air quality forecasting.

PDF Details

ECAI Conference 2016 Conference Paper

Explanatory Diagnosis of an Ontology Stream via Reasoning About Actions

Quan Yu
Hai Wan
Jiangtao Xu
Freddy Lécué
Liang Chang 0003

Explanatory diagnosis of an ontology stream aims to explain the changes hidden in the ontology stream by a sequence of actions. In this paper, we present a framework for explanatory diagnosis of an ontology stream, which allows the actions to be uncertain. In order to capture the semantics of actions, we introduce a new update operator and effect-guided bold-repair. By combining these operators with a query mechanism of description logicssupporting inconsistency-tolerant semantics, we present a formal definition for the explanatory diagnosis problem of ontology streams.

Details

ECAI Conference 2014 Conference Paper

Exploiting the Semantic Web for Systems Diagnosis

Anika Schumann
Freddy Lécué
Joern Ploennigs

Diagnosis is the task of explaining abnormal behaviors of systems like telecommunication, transportation or energy systems. Given a sequence of observations the problem is to determine, online, all faults that are in line with these observations. Many approaches tackle this problem but they either require domain expertise or a formal description of how observations and faults are connected. This limits their scope to the diagnosis of well-understood faults. We address the problem of diagnosing faults that may occur for the first time and present a new diagnosis approach that integrates techniques for analyzing semantic descriptions of observations and faults.

Details

ECAI Conference 2014 Conference Paper

Extending Semantic Sensor Networks for Automatically Tackling Smart Building Problems

Joern Ploennigs
Anika Schumann
Freddy Lécué

Sensor systems are constantly growing in all application areas and become elements of our environment. Semantic Sensor Networks (SSN) support this development and provide standardized semantic access for reasoning on this information. Unfortunately they do not model internal system knowledge or simple correlations between sensors and hence they cannot be used to automatically perform analytics tasks based on sensor data only. We show how SSN ontology can be extended and demonstrate its benefits for the task of diagnosing smart building problems using real-world data.

Details

IJCAI Conference 2013 Conference Paper

Predicting Knowledge in an Ontology Stream

Freddy Lécué
Jeff Z. Pan

Recently, ontology stream reasoning has been introduced as a multidisciplinary approach, merging synergies from Artiﬁcial Intelligence, Database, World-Wide-Web to reason on semantic augmented data streams. Although knowledge evolution and real-time reasoning have been largely addressed in ontology streams, the challenge of predicting its future (or missing) knowledge remains open and yet unexplored. We tackle predictive reasoning as a correlation and interpretation of past semanticsaugmented data over exogenous ontology streams. Consistent predictions are constructed as Description Logics entailments by selecting and applying relevant cross-streams association rules. The experiments have shown accurate prediction with real and live stream data from Dublin City in Ireland.

PDF Details DOI

ECAI Conference 2008 Conference Paper

Optimizing Causal Link Based Web Service Composition

Freddy Lécué
Alexandre Delteil
Alain Léger

Automation of Web service composition is one of the most interesting challenges facing the Semantic Web today. Since Web services have been enhanced with formal semantic descriptions, it becomes conceivable to exploit causal links i. e. , semantic matching between their functional parameters (i. e. , outputs and inputs). The semantic quality of causal links involved in a composition can be then used as a innovative and distinguishing criterion to estimate its overall semantic quality. Therefore non functional criteria such as quality of service (QoS) are no longer considered as the only criteria to rank compositions satisfying the same goal. In this paper we focus on semantic quality of causal link based semantic Web service composition. First of all, we present a general and extensible model to evaluate quality of both elementary and composition of causal links. From this, we introduce a global causal link selection based approach to retrieve the optimal composition. This problem is formulated as an optimization problem which is solved using efficient integer linear programming methods. The preliminary evaluation results showed that our global selection based approach is not only more suitable than the local approach but also outperforms the naive approach.

Details