As Large Langue Models have been shown to memorize real-world facts, the need to update this knowledge in a controlled and efficient manner arises.Designed with these constraints in mind, Knowledge Editing (KE) approaches propose to alter specific facts in pretrained models.However, they have been shown to suffer from several limitations, including their lack of contextual robustness and their failure to generalize to logical implications related to the fact.To overcome these issues, we propose SAKE, a steering activation method that models a fact to be edited as a distribution rather than a single prompt.Leveraging Optimal Transport, SAKE alters the LLM behavior over a whole fact-related distribution, defined as paraphrases and logical implications.Several numerical experiments demonstrate the effectiveness of this method: SAKE is thus able to perform more robust edits than its existing counterparts.The code to reproduce all experiments is made available on a repository 1 .

In the field of algorithmic fairness, significant attention has been put on group fairness criteria, such as Demographic Parity and Equalized Odds. Nevertheless, these objectives, measured as global averages, have raised concerns about persistent local disparities between sensitive groups. In this work, we address the problem of local fairness, which ensures that the predictor is unbiased not only in terms of expectations over the whole population, but also within any subregion of the feature space, unknown at training time. To enforce this objective, we introduce ROAD, a novel approach that leverages the Distributionally Robust Optimization (DRO) framework within a fair adversarial learning objective, where an adversary tries to infer the sensitive attribute from the predictions. Using an instance-level re-weighting strategy, ROAD is designed to prioritize inputs that are likely to be locally unfair, i.e. where the adversary faces the least difficulty in reconstructing the sensitive attribute. Numerical experiments demonstrate the effectiveness of our method: it achieves Pareto dominance with respect to local fairness and accuracy for a given global fairness level across three standard datasets, and also enhances fairness generalization under distribution shift.

We propose a simple way to use large language models (LLMs) in education. Specifically, our method aims to improve individual comprehension by adding a novel feature to online videos. We combine the low threshold for interactivity in digital experiences with the benefits of rephrased and elaborated explanations typical of face-to-face interactions, thereby supporting to close knowledge gaps at scale. To demonstrate the technical feasibility of our approach, we conducted a proof-of-concept experiment and implemented a prototype which is available for testing online. Through the use case, we also show how caching can be applied in LLM-powered applications to reduce their carbon footprint.

Few-Shot Object Detection (FSOD) allows fast adaptation of an object detection model to new classes of objects using few examples per class. This has many applications, in particular in satellite and aerial observation, as it allows learning from experts who can only annotate a few examples for new classes and helps migrate models across tasks. In this work, we present a technique to improve the performance of FSOD in remote sensing by defining a contrastive loss that utilizes parts of objects. For this, we generate, what we call, Object Parts Proposals (OPPs) on the fly for each novel class, and use them to learn more robust features with an additional contrastive objective. We observe that training with OPPs brings a consistent improvement over the state-of-the-art when evaluating on the DIOR dataset.The code is available at https://github.com/arthurchevalley/Improving-FSOD-on-RSI-using-Sub-Parts.

Plural counterfactual examples have been proposed to explain the prediction of a classifier by offering a user several instances of minimal modifications that may be performed to change the prediction. Yet, such explanations may provide too much information, generating potential confusion for the end-users with no specific knowledge, neither on the machine learning, nor on the application domains. In this paper, we investigate the design of explanation user interfaces for plural counterfactual examples offering comparative analysis features to mitigate this potential confusion and improve the intelligibility of such explanations for non-expert users. We propose an implementation of such an enhanced explanation user interface, illustrating it in a financial scenario related to a loan application. We then present the results of a lab user study conducted with 112 participants to evaluate the effectiveness of having plural examples and of offering comparative analysis principles, both on the objective understanding and satisfaction of such explanations. The results demonstrate the effectiveness of the plural condition, both on objective understanding and satisfaction scores, as compared to having a single counterfactual example. Beside the statistical analysis, we perform a thematic analysis of the participants’ responses to the open-response questions, that also shows encouraging results for the comparative analysis features on the objective understanding.

In recent years, most fairness strategies in machine learning have focused on mitigating unwanted biases by assuming that the sensitive information is available. However, in practice this is not always the case: due to privacy purposes and regulations such as RGPD in EU, many personal sensitive attributes are frequently not collected. Yet, only a few prior works address the issue of mitigating bias in such a difficult setting, in particular to meet classical fairness objectives such as Demographic Parity and Equalized Odds. By leveraging recent developments for approximate inference, we propose in this paper an approach to fill this gap. To infer a sensitive information proxy, we introduce a new variational auto-encoding-based framework named SRCVAE that relies on knowledge of the underlying causal graph. The bias mitigation is then done in an adversarial fairness approach. Our proposed method empirically achieves significant improvements over existing works in the field. We observe that the generated proxy’s latent space correctly recovers sensitive information and that our approach achieves a higher accuracy while obtaining the same level of fairness on two real datasets.

The increasing usage of complex Machine Learning models for decision-making has raised interest in explainable artificial intelligence (XAI). In this work, we focus on the effects of providing accessible and useful explanations to non-expert users. More specifically, we propose generic XAI design principles for contextualizing and allowing the exploration of explanations based on local feature importance. To evaluate the effectiveness of these principles for improving users’ objective understanding and satisfaction, we conduct a controlled user study with 80 participants using 4 different versions of our XAI system, in the context of an insurance scenario. Our results show that the contextualization principles we propose significantly improve user’s satisfaction and is close to have a significant impact on user’s objective understanding. They also show that the exploration principles we propose improve user’s satisfaction. On the other hand, the interaction of these principles does not appear to bring improvement on both dimensions of users’ understanding.

The past few years have seen a dramatic rise of academic and societal interest in fair machine learning. While plenty of fair algorithms have been proposed recently to tackle this challenge for discrete variables, only a few ideas exist for continuous ones. The objective in this paper is to ensure some independence level between the outputs of regression models and any given continuous sensitive variables. For this purpose, we use the Hirschfeld-Gebelein-Rényi (HGR) maximal correlation coefficient as a fairness metric. We propose to minimize the HGR coefficient directly with an adversarial neural network architecture. The idea is to predict the output Y while minimizing the ability of an adversarial neural network to find the estimated transformations which are required to predict the HGR coefficient. We empirically assess and compare our approach and demonstrate significant improvements on previously presented work in the field.

Abstract Fair classification has become an important topic in machine learning research. While most bias mitigation strategies focus on neural networks, we noticed a lack of work on fair classifiers based on decision trees even though they have proven very efficient. In an up-to-date comparison of state-of-the-art classification algorithms in tabular data, tree boosting outperforms deep learning (Zhang et al. in Expert Syst Appl 82:128–150, 2017). For this reason, we have developed a novel approach of adversarial gradient tree boosting. The objective of the algorithm is to predict the output Y with gradient tree boosting while minimizing the ability of an adversarial neural network to predict the sensitive attribute S. The approach incorporates at each iteration the gradient of the neural network directly in the gradient tree boosting. We empirically assess our approach on four popular data sets and compare against state-of-the-art algorithms. The results show that our algorithm achieves a higher accuracy while obtaining the same level of fairness, as measured using a set of different common fairness definitions.

Fair classification has become an important topic in machine learning research. While most bias mitigation strategies focus on neural networks, we noticed a lack of work on fair classifiers based on decision trees even though they have proven very efficient. In an up-to-date comparison of state-of-the-art classification algorithms in tabular data, tree boosting outperforms deep learning. For this reason, we have developed a novel approach of adversarial gradient tree boosting. The objective of the algorithm is to predict the output Y with gradient tree boosting while minimizing the ability of an adversarial neural network to predict the sensitive attribute S. The approach incorporates at each iteration the gradient of the neural network directly in the gradient tree boosting. We empirically assess our approach on 4 popular data sets and compare against state-of-the-art algorithms. The results show that our algorithm achieves a higher accuracy while obtaining the same level of fairness, as measured using a set of different common fairness definitions.

Post-hoc interpretability approaches have been proven to be powerful tools to generate explanations for the predictions made by a trained black-box model. However, they create the risk of having explanations that are a result of some artifacts learned by the model instead of actual knowledge from the data. This paper focuses on the case of counterfactual explanations and asks whether the generated instances can be justified, i.e. continuously connected to some ground-truth data. We evaluate the risk of generating unjustified counterfactual examples by investigating the local neighborhoods of instances whose predictions are to be explained and show that this risk is quite high for several datasets. Furthermore, we show that most state of the art approaches do not differentiate justified from unjustified counterfactual examples, leading to less useful explanations.

Towards conversational agents that are capable of handling more complex questions on contractual conditions, formalizing contract statements in a machine readable way is crucial. However, constructing a formal model which captures the full scope of a contract proves difficult due to the overall complexity its set of rules represent. Instead, this paper presents a top-down approach to the problem. After identifying the most relevant contract statements, we model their underlying rules in a novel knowledge engineering method. A user-friendly tool we developed for this purpose allows to do so easily and at scale. Then, we expose the statements as service so they can get smoothly integrated in any chatbot framework.

Stress is a complex phenomenon that impacts the body and the mind at several levels. It has been studied for more than a century from different perspectives, which result in different definitions and different ways to assess the presence of stress. This paper introduces a methodology for analyzing multimodal stress detection results by taking into account the variety of stress assessments. As a first step, we have collected video, depth and physiological data from 25 subjects in a stressful situation: a socially evaluated mental arithmetic test. As a second step, we have acquired three different assessments of stress: self-assessment, assessments from external observers and assessment from a physiology expert. Finally, we extract 101 behavioural and physiological features and evaluate their predictive power for the three collected assessments using a classification task. Using multimodal features, we obtain average F1 scores up to 0.85. By investigating the composition of the best selected feature subsets and the individual feature classification performances, we show that several features provide valuable information for the classification of the three assessments: features related to body movement, blood volume pulse and heart rate. From a methodological point of view, we argue that a multiple assessment approach provide more robust results.

This paper proposes an experimental comparison between human imprecise calculation and fuzzy arithmetic: an empirical study has been conducted to collect real intervals resulting from products and additions with imprecise operands from participants. Fuzzy intervals are elicited from these data and fuzzy arithmetic is applied to the collected imprecise operands. Comparisons show that the fuzzy product and addition differ from the way human beings perform these operations. Moreover, they show that the participants, rather than taking into account the imprecisions in the calculations, realise exact calculation and in the end approximate the exact result.

Imprecise numerical expressions, such as "about 100 meters", are pervasive in natural language. Mobile robotics, Geographic Information Systems, intelligent personal assistants as well as database querying applications are required to automatically and accurately interpret such expressions, called Approximate Numerical Expressions (ANE). The main challenge is to determine their numerical boundaries that sound plausible to users. The aim of this paper is to provide guidelines to interpret ANEs that are independent from the domain and the formal representations. We identified three arithmetical properties and examined their involvement in ANE interpretation as intervals of denoted values. The implicit assumption of symmetry of the intervals was also tested. To do so, 146 participants were asked to provide the intervals corresponding to 24 ANEs in a semantically neutral context. Results suggest that the properties of ANEs we identified are key factors in their interpretation while symmetry is not always maintained. This study contributes towards an understanding of how users process ANEs and its results can be used to improve intelligent interfaces that lead to better users' satisfaction and natural interaction between him/her and the system.

Generalised Modus Ponens (GMP) allows to perform logical inference in the case where an observation partially matches the premise of an implication, enriching the rule exploitation as compared to binary classical logic. This paper proposes to further enhance the rule exploitation, integrating additional constraints to guide the inference, both to reduce uncertainty in case of partial match and to perform inference in the case of an observation disjoint from the rule premise. These constraints are expressed as logical predicates derived from properties that characterise the observation, in an absolute way or relatively to the rule premise. An extension of GMP is proposed, to take into account the constraints, based on transformation operations applied to the fuzzy sets involved in the rule and the observation. An instantiation to a GMP preserving graduality and ambiguity is established and its validity is proven.

Resorting to crowdsourcing platforms is a popular way to obtain annotations. Multiple potentially noisy answers can thus be aggregated to retrieve an underlying ground truth. However, it may be irrelevant to look for a unique ground truth when we ask crowd workers for opinions, notably when dealing with subjective phenomena such as stress. In this paper, we discuss how we can better use crowdsourced annotations with an application to automatic detection of perceived stress. Towards this aim, we first acquired video data from 44 subjects in a stressful situation and gathered answers to a binary question using a crowdsourcing platform. Then, we propose to integrate two measures derived from the set of gathered answers into the machine learning framework. First, we highlight that using the consensus level among crowd worker answers substantially increases classification accuracies. Then, we show that it is suitable to directly predict for each video the proportion of positive answers to the question from the different crowd workers. Hence, we propose a thorough study on how crowdsourced annotations can be used to enhance performance of classification and regression methods.

This paper addresses the task of community detection and proposes a local approach based on a distributed list building, where each vertex broadcasts basic information that only depends on its degree and that of its neighbours. A decentralised external process then unveils the community structure. The relevance of the proposed method is experimentally shown on both artificial and real data.

In this paper a new on-line algorithm is proposed (the Droplets algorithm) for dealing with concept drifts and to produce reliable predictions. The two main characteristics of this algorithm are that it is able to adapt to different types of drifts without making any assumptions regarding their type or when they occur, and can provide reliable predictions in a non-stationary environment without using a fixed confidence threshold. Experimental results on five datasets based on Random RBF and Rotating Hyperplane generators as well as a new semi-synthetic dataset based weather temperatures show that, by discarding difficult observations, the Droplets algorithm manages to obtain the best average accuracy against ten classifiers. The results also indicate that the algorithm manages to provide reliable prediction by accurately distinguishing which observations are easily classifiable.

Time series shapelets proposes an approach to extract subsequences most suitable to discriminate time series belonging to distinct classes. Computational complexity is the major issue with shapelets: the time required to identify interesting subsequences can be intractable for large cases. In fact, it is required to evaluate all the subsequences of all the time series of the training dataset. In the literature, improvements have been proposed to accelerate the process, but few provide a solution that dramatically reduces the time required to find a solution. We propose a random-based approach that reduces the time necessary to find a solution, in our experimentation until 3 orders of magnitude compared to the original method. Based on extensive experimentations on several data sets from the literature, we show that even with a few time available, random-shapelet algorithm is able to find very competitive shapelets.

Gradual relationship between premises and conclusions is often an underlying property of fuzzy rules. In this paper, we propose to integrate the gradual hypothesis, sometimes called monotonicity, to Generalized Modus Ponens (GMP). To achieve this objective, we defined the Gradual Generalized Modus Ponens (GGMP), based on a partioning of the universe of discourse. Moreover, we prove, for this formulation, some major preservation properties, such as for the convexity, the continuity and the normality. Finally, we show that the ordering of different fuzzy observations is conserved for the associated conclusions.

This paper introduces a racing mechanism in the cluster selection process for one-pass clustering algorithms. We focus on cases where data are not numerical vectors and where it is not necessarily possible to compute a mean for each cluster. In this case, the distance of each point to existing clusters can be computed exhaustively with a quadratic complexity which is not tractable in most of nowadays use cases. In this paper we first introduce a stochastic approach for estimating the distance of each new data point to existing clusters based on Hoeffding and Bernstein bounds, that reduces the number of computations by simultaneously selecting the quantity of data to be sampled and by eliminating the non-competitive clusters. Second, this paper shows that it is possible to improve the efficiency of our approach by reducing the theoretical values of the Hoeffding and Bernstein bounds. Our algorithms, tested on real data sets, provide significant acceleration of the one-pass clustering algorithms, while making less error (or any depending on parameters) than one-pass clustering algorithm with fixed number of comparisons with each cluster.

Photovoltaic solar panels are effective energy sources during periods of bright sunlight. Excess energy can be stored for later use at night or on cloudy days. The decision to use the stored energy now or later depends largely on being able to predict the weather on different timescales. Short term prediction of stored energy is challenging due to the non-trivial I-V characteristic of the solar cell. The erratic nature of the weather makes long term predictive energy management difficult. In this paper, we address these issues based on data collected from a solar panel, as well as its relationship to observations made of the weather. We observe that prediction, based on fuzzy decision trees, reduces the energy error by 22% compared to a constant prediction equal to the average on the studied period. Thus, exploiting the fuzzy classification provided by a fuzzy decision tree is a good improvement compared to the baseline.

Investors are constantly confronted with deciding between a multitude of different investments. The characteristics, especially the estimated return, of each alternative is never precisely known. In this paper, we propose to use fuzzy present values to model this uncertainty. We extend previous work with the possibility to account for uncertain project durations, which become increasingly important for long-term projects. The results allowed a detailed assessment of the cost of hydrogen production using a thermo-chemical cycle which is still in the early phase of research. On the theoretical side, we propose a sound fuzzification over crisp domains, avoiding in particular the unsteady behaviour of existing approaches.

In this paper, we present an improved evolutionary method for the task of locating a group of buildings based solely on their relative spatial relationships. This problem arises in the general text-to-sketch problem of conflating a hand or machine drafted sketch of building locations to a satellite image. We use the histograms of forces to capture the relative position information between buildings and develop a method to compare building sets. This represents an extension to our previous work, allowing for larger placement perturbations and changes in orientation.

In this paper, we present an original fuzzy interpolation method. In contrast to existing approaches, our method is able to always construct an interpolated fuzzy interval without a need of a special step dedicated to the "standardization" of non viable solutions, which fractures the sense of the interpolation. In fact, these "standardization" steps imply that, for instance, a point obtained from the interpolation of the upper limit (right side) of the fuzzy sets, is used to build the lower limit (left side) of the interpolated conclusion, breaking the underlying hypothesis of (linear) graduality. To achieve the direct interpolation, our method is based on the deviation of the observation from the expected linearly interpolated solution and constrains of the constructed solution between extreme cases. We illustrate and discuss the behavior of our method by comparison to other well known fuzzy interpolation methods.

The noise-robust matching of two graphs is a hard combinatorial problem with practical importance in several domains. In practical applications, a unique solution for a given instance can not be defined, i.e. the actual solution may be outscored by some other due to noise effects arising during feature extraction. Soft computing approaches in general provide fast but not necessarily globally optimal solutions. In this case, the lack of guarantee of the global optimum is not a real drawback, since the uncertainty already arises in the problem definition. This paper discusses the application of memetic algorithms on the error-correcting graph isomorphism problem. We show that permutation encoding is robust enough to allow addressing both, the matching problem for graphs of the same size, and the subgraph matching problem. Since gene order information is meaningless in this particular case, a strict position based crossover is applied providing better performance than the popular PMX. We evaluate our algorithm on a synthetic data set with larger graph sizes than used in traditional, exact approaches and other permutation-based genetic approaches.

Facing a huge amount of multimedia information available today, it becomes inevitably necessary to develop efficient methods for accessing, searching, structuring, and representing it. Multimedia retrieval systems especially in the case of video should support users in all of these tasks. Therefore, specialized systems that focus on each of these aspects have been developed. However, an open research perspective issue is the development of a retrieval tool that integrates all user interactions in a single interface. In this paper, we present a system that focuses on the summarization of one single video. We consider the structuring and visualization components including reasonable user interactions to have the most significant influence on the systems usability. Our prototype is based on growing self-organizing maps. The emphasis is on intuitive hierarchical interaction for content visualization exploiting the features of the maps and integrating additional potentially useful information.

In this article, we propose a new method to retrieve images containing a request set of regions. The user is asked to specify a set of regions belonging to a single image. Then this request set of regions is compared to the sets of the regions of the images in the database. We propose a comparison measure that not only evaluates the similarity of regions one to the other, but that also takes into account the spatial configuration of the regions. The spatial structure of the regions is represented by means of fuzzy spatial relations, like horizontal and vertical disposal and connexity. © 2006 Wiley Periodicals, Inc. Int J Int Syst 21: 711–723, 2006.

We first introduce the fuzzy similarity measures in the context of a CBIR system. This leads to the observation of an invariance in the ranking for different similarity measures. We then propose an explanation to this phenomenon, and a larger theory about order invariance for fuzzy similarity measures. We introduce a definition for equivalence classes based on order conservation between these measures. We then study the consequences of this theory on the evaluation of document retrieval by fuzzy similarity.

In this paper, we present the discovery of rules for different challenges encountered in video indexing. These rules should be considered as knowledge that can be used as a guideline for the development of better indexing tools. We use a fuzzy decision tree to extract the rules based on color proportions of key-frames extracted from one single video-news. Experimental results and comparisons with other data mining tools are presented.

Video as format of computer related material is becoming more and more common [1,2]. Every day new multimedia information systems appear on the market containing more and more video. Also the format of information on Internet is clearly evolving to a video form. Initially we saw the embedding of images on text pages, now we see simple animation on almost every web page. We also know that the amount of information stored in computers is growing. So, the question that naturally arises is: “How to get the information you want?” We propose here a new tool for leading the user to make the right question to retrieve the information he wants inside a video. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

We studied here on some simple examples the interaction between valuation family, parameters and ranking result. The ranking method studied is based upon the idea of associating with a fuzzy number a scalar value, its valuation, and using this valuation to compare and order fuzzy numbers. The valuation method considered was introduced initially by the Yager and Filev. This valuation consists in the integration over α-levels, of the average of each α-cut weighted by a weight distribution function. We finish by introducing a new weight distribution function.