In this paper, we introduce SWE-QA, a text and code corpus aimed at benchmarking multi-hop code comprehension, addressing the gap between simplified evaluation tasks and the complex reasoning required in real-world software development. While existing code understanding benchmarks focus on isolated snippets, developers must routinely connect information across multiple dispersed code segments. The dataset comprises 9,072 multiple-choice questions systematically generated from 12 Python repositories of SWE-bench, evaluating several recurrent reasoning patterns like Declaration-and-Call questions that link entity definitions to their usage, and Interacting-Entity questions that examine the dynamic relationships among multiple collaborating components. Generated through parsing-based entity extraction and Large Language Model assisted question construction with carefully validated distractors, the benchmark distinguishes genuine comprehension from superficial pattern matching. Evaluation of 15 language models (360M to 671B parameters) reveals significant challenges in multi-hop reasoning, with best performance reaching 74.41% accuracy. Dense architectures consistently outperform mixture-of-experts models by 10-14 percentage points, while reasoning-enhanced variants show inconsistent benefits.

International audience

Statistical anomaly detection is critical across various domains, including healthcare, finance, industry, and cybersecurity. While supervised methods often achieve high performance, the limited availability of labeled data requires effective unsupervised techniques. In this paper, we introduce Dataset Sampling Iterative Learning (DSIL), a novel iterative learning framework for unsu-pervised anomaly detection leveraging generative modeling with diffusion. Our approach progressively refines an unlabeled dataset by identifying and removing anomalies, effectively approximating a semi-supervised setup. We demonstrate the efficiency of our framework with Diffusion Time Estimation (DTE). Furthermore, it enables better explainability through a novel approach of noised-feature discovery. Extensive experiments against unsupervised methods on both synthetic and real-world datasets demonstrate improved state-of-the-art performance. Finally, we suggest a novel usage of existing metrics to evaluate the explainability of anomaly detection models.

Self-supervised skill learning aims to acquire useful behaviors that leverage the underlying dynamics of the environment. Latent variable models, based on mutual information maximization, have been successful in this task but still struggle in the context of robotic manipulation. As it requires impacting a possibly large set of degrees of freedom composing the environment, mutual information maximization fails alone in producing useful and safe manipulation behaviors. Furthermore, tackling this by augmenting skill discovery rewards with additional rewards through a naive combination might fail to produce desired behaviors. To address this limitation, we introduce SLIM, a multi-critic learning approach for skill discovery with a particular focus on robotic manipulation. Our main insight is that utilizing multiple critics in an actor-critic framework to gracefully combine multiple reward functions leads to a significant improvement in latent-variable skill discovery for robotic manipulation while overcoming possible interference occurring among rewards which hinders convergence to useful skills. Furthermore, in the context of tabletop manipulation, we demonstrate the applicability of our novel skill discovery approach to acquire safe and efficient motor primitives in a hierarchical reinforcement learning fashion and leverage them through planning, significantly surpassing baseline approaches for skill discovery.

The robotic manipulation of deformable objects, such as clothes and fabric, is known as a complex task from both the perception and planning perspectives. Indeed, the stochastic nature of the underlying environment dynamics makes it an interesting research field for statistical learning approaches and neural policies. In this work, we introduce a novel attention-based neural architecture capable of solving a smoothing task for such objects by means of a single robotic arm. To train our network, we leverage an oracle policy, executed in simulation, which uses the topological description of a mesh of points for representing the object to smooth. In a second step, we transfer the resulting behavior in the real world with imitation learning using the cloth point cloud as decision support, which is captured from a single RGBD camera placed egocentrically on the wrist of the arm. This approach allows fast training of the real-world manipulation neural policy while not requiring scene reconstruction at test time, but solely a point cloud acquired from a single RGBD camera. Our resulting policy first predicts the desired point to choose from the given point cloud and then the correct displacement to achieve a smoothed cloth. Experimentally, we first assess our results in a simulation environment by comparing them with an existing heuristic policy, as well as several baseline attention architectures. Then, we validate the performance of our approach in a real-world scenario. Project website: link

Programming manipulation behaviors can become increasingly difficult with a growing number and complexity of manipulation tasks, particularly in a dynamic and unstructured environment. Recent progress in unsupervised skill discovery algorithms has shown great promise in learning an extensive collection of behaviors without extrinsic supervision. On the other hand, safety is one of the most critical factors for real- world robot applications. As skill discovery methods typically encourage exploratory and dynamic behaviors, it can often be the case that a large portion of learned skills remain too dangerous and unsafe. In this paper, we introduce the novel problem of Safety-Aware Skill Discovery, which aims to learn, in a task-agnostic fashion, a repertoire of reusable skills that are inherently safe to be composed for solving downstream tasks. We present a computationally tractable algorithm that learns a latent-conditioned skill policy that maximizes intrinsic rewards regularized with a safety-critic that can model any user-defined safety constraints. Using the pretrained safe skill repertoire, hierarchical reinforcement learning can solve multiple downstream tasks without the need for explicit consideration of safety during training and testing. We evaluate our algorithm on a collection of force-controlled robotic manipulation tasks in simulation and show promising downstream task performance while satisfying safety constraints.

Fine-tuning a large language model on downstream tasks has become a commonly adopted process in the Natural Language Processing (NLP) (Wang et al., 2019).However, such a process, when associated with the current transformer-based (Vaswani et al., 2017) architectures, shows several limitations when the target task requires to reason with long documents.In this work, we introduce a novel hierarchical propagation layer that spreads information between multiple transformer windows.We adopt a hierarchical approach where the input is divided in multiple blocks independently processed by the scaled dot-attentions and combined between the successive layers.We validate the effectiveness of our approach on three extractive summarization corpora of long scientific papers and news articles.We compare our approach to standard and pre-trained language-model-based summarizers and report state-of-the-art results for long document summarization and comparable results for smaller document summarization.

In few-shot imitation, an agent is given a few demonstrations of a previously unseen task, and must then successfully perform that task. We propose a novel approach to learning few-shot-imitation agents that we call demonstration-conditioned reinforcement learning (DCRL). Given a training set consisting of demonstrations, reward functions and transition distributions for multiple tasks, the idea is to work with a policy that takes demonstrations as input, and to train this policy to maximize the average of the cumulative reward over the set of training tasks. Relative to previously proposed few-shot imitation methods that use behaviour cloning or infer reward functions from demonstrations, our method has the disadvantage that it requires reward functions at training time. However, DCRL also has several advantages, such as the ability to improve upon suboptimal demonstrations, to operate given state-only demonstrations, and to cope with a domain shift between the demonstrator and the agent. Moreover, we show that DCRL outperforms methods based on behaviour cloning by a large margin, on navigation tasks and on robotic manipulation tasks from the Meta-World benchmark.

Validating the kinematic feasibility of a planned robot motion and finding corresponding inverse solutions are time-consuming processes, especially for long-horizon manipulation tasks. Most existing approaches are based on solving iterative gradient-based optimization, so the processes are time-consuming and have a high risk of falling in local minima. In this work, we propose a unified framework to learn a kinematic feasibility model and a one-shot inverse mapping model for a redundant robot manipulator. Once they are trained, the models can compute the kinematic reachability of a target pose and its inverse solutions without iterative process. We validate our approach using a 7-DOF robot arm with an object grasping application.

Abstract Distributional semantic word representations are at the basis of most modern NLP systems. Their usefulness has been proven across various tasks, particularly as inputs to deep learning models. Beyond that, much work investigated fine-tuning the generic word embeddings to leverage linguistic knowledge from large lexical resources. Some work investigated context-dependent word token embeddings motivated by word sense disambiguation, using sequential context and large lexical resources. More recently, acknowledging the need for an in-context representation of words, some work leveraged information derived from language modelling and large amounts of data to induce contextualised representations. In this paper, we investigate Syntax-Aware word Token Embeddings (SATokE) as a way to explicitly encode specific information derived from the linguistic analysis of a sentence in vectors which are input to a deep learning model. We propose an efficient unsupervised learning algorithm based on tensor factorisation for computing these token embeddings given an arbitrary graph of linguistic structure. Applying this method to syntactic dependency structures, we investigate the usefulness of such token representations as part of deep learning models of text understanding. We encode a sentence either by learning embeddings for its tokens and the relations between them from scratch or by leveraging pre-trained relation embeddings to infer token representations. Given sufficient data, the former is slightly more accurate than the latter, yet both provide more informative token embeddings than standard word representations, even when the word representations have been learned on the same type of context from larger corpora (namely pre-trained dependency-based word embeddings). We use a large set of supervised tasks and two major deep learning families of models for sentence understanding to evaluate our proposal. We empirically demonstrate the superiority of the token representations compared to popular distributional representations of words for various sentence and sentence pair classification tasks.

There have been many attempts at automatically recognising author personality traits from text, typically incorporating linguistic features with conventional machine learning models, e.g.linear regression or Support Vector Machines.In this work, we propose to use deep-learningbased models with atomic features of text -the characters -to build hierarchical, vectorial word and sentence representations for the task of trait inference.On a corpus of tweets, this method shows stateof-the-art performance across five traits and three languages (English, Spanish and Italian) compared with prior work in author profiling.The results, supported by preliminary visualisation work, are encouraging for the ability to detect complex human traits.

Machine reading using differentiable reasoning models has recently shown remarkable progress.In this context, End-to-End trainable Memory Networks (MemN2N) have demonstrated promising performance on simple natural language based reasoning tasks such as factual reasoning and basic deduction.However, other tasks, namely multi-fact questionanswering, positional reasoning or dialog related tasks, remain challenging particularly due to the necessity of more complex interactions between the memory and controller modules composing this family of models.In this paper, we introduce a novel end-to-end memory access regulation mechanism inspired by the current progress on the connection short-cutting principle in the field of computer vision.Concretely, we develop a Gated End-to-End trainable Memory Network architecture (GMemN2N).From the machine learning perspective, this new capability is learned in an end-to-end fashion without the use of any additional supervision signal which is, as far as our knowledge goes, the first of its kind.Our experiments show significant improvements on the most challenging tasks in the 20 bAbI dataset, without the use of any domain knowledge.Then, we show improvements on the Dialog bAbI tasks including the real human-bot conversion-based Dialog State Tracking Challenge (DSTC-2) dataset.On these two datasets, our model sets the new state of the art.

In an end-to-end dialog system, the aim of dialog state tracking is to accurately estimate a compact representation of the current dialog status from a sequence of noisy observations produced by the speech recognition and the natural language understanding modules.This paper introduces a novel method of dialog state tracking based on the general paradigm of machine reading and proposes to solve it using an End-to-End Memory Network, MemN2N, a memory-enhanced neural network architecture.We evaluate the proposed approach on the second Dialog State Tracking Challenge (DSTC-2) dataset.The corpus has been converted for the occasion in order to frame the hidden state variable inference as a questionanswering task based on a sequence of utterances extracted from a dialog.We show that the proposed tracker gives encouraging results.Then, we propose to extend the DSTC-2 dataset and the definition of this dialog state task with specific reasoning capabilities like counting, list maintenance, yes-no question answering and indefinite knowledge management.Finally, we present encouraging results using our proposed MemN2N based tracking model.

Language use is known to be influenced by personality traits as well as by sociodemographic characteristics such as age or mother tongue.As a result, it is possible to automatically identify these traits of the author from her texts.It has recently been shown that knowledge of such dimensions can improve performance in NLP tasks such as topic and sentiment modeling.We posit that machine translation is another application that should be personalized.In order to motivate this, we explore whether translation preserves demographic and psychometric traits.We show that, largely, both translation of the source training data into the target language, and the target test data into the source language has a detrimental effect on the accuracy of predicting author traits.We argue that this supports the need for personal and personality-aware machine translation models.

As current server capacity and network bandwidth become increasingly overloaded by the rapid growth of high quality emerging multimedia services such as mobile online gaming, social networking or IPTV, a critical factor of success of these multimedia services becomes the end-user perception of quality while them using the service. As a result, user-centered approaches that consider quality of experience (QoE) constitute the current design trend for network systems of content providers and network operators. A content distribution network (CDN) that replicates the content from original servers to the replicated servers close to end users is actually an effective solution to improve network quality. We propose a QoE-based server selection algorithm in the context of a CDN architecture. Using realistic characteristics of the server selection process, we formalize our selection model as a sequential decision problem solved by the multi-armed bandit (MAB) paradigm. By using realistic experiments, we demonstrate that our approach yields significant improvements in term of user perception compared to traditional methods (such as Fastest, Closest and Round Robin).

A main objective of an Information Centric Network (ICN) is to improve the network by placing the knowledge in center of the network design. This vision of the network needs an efficient distributed and decentralized knowledge plane. So, an important amount of knowledge should be disseminated over the supervised network, which remains an open problem. Indeed, the dissemination infrastructure must be able to ensure the transport of all information types including knowledge information, throughout the network, and guarantee its freshness. Another crucial aspect of the problem is related to the network robustness with respect to network failures. In this paper, we propose a new model of knowledge dissemination based on super peers architecture. We formalize the super peer selection problem as a K-medoids clustering task. Furthermore, to handle the dynamicity of the network and especially the changes of the end-user network topology, we improved the selection mechanism by adding an adaptive mechanism based on the Page-Hinkley statistical test. Experimental results show that the proposed approach significantly improves performances compared to other current approaches.

The Affinity Propagation (AP) clustering algorithm proposed by Frey and Dueck (2007) provides an understandable, nearly optimal summary of a dataset, albeit with quadratic computational complexity. This paper, motivated by Autonomic Computing, extends AP to the data streaming framework. Firstly a hierarchical strategy is used to reduce the complexity to O(N1+ε); the distortion loss incurred is analyzed in relation with the dimension of the data items. Secondly, a coupling with a change detection test is used to cope with non-stationary data distribution, and rebuild the model as needed. The presented approach StrAP is applied to the stream of jobs submitted to the EGEE Grid, providing an understandable description of the job flow and enabling the system administrator to spot online some sources of failures.

Large scale production grids are a major case for autonomic computing. Following the classical definition of Kephart, an autonomic computing system should optimize its own behavior in accordance with high level guidance from humans. This central tenet of this paper is that the combination of utility functions and reinforcement learning (RL) can provide a general and efficient method for dynamically allocating grid resources in order to optimize the satisfaction of both end-users and participating institutions. The flexibility of an RL-based system allows to model the state of the grid,the jobs to be scheduled, and the high-level objectives of the various actors on the grid. RL-based scheduling can seamlessly adapt its decisions to changes in the distributions ofinter-arrival time, QoS requirements, and resource availability. Moreover, it requires minimal prior knowledge about thetarget environment, including user requests and infrastructure. Our experimental results, both on a synthetic workloadand a real trace, show that RL is not only a realistic alternative to empirical scheduler design, but is able to outperform them.