Recent work on predicting category structure with distributional models, using either static word embeddings (Heyman and Heyman, 2019) or contextualized language models (CLMs) (Misra et al., 2021), report low correlations with human ratings, thus calling into question their plausibility as models of human semantic memory. In this work, we revisit this question testing a wider array of methods for probing CLMs for predicting typicality scores. Our experiments, using BERT (Devlin et al., 2018), show the importance of using the right type of CLM probes, as our best BERT-based typicality prediction methods substantially improve over previous works. Second, our results highlight the importance of polysemy in this task: our best results are obtained when using a disambiguation mechanism. Finally, additional experiments reveal that Information Contentbased WordNet (Miller, 1995), also endowed with disambiguation, match the performance of the best BERT-based method, and in fact capture complementary information, which can be combined with BERT to achieve enhanced typicality predictions.

Lexical Semantic Change is the study of how the meaning of words evolves through time. Another related question is whether and how lexical relations over pairs of words, such as synonymy, change over time. There are currently two competing, apparently opposite hypotheses in the historical linguistic literature regarding how synonymous words evolve: the Law of Differentiation (LD) argues that synonyms tend to take on different meanings over time, whereas the Law of Parallel Change (LPC) claims that synonyms tend to undergo the same semantic change and therefore remain synonyms. So far, there has been little research using distributional models to assess to what extent these laws apply on historical corpora. In this work, we take a first step toward detecting whether LD or LPC operates for given word pairs. After recasting the problem into a more tractable task, we combine two linguistic resources to propose the first complete evaluation framework on this problem and provide empirical evidence in favor of a dominance of LD. We then propose various computational approaches to the problem using Distributional Semantic Models and grounded in recent literature on Lexical Semantic Change detection. Our best approaches achieve a balanced accuracy above 0.6 on our dataset. We discuss challenges still faced by these approaches, such as polysemy or the potential confusion between synonymy and hypernymy.

We propose a simple unsupervised approach which exclusively relies on WordNet (Miller, 1995) for predicting graded lexical entailment (GLE) in English. Inspired by the seminal work of Resnik (1995), our method models GLE as the sum of two information-theoretic scores: a symmetric semantic similarity score and an asymmetric specificity loss score, both exploiting the hierarchical synset structure of Word-Net. Our approach also includes a simple disambiguation mechanism to handle polysemy in a given word pair. Despite its simplicity, our method achieves performance above the state of the art (Spearman $\rho$ = 0.75) on HyperLex (Vulic et al., 2017), the largest GLE dataset, outperforming all previous methods, including specialized word embeddings approaches that use WordNet as weak supervision.

In multimodal understanding tasks, visual and linguistic ambiguities can arise. Visual ambiguity can occur when visual objects require a model to ground a referring expression in a video without strong supervision, while linguistic ambiguity can occur from changes in entities in action flows. As an example from the cooking domain, ”oil” mixed with ”salt” and ”pepper” could later be referred to as a ”mixture”. Without a clear visual-linguistic alignment, we cannot know which among several objects shown is referred to by the language expression ”mixture”, and without resolved antecedents, we cannot pinpoint what the mixture is. We define this chicken-and-egg problem as visual-linguistic ambiguity. In this paper, we present Find2Find, a joint anaphora resolution and object localization dataset targeting the problem of visual-linguistic ambiguity, consisting of 500 anaphora-annotated recipes with corresponding videos. We present experimental results of a novel end-to-end joint multitask learning framework for Find2Find that fuses visual and textual information and shows improvements both for anaphora resolution and object localization as compared to a strong single-task baseline.

Words of Estimative Probability (WEP) are phrases used to express the plausibility of a statement. Examples include terms like probably, maybe, likely, doubt, unlikely, and impossible. Surveys have shown that human evaluators tend to agree when assigning numerical probability levels to these WEPs. For instance, the term highly likely equates to a median probability of 0.90$\pm$0.08 according to a survey by Fagen-Ulmschneider (2015). In this study, our focus is to gauge the competency of neural language processing models in accurately capturing the consensual probability level associated with each WEP. Our first approach is utilizing the UNLI dataset (Chen et al., 2020), which links premises and hypotheses with their perceived joint probability p. From this, we craft prompts in the form: ”[PREMISE]. [WEP], [HYPOTHESIS].” This allows us to evaluate whether language models can predict if the consensual probability level of a WEP aligns closely with p. In our second approach, we develop a dataset based on WEP-focused probabilistic reasoning to assess if language models can logically process WEP compositions. For example, given the prompt ”[EVENTA] is likely. [EVENTB] is impossible.”, a wellfunctioning language model should not conclude that [EVENTA&B] is likely. Through our study, we observe that both tasks present challenges to out-of-the-box English language models. However, we also demonstrate that fine-tuning these models can lead to significant and transferable improvements.

Theory of Mind (ToM) is a critical component of intelligence, yet accurately measuring it continues to be a subject of debate. Prior research has attempted to apply human ToM assessments to natural language processing models using either human-created standardized tests or rule-based templates. However, these methods primarily focus on simplistic reasoning and require further validation. In this study, we utilize dynamic epistemic logic, which has established overlaps with ToM, to generate more intricate problems. We also introduce novel verbalization techniques to express these problems using natural language. Our findings indicate that certain language model scaling (from 70M to 6B and 350M to 174B) does not consistently yield results better than random chance. While GPT-4 demonstrates improved epistemic reasoning capabilities, there is still room for enhancement. Our code and datasets are publicly available here and there.

The HuggingFace Datasets Hub hosts thousands of datasets, offering exciting opportunities for language model training and evaluation. However, datasets for a specific task type often have different schemas, making harmonization challenging. Multi-task training or evaluation necessitates manual work to fit data into task templates. Several initiatives independently tackle this issue by releasing harmonized datasets or providing harmonization codes to preprocess datasets into a consistent format. We identify patterns across previous preprocessing efforts, such as column name mapping and extracting specific sub-fields from structured data in a column. We then propose a structured annotation framework that ensures our annotations are fully exposed and not hidden within unstructured code. We release a dataset annotation framework and dataset annotations for more than 500 English tasks1. These annotations include metadata, such as the names of columns to be used as input or labels for all datasets, which can save time for future dataset preprocessing, regardless of whether our framework is utilized. We fine-tune a multi-task text encoder on all tasksource tasks, outperforming every publicly available text encoder of comparable size in an external evaluation.

Modal verbs in English communicate delicate shades of meaning, there being a large range of verbs both on the necessity side (must, have to, should, ought to, need, need to) and the possibility side (can, may, could, might, be able to). They therefore constitute excellent test ground to apply and compare different methodologies that can lay bare the factors that drive the speaker’s choice of modal verb. This book is not merely concerned with a purely grammatical description of the use of modal verbs, but aims at advancing our understanding of lexical and grammatical units in general and of linguistic methodologies to explore these. It thus involves a genuine effort to compare, assess and combine a variety of approaches. It complements the leading descriptive qualitative work on modal verbs by testing a diverse range of quantitative methods, while not ignoring qualitative issues pertaining to the semantics-pragmatics interface. Starting from a critical assessment of what constitutes the meaning of modal verbs, different types of empirical studies (usage-based, data-driven and experimental), drawing considerably on the same data sets, shows how method triangulation can contribute to an enhanced understanding. Due attention is also given to individual variation as well as the degree to which modals can predict L2 proficiency level.

The race to train language models on vast, diverse, and inconsistently documented datasets has raised pressing concerns about the legal and ethical risks for practitioners. To remedy these practices threatening data transparency and understanding, we convene a multidisciplinary effort between legal and machine learning experts to systematically audit and trace 1800+ text datasets. We develop tools and standards to trace the lineage of these datasets, from their source, creators, series of license conditions, properties, and subsequent use. Our landscape analysis highlights the sharp divides in composition and focus of commercially open vs closed datasets, with closed datasets monopolizing important categories: lower resource languages, more creative tasks, richer topic variety, newer and more synthetic training data. This points to a deepening divide in the types of data that are made available under different license conditions, and heightened implications for jurisdictional legal interpretations of copyright and fair use. We also observe frequent miscategorization of licenses on widely used dataset hosting sites, with license omission of 70%+ and error rates of 50%+. This points to a crisis in misattribution and informed use of the most popular datasets driving many recent breakthroughs. As a contribution to ongoing improvements in dataset transparency and responsible use, we release our entire audit, with an interactive UI, the Data Provenance Explorer, which allows practitioners to trace and filter on data provenance for the most popular open source finetuning data collections: www.dataprovenance.org.

One of the key challenges in decentralized and federated learning is to design algorithms that efficiently deal with highly heterogeneous data distributions across agents. In this paper, we revisit the analysis of the popular Decentralized Stochastic Gradient Descent algorithm (D-SGD) under data heterogeneity. We exhibit the key role played by a new quantity, called neighborhood heterogeneity, on the convergence rate of D-SGD. By coupling the communication topology and the heterogeneity, our analysis sheds light on the poorly understood interplay between these two concepts. We then argue that neighborhood heterogeneity provides a natural criterion to learn data-dependent topologies that reduce (and can even eliminate) the otherwise detrimental effect of data heterogeneity on the convergence time of D-SGD. For the important case of classification with label skew, we formulate the problem of learning such a good topology as a tractable optimization problem that we solve with a Frank-Wolfe algorithm. As illustrated over a set of simulated and real-world experiments, our approach provides a principled way to design a sparse topology that balances the convergence speed and the per-iteration communication costs of D-SGD under data heterogeneity.

In this paper, we introduce a conformal prediction method to construct prediction sets in a oneshot federated learning setting. More specifically, we define a quantile-of-quantiles estimator and prove that for any distribution, it is possible to output prediction sets with desired coverage in only one round of communication. To mitigate privacy issues, we also describe a locally differentially private version of our estimator. Finally, over a wide range of experiments, we show that our method returns prediction sets with coverage and length very similar to those obtained in a centralized setting. Overall, these results demonstrate that our method is particularly well-suited to perform conformal predictions in a one-shot federated learning setting.

Sharing real-world speech utterances is key to the training and deployment of voice-based services. However, it also raises privacy risks as speech contains a wealth of personal data. Speaker anonymization aims to remove speaker information from a speech utterance while leaving its linguistic and prosodic attributes intact. State-of-the-art techniques operate by disentangling the speaker information (represented via a speaker embedding) from these attributes and re-synthesizing speech based on the speaker embedding of another speaker. Prior research in the privacy community has shown that anonymization often provides brittle privacy protection, even less so any provable guarantee. In this work, we show that disentanglement is indeed not perfect: linguistic and prosodic attributes still contain speaker information. We remove speaker information from these attributes by introducing differentially private feature extractors based on an autoencoder and an automatic speech recognizer, respectively, trained using noise layers. We plug these extractors in the state-of-the-art anonymization pipeline and generate, for the first time, differentially private utterances with a provable upper bound on the speaker information they contain. We evaluate empirically the privacy and utility resulting from our differentially private speaker anonymization approach on the LibriSpeech data set. Experimental results show that the generated utterances retain very high utility for automatic speech recognition training and inference, while being much better protected against strong adversaries who leverage the full knowledge of the anonymization process to try to infer the speaker identity.

We study differentially private (DP) machine learning algorithms as instances of noisy fixed-point iterations, in order to derive privacy and utility results from this well-studied framework. We show that this new perspective recovers popular private gradient-based methods like DP-SGD and provides a principled way to design and analyze new private optimization algorithms in a flexible manner. Focusing on the widely-used Alternating Directions Method of Multipliers (ADMM) method, we use our general framework to derive novel private ADMM algorithms for centralized, federated and fully decentralized learning. For these three algorithms, we establish strong privacy guarantees leveraging privacy amplification by iteration and by subsampling. Finally, we provide utility guarantees using a unified analysis that exploits a recent linear convergence result for noisy fixed-point iterations.

State-of-the-art approaches for training Differentially Private (DP) Deep Neural Networks (DNN) faces difficulties to estimate tight bounds on the sensitivity of the network's layers, and instead rely on a process of per-sample gradient clipping. This clipping process not only biases the direction of gradients but also proves costly both in memory consumption and in computation. To provide sensitivity bounds and bypass the drawbacks of the clipping process, our theoretical analysis of Lipschitz constrained networks reveals an unexplored link between the Lipschitz constant with respect to their input and the one with respect to their parameters. By bounding the Lipschitz constant of each layer with respect to its parameters we guarantee DP training of these networks. This analysis not only allows the computation of the aforementioned sensitivities at scale but also provides leads on to how maximize the gradient-to-noise ratio for fixed privacy guarantees. To facilitate the application of Lipschitz networks and foster robust and certifiable learning under privacy guarantees, we provide a Python package that implements building blocks allowing the construction and private training of such networks.

In this paper, we study the problem of learning Graph Neural Networks (GNNs) with Differential Privacy (DP). We propose a novel differentially private GNN based on Aggregation Perturbation (GAP), which adds stochastic noise to the GNN's aggregation function to statistically obfuscate the presence of a single edge (edge-level privacy) or a single node and all its adjacent edges (node-level privacy). Tailored to the specifics of private learning, GAP's new architecture is composed of three separate modules: (i) the encoder module, where we learn private node embeddings without relying on the edge information; (ii) the aggregation module, where we compute noisy aggregated node embeddings based on the graph structure; and (iii) the classification module, where we train a neural network on the private aggregations for node classification without further querying the graph edges. GAP's major advantage over previous approaches is that it can benefit from multi-hop neighborhood aggregations, and guarantees both edge-level and node-level DP not only for training, but also at inference with no additional costs beyond the training's privacy budget. We analyze GAP's formal privacy guarantees using Rényi DP and conduct empirical experiments over three real-world graph datasets. We demonstrate that GAP offers significantly better accuracy-privacy trade-offs than state-of-the-art DP-GNN approaches and naive MLP-based baselines. Our code is publicly available here.

In this paper, we study differentially private empirical risk minimization (DP-ERM). It has been shown that the worst-case utility of DP-ERM reduces polynomially as the dimension increases. This is a major obstacle to privately learning large machine learning models. In high dimension, it is common for some model's parameters to carry more information than others. To exploit this, we propose a differentially private greedy coordinate descent (DP-GCD) algorithm. At each iteration, DP-GCD privately performs a coordinate-wise gradient step along the gradients' (approximately) greatest entry. We show theoretically that DP-GCD can achieve a logarithmic dependence on the dimension for a wide range of problems by naturally exploiting their structural properties (such as quasi-sparse solutions). We illustrate this behavior numerically, both on synthetic and real datasets.

We theoretically study the impact of differential privacy on fairness in classification. We prove that, given a class of models, popular group fairness measures are pointwise Lipschitz-continuous with respect to the parameters of the model. This result is a consequence of a more general statement on accuracy conditioned on an arbitrary event (such as membership to a sensitive group), which may be of independent interest. We use this Lipschitz property to prove a non-asymptotic bound showing that, as the number of samples increases, the fairness level of private models gets closer to the one of their non-private counterparts. This bound also highlights the importance of the confidence margin of a model on the disparate impact of differential privacy.

Pufferfish privacy is a flexible generalization of differential privacy that allows to model arbitrary secrets and adversary's prior knowledge about the data. Unfortunately, designing general and tractable Pufferfish mechanisms that do not compromise utility is challenging. Furthermore, this framework does not provide the composition guarantees needed for a direct use in iterative machine learning algorithms. To mitigate these issues, we introduce a Rényi divergence-based variant of Pufferfish and show that it allows us to extend the applicability of the Pufferfish framework. We first generalize the Wasserstein mechanism to cover a wide range of noise distributions and introduce several ways to improve its utility. We also derive stronger guarantees against out-ofdistribution adversaries. Finally, as an alternative to composition, we prove privacy amplification results for contractive noisy iterations and showcase the first use of Pufferfish in private convex optimization. A common ingredient underlying our results is the use and extension of shift reduction lemmas.

In the past decades, concerns about the societal impact of machine learning have been growing. Indeed, if machine learning has proven its usefulness in science, day-to-day applications, and many other domains, its success is principally due to the availability of large datasets. This raises two concerns, the first about the confidentiality of the training data, and the second, about possible discrimination in a model's predictions. Trustworthy machine learning aims at providing technical answers to these concerns.Unfortunately, guaranteeing the privacy of the training data and the fairness of the predictions often decreases the utility of the learned model. This problem has drawn significant interest in the past years, but most of existing methods (usually based on stochastic gradient descent) tend to fail in some common scenarios, like training of high-dimensional models. In this thesis, we study how structural properties of machine learning problems can be exploited to improve the trade-off between privacy and utility, and how this can impact the fairness of the predictions.The first two contributions of this thesis are two new differentially private optimization algorithms, that are both based on coordinate descent. They aim at exploiting different structural properties of the problem at hand. The first algorithm is based on stochastic coordinate descent, and can exploit imbalance in the scale of the gradient's coordinates by using large step sizes. This allows our algorithm to obtain useful models in difficult problems, where stochastic gradient descent quickly stalls. The second algorithm is based on greedy coordinate descent. Its greedy updates allow to focus on the most important coordinates of the problem, which can sometimes drastically improve utility (e.g. when the solution of the problem is sparse).The third contribution of this thesis studies the interplay of differential privacy and fairness in machine learning. These two notions have rarely been studied simultaneously, and there are growing concerns that differential privacy may exacerbate unfairness. We show that group fairness measures have interesting regularity properties, provided that the predictions of the model are Lipschitz-continuous in its parameters. This result allows to derive a bound on the difference in fairness levels between a private model and its non-private counterpart.

In recent years, due to the growing importance of network applications and the growing concerns for privacy, there is an increasing interest in decentralized forms of machine learning. In this dissertation, we study the setting that involves a communication network of agents, where each agent locally privatizes (adds noise to) its data, and where the agents aim to collaboratively learn statistical models over their data. Such local privatization is in line with a standard of data privacy known as local differential privacy, and local differential privacy is useful when alternatives, such as secure multi-party computation or central differential privacy performed by a trusted curator, are infeasible. However, local differential privacy results, typically, in worse utility (less accurate statistical models) compared to central differential privacy because, for the same privacy budget, local differential privacy adds more privatization noise than central differential privacy. The principal question of this dissertation is the following: given that local differential privacy must be used, how could the agents maximize the utility they achieve? We study two cases to address the stated principal question.In the first case, we consider the problem of distributed averaging, where each agent intends to collaboratively compute the unbiased average over the individual values of all agents without revealing neither their sensitive attributes nor their degree (number of neighbors). Usually, existing works solve this problem by assuming that either (i) each agent reveals its degree to its neighbors or (ii) every two neighboring agents can perform handshakes (requests that rely on replies) in every exchange of information. Since such assumptions are not always desirable, we propose an approach that is handshake-free and where the degrees are privatized. In particular, we use a gossip algorithm that computes averages that are biased when the graph of agents is non-regular (when the vertices have unequal degrees) and then perform a procedure combining multiple biased averages for bias correction. We apply the proposed approach for fitting linear regression models. We prove the asymptotic guarantee that the mean squared error between the average of privatized attributes computed by our approach and the average of sensitive attributes is $𝒪\left(\frac{1}{n}\right)$, where $n$ is the number of agents.In the second case, we consider a group of agents, where features (for fitting regression models) are computed by transforming sensitive attributes, and where the transformations have high-magnitude gradients or singularities. In such setting, there is a risk to magnify the privatization noise if the perturbed data is in an interval where the feature function has high-magnitude gradients. We provide a tailored noise mechanism for privatizing features by solving a convex program in such a way that (i) only pertinent intervals of transformations are selected, (ii) the variance of privatization noise is minimized, and (iii) the biasedness of privatization noise is minimized.

This paper is an overview of the EU-funded project TRUMPET (see here), and gives an outline of its scope and main technical aspects and objectives. In recent years, Federated Learning has emerged as a revolutionary privacy-enhancing technology. However, further research has cast a shadow of doubt on its strength for privacy protection. The goal of TRUMPET is to research and develop novel privacy enhancement methods for Federated Learning, and to deliver a highly scalable Federated AI service platform for researchers, that will enable AI-powered studies of siloed, multi-site, crossdomain, cross-border European datasets with privacy guarantees that follow the requirements of GDPR. The generic TRUMPET platform will be piloted, demonstrated and validated in the specific use case of European cancer hospitals, allowing researchers and policymakers to extract AI-driven insights from previously inaccessible cross-border, cross-organization cancer data, while ensuring the patients' privacy.

In this work, we consider the problem of intersectional group fairness in the classification setting, where the objective is to learn discrimination-free models in the presence of several intersecting sensitive groups. First, we illustrate various shortcomings of existing fairness measures commonly used to capture intersectional fairness. Then, we propose a new definition called the $\alpha$-Intersectional Fairness, which combines the absolute and the relative performance across sensitive groups and can be seen as a generalization of the notion of differential fairness. We highlight several desirable properties of the proposed definition and analyze its relation to other fairness measures. Finally, we benchmark multiple popular in-processing fair machine learning approaches using our new fairness definition and show that they do not achieve any improvement over a simple baseline. Our results reveal that the increase in fairness measured by previous definitions hides a ”leveling down” effect, i.e., degrading the best performance over groups rather than improving the worst one.

We tackle the problem of group fairness in classification, where the objective is to learn models that do not unjustly discriminate against subgroups of the population. Most existing approaches are limited to simple binary tasks or involve difficult to implement training mechanisms. This reduces their practical applicability. In this paper, we propose FairGrad, a method to enforce fairness based on a reweighting scheme that iteratively learns group specific weights based on whether they are advantaged or not. FairGrad is easy to implement and can accommodate various standard fairness definitions. Furthermore, we show that it is comparable to standard baselines over various datasets including ones used in natural language processing and computer vision.

In this paper we study verifiable sampling from probability distributions in the context of multi-party computation. This has various applications in randomized algorithms performed collaboratively by parties not trusting each other. One example is differentially private machine learning where noise should be drawn, typically from a Laplace or Gaussian distribution, and it is desirable that no party can bias this process. In particular, we propose algorithms to draw random numbers from uniform, Laplace, Gaussian and arbitrary probability distributions, and to verify honest execution of the protocols through zero-knowledge proofs. We propose protocols that result in one party knowing the drawn number and protocols that deliver the drawn random number as a shared secret.

Comparison-based learning addresses the problem of learning when, instead of explicit features or pairwise similarities, one only has access to comparisons of the form: Object $A$ is more similar to $B$ than to $C$. Recently, it has been shown that, in Hierarchical Clustering, single and complete linkage can be directly implemented using only such comparisons while several algorithms have been proposed to emulate the behaviour of average linkage. Hence, finding hierarchies (or dendrograms) using only comparisons is a well understood problem. However, evaluating their meaningfulness when no ground-truth nor explicit similarities are available remains an open question. In this paper, we bridge this gap by proposing a new revenue function that allows one to measure the goodness of dendrograms using only comparisons. We show that this function is closely related to Dasgupta's cost for hierarchical clustering that uses pairwise similarities. On the theoretical side, we use the proposed revenue function to resolve the open problem of whether one can approximately recover a latent hierarchy using few triplet comparisons. On the practical side, we present principled algorithms for comparison-based hierarchical clustering based on the maximisation of the revenue and we empirically compare them with existing methods.

Maximum common substructures (MCS) have received a lot of attention in the chemoinformatics community. They are typically used as a similarity measure between molecules, showing high predictive performance when used in classification tasks, while being easily explainable substructures. In the present work, we applied the Pairwise Maximum Common Subgraph Feature Generation (PMCSFG) algorithm to automatically detect toxicophores (structural alerts) and to compute fingerprints based on MCS. We present a comparison between our MCS-based fingerprints and 12 well-known chemical fingerprints when used as features in machine learning models. We provide an experimental evaluation and discuss the usefulness of the different methods on mutagenicity data. The features generated by the MCS method have a state-of-the-art performance when predicting mutagenicity, while they are more interpretable than the traditional chemical fingerprints.

Gait analysis is important for evaluating neurological disorders such as stroke and Parkinson's disease. Traditionally, healthcare professionals had to rely on subjective assessments (i.e., human-based) of gait which were time consuming and not very reproducible. However, with the advent of IoT and indeed more objective (e.g., measurement-based) assessment methods, gait analysis can now be performed more accurately and effectively. It is worth noting, however, that there are still limitations to these objective methods, especially the lack of privacy-preserving continuous data collection. To overcome this limitation, we present in this paper a privacy-by-design monitoring application for post-stroke patients to evaluate their gait before, during, and after a rehabilitation program. Gait measurements are collected by a mobile application that continuously captures spatiotemporal parameters in the background using the built-in smartphone accelerometer. Statistical techniques are then applied to extract general indicators about the performed activity, as well as some more specific gait metrics in real-time such as regularity, symmetry and walking speed. These metrics are calculated based on the detected steps while patients are performing an activity. Additionally, a deep learning approach based on an auto-encoder is implemented to detect abnormal activities in the gait of patients. These analyses provides both valuable insights and statistical information about the activities performed by the patient, and a useful tool for practitioners to monitor the progression of neurological disorders and detect anomalies. We conducted experiments using this application in real conditions to monitor post-stroke patients in collaboration with a hospital, demonstrating its ability to compute valuable metrics and detect abnormal events patient's gait.

In this paper, we study the problem of optimizing a linear program whose variables are the answers to a conjunctive query. For this we propose the language LP(CQ) for specifying linear programs whose constraints and objective functions depend on the answer sets of conjunctive queries. We contribute an efficient algorithm for solving programs in a fragment of LP(CQ). The natural approach constructs a linear program having as many variables as there are elements in the answer set of the queries. Our approach constructs a linear program having the same optimal value but fewer variables. This is done by exploiting the structure of the conjunctive queries using generalized hypertree decompositions of small width to factorize elements of the answer set together. We illustrate the various applications of LP(CQ) programs on three examples: optimizing deliveries of resources, minimizing noise for differential privacy, and computing the s-measure of patterns in graphs as needed for data mining.