Search

Model pruning is attracting increasing interest because of its positive implications in terms of resource consumption and costs. A variety of methods have been developed in the past years. In particular, structured pruning techniques discern the importance of nodes in neural networks (NNs) and filters in convolutional neural networks (CNNs). Global versions of these rank all nodes in a network and select the top-k, offering an advantage over local methods that rank nodes only within individual layers. By evaluating all nodes simultaneously, global techniques provide greater control over the network architecture, which improves performance. However, the ranking and selecting process carried out during global pruning can have several major drawbacks. First, the ranking is not updated in real time based on the pruning already performed, making it unable to account for inter-node interactions. Second, it is not uncommon for whole layers to be removed from a model, which leads to untrainable networks. Lastly, global pruning methods do not offer any guarantees regarding re-training. In order to address these issues, we introduce Mutual Information Preserving Pruning (MIPP). The fundamental principle of our method is to select nodes such that the mutual information (MI) between the activations of adjacent layers is maintained. We evaluate MIPP on an array of vision models and datasets, including a pre-trained ResNet50 on ImageNet, where we demonstrate MIPP's ability to outperform state-of-the-art methods. The implementation of MIPP will be made available upon publication.

We introduce a machine learning approach to model checking temporal logic, with application to formal hardware verification. Model checking answers the question of whether every execution of a given system satisfies a desired temporal logic specification. Unlike testing, model checking provides formal guarantees. Its application is expected standard in silicon design and the EDA industry has invested decades into the development of performant symbolic model checking algorithms. Our new approach combines machine learning and symbolic reasoning by using neural networks as formal proof certificates for linear temporal logic. We train our neural certificates from randomly generated executions of the system and we then symbolically check their validity using satisfiability solving which, upon the affirmative answer, establishes that the system provably satisfies the specification. We leverage the expressive power of neural networks to represent proof certificates as well as the fact that checking a certificate is much simpler than finding one. As a result, our machine learning procedure for model checking is entirely unsupervised, formally sound, and practically effective. We experimentally demonstrate that our method outperforms the state-of-the-art academic and commercial model checkers on a set of standard hardware designs written in SystemVerilog., Comment: To appear in NeurIPS 2024

Understanding the complex dynamics of high-dimensional, contingent, and strongly nonlinear economic data, often shaped by multiplicative processes, poses significant challenges for traditional regression methods as such methods offer limited capacity to capture the structural changes they feature. To address this, we propose leveraging the potential of knowledge graph embeddings for economic trade data, in particular, to predict international trade relationships. We implement KonecoKG, a knowledge graph representation of economic trade data with multidimensional relationships using SDM-RDFizer, and transform the relationships into a knowledge graph embedding using AmpliGraph.

In the last decade, the generalization and adaptation abilities of deep learning models were typically evaluated on fixed training and test distributions. Contrary to traditional deep learning, large language models (LLMs) are (i) even more overparameterized, (ii) trained on unlabeled text corpora curated from the Internet with minimal human intervention, and (iii) trained in an online fashion. These stark contrasts prevent researchers from transferring lessons learned on model generalization and adaptation in deep learning contexts to LLMs. To this end, our short paper introduces empirical observations that aim to shed light on further training of already pretrained language models. Specifically, we demonstrate that training a model on a text domain could degrade its perplexity on the test portion of the same domain. We observe with our subsequent analysis that the performance degradation is positively correlated with the similarity between the additional and the original pretraining dataset of the LLM. Our further token-level perplexity observations reveals that the perplexity degradation is due to a handful of tokens that are not informative about the domain. We hope these findings will guide us in determining when to adapt a model vs when to rely on its foundational capabilities., Comment: Accepted to EMNLP 2024 Main Conference

We propose a novel approach for humming transcription that combines a CNN-based architecture with a dynamic programming-based post-processing algorithm, utilizing the recently introduced HumTrans dataset. We identify and address inherent problems with the offset and onset ground truth provided by the dataset, offering heuristics to improve these annotations, resulting in a dataset with precise annotations that will aid future research. Additionally, we compare the transcription accuracy of our method against several others, demonstrating state-of-the-art (SOTA) results. All our code and corrected dataset is available at https://github.com/shubham-gupta-30/humming_transcription

Decision-making agents based on pre-trained Large Language Models (LLMs) are increasingly being deployed across various domains of human activity. While their applications are currently rather specialized, several research efforts are under way to develop more generalist agents. As LLM-based systems become more agentic, their influence on human activity will grow and the transparency of this will decrease. Consequently, developing effective methods for aligning them to human values is vital. The prevailing practice in alignment often relies on human preference data (e.g., in RLHF or DPO), in which values are implicit and are essentially deduced from relative preferences over different model outputs. In this work, instead of relying on human feedback, we introduce the design of reward functions that explicitly encode core human values for Reinforcement Learning-based fine-tuning of foundation agent models. Specifically, we use intrinsic rewards for the moral alignment of LLM agents. We evaluate our approach using the traditional philosophical frameworks of Deontological Ethics and Utilitarianism, quantifying moral rewards for agents in terms of actions and consequences on the Iterated Prisoner's Dilemma (IPD) environment. We also show how moral fine-tuning can be deployed to enable an agent to unlearn a previously developed selfish strategy. Finally, we find that certain moral strategies learned on the IPD game generalize to several other matrix game environments. In summary, we demonstrate that fine-tuning with intrinsic rewards is a promising general solution for aligning LLM agents to human values, and it might represent a more transparent and cost-effective alternative to currently predominant alignment techniques.

In audio and speech processing, tasks usually focus on either the audio or speech modality, even when both sounds and human speech are present in the same audio clip. Recent Auditory Large Language Models (ALLMs) have made it possible to process audio and speech simultaneously within a single model, leading to further considerations of joint audio-speech tasks. In this paper, we investigate how well ALLMs can perform joint audio-speech processing. Specifically, we introduce Joint Audio-Speech Co-Reasoning (JASCO), a novel task that unifies audio and speech processing, strictly requiring co-reasoning across both modalities. We release a scene-reasoning dataset called "What Are They Doing" and establish a joint audio-speech benchmark to evaluate the joint reasoning capability of popular ALLMs. Additionally, we provide deeper insights into the models' behaviors by analyzing their dependence on each modality., Comment: Submitted to ICASSP 2025

Human activities exhibit a strong correlation between actions and the places where these are performed, such as washing something at a sink. More specifically, in daily living environments we may identify particular locations, hereinafter named activity-centric zones, which may afford a set of homogeneous actions. Their knowledge can serve as a prior to favor vision models to recognize human activities. However, the appearance of these zones is scene-specific, limiting the transferability of this prior information to unfamiliar areas and domains. This problem is particularly relevant in egocentric vision, where the environment takes up most of the image, making it even more difficult to separate the action from the context. In this paper, we discuss the importance of decoupling the domain-specific appearance of activity-centric zones from their universal, domain-agnostic representations, and show how the latter can improve the cross-domain transferability of Egocentric Action Recognition (EAR) models. We validate our solution on the EPIC-Kitchens-100 and Argo1M datasets, Comment: Project webpage: https://gabrielegoletto.github.io/EgoZAR/

Structural network embedding is a crucial step in enabling effective downstream tasks for complex systems that aims to project a network into a lower-dimensional space while preserving similarities among nodes. We introduce a simple and efficient embedding technique based on approximate variants of equitable partitions. The approximation consists in introducing a user-tunable tolerance parameter relaxing the otherwise strict condition for exact equitable partitions that can be hardly found in real-world networks. We exploit a relationship between equitable partitions and equivalence relations for Markov chains and ordinary differential equations to develop a partition refinement algorithm for computing an approximate equitable partition in polynomial time. We compare our method against state-of-the-art embedding techniques on benchmark networks. We report comparable -- when not superior -- performance for visualization, classification, and regression tasks at a cost between one and three orders of magnitude smaller using a prototype implementation, enabling the embedding of large-scale networks which could not be efficiently handled by most of the competing techniques., Comment: Accepted at ICDM 2024

Neural networks are typically black-boxes that remain opaque with regards to their decision mechanisms. Several works in the literature have proposed post-hoc explanation methods to alleviate this issue. This paper proposes LMAC-TD, a post-hoc explanation method that trains a decoder to produce explanations directly in the time domain. This methodology builds upon the foundation of L-MAC, Listenable Maps for Audio Classifiers, a method that produces faithful and listenable explanations. We incorporate SepFormer, a popular transformer-based time-domain source separation architecture. We show through a user study that LMAC-TD significantly improves the audio quality of the produced explanations while not sacrificing from faithfulness., Comment: The first two authors contributed equally to this research. Author order is alphabetical

Graph path search is a classic computer science problem that has been recently approached with Reinforcement Learning (RL) due to its potential to outperform prior methods. Existing RL techniques typically assume a global view of the network, which is not suitable for large-scale, dynamic, and privacy-sensitive settings. An area of particular interest is search in social networks due to its numerous applications. Inspired by seminal work in experimental sociology, which showed that decentralized yet efficient search is possible in social networks, we frame the problem as a collaborative task between multiple agents equipped with a limited local view of the network. We propose a multi-agent approach for graph path search that successfully leverages both homophily and structural heterogeneity. Our experiments, carried out over synthetic and real-world social networks, demonstrate that our model significantly outperforms learned and heuristic baselines. Furthermore, our results show that meaningful embeddings for graph navigation can be constructed using reward-driven learning.

Large Language Models (LLMs) have shown their ability to improve the performance of speech recognizers by effectively rescoring the n-best hypotheses generated during the beam search process. However, the best way to exploit recent generative instruction-tuned LLMs for hypothesis rescoring is still unclear. This paper proposes a novel method that uses instruction-tuned LLMs to dynamically expand the n-best speech recognition hypotheses with new hypotheses generated through appropriately-prompted LLMs. Specifically, we introduce a new zero-shot method for ASR n-best rescoring, which combines confidence scores, LLM sequence scoring, and prompt-based hypothesis generation. We compare Llama-3-Instruct, GPT-3.5 Turbo, and GPT-4 Turbo as prompt-based generators with Llama-3 as sequence scorer LLM. We evaluated our approach using different speech recognizers and observed significant relative improvement in the word error rate (WER) ranging from 5% to 25%., Comment: IEEE Spoken Language Technology Workshop

The area of study concerning the estimation of spatial sound, i.e., the distribution of a physical quantity of sound such as acoustic pressure, is called sound field estimation, which is the basis for various applied technologies related to spatial audio processing. The sound field estimation problem is formulated as a function interpolation problem in machine learning in a simplified scenario. However, high estimation performance cannot be expected by simply applying general interpolation techniques that rely only on data. The physical properties of sound fields are useful a priori information, and it is considered extremely important to incorporate them into the estimation. In this article, we introduce the fundamentals of physics-informed machine learning (PIML) for sound field estimation and overview current PIML-based sound field estimation methods., Comment: Accepted to IEEE Signal Processing Magazine, Special Issue on Model-based and Data-Driven Audio Signal Processing

Quantum imaging with undetected photons spatially transfers amplitude and phase information from one spectral region of physical interest to another spectral region that is easy to detect. The photon energy of the two spectral regions can, in principle, be separated by several orders of magnitude. However, quantum imaging with undetected photons has so far only been demonstrated in spectral regions of similar order of magnitude in frequency (and for which cameras are commercially available). Here, we demonstrate amplitude- and phase-sensitive imaging in the terahertz spectral region (1.5 THz center frequency) by detecting only visible photons (center wavelength 662.2 nm, 452.7 THz center frequency) more than two orders of magnitude apart. As a result, terahertz spectral information can be reliably detected with a standard CMOS camera without cooling, achieving a spatial resolution close to the wavelength. By taking advantage of quantum distillation in a nonlinear interferometer, the influence of ubiquitous thermal terahertz photons can be neglected. Our results are in good agreement with numerical simulations of the imaging process and demonstrate the huge potential of this method to address otherwise challenging spectral regions where cameras do not exist.

Droplet-confined electrodeposition enables a precise deposition of three dimensional, nanoscopic and high purity metal structures. It aspires to fabricate intricate microelectronic devices, metamaterials, plasmonic structures and functionalized surfaces. Yet, a major handicap of droplet-confined electrodeposition is the current lack of control over the process, which is owed to its dynamic nature and the nanoscopic size of the involved droplets. The deposition current offers itself as an obvious and real-time window into the deposition. Therefore, the current during droplet-confined deposition is analysed. Nucleation and growth dynamics are evaluated systematically. Our results indicate different deposition regimes and link current to both volume and morphology of deposited copper. This allows for optimized electroplating strategies and to calibrate the slicing algorithms necessary for a controlled deposition of 3D structures. The potential of selecting appropriate solvents further readies this novel technique for the reliable deposition of functional structures with submicron resolution.

The proliferation of human-AI ecosystems involving human interaction with algorithms, such as assistants and recommenders, raises concerns about large-scale social behaviour. Despite evidence of such phenomena across several contexts, the collective impact of GPS navigation services remains unclear: while beneficial to the user, they can also cause chaos if too many vehicles are driven through the same few roads. Our study employs a simulation framework to assess navigation services' influence on road network usage and CO2 emissions. The results demonstrate a universal pattern of amplified conformity: increasing adoption rates of navigation services cause a reduction of route diversity of mobile travellers and increased concentration of traffic and emissions on fewer roads, thus exacerbating an unequal distribution of negative externalities on selected neighbourhoods. Although navigation services recommendations can help reduce CO2 emissions when their adoption rate is low, these benefits diminish or even disappear when the adoption rate is high and exceeds a certain city- and service-dependent threshold. We summarize these discoveries in a non-linear function that connects the marginal increase of conformity with the marginal reduction in CO2 emissions. Our simulation approach addresses the challenges posed by the complexity of transportation systems and the lack of data and algorithmic transparency.

Spherical microphone arrays are convenient tools for capturing the spatial characteristics of a sound field. However, achieving superior spatial resolution requires arrays with numerous capsules, consequently leading to expensive devices. To address this issue, we present a method for spatially upsampling spherical microphone arrays with a limited number of capsules. Our approach exploits a physics-informed neural network with Rowdy activation functions, leveraging physical constraints to provide high-order microphone array signals, starting from low-order devices. Results show that, within its domain of application, our approach outperforms a state of the art method based on signal processing for spherical microphone arrays upsampling., Comment: Accepted for publication at IWAENC 2024

We explore the optical dipole orientation of single photon emitters in monolayer MoS2 as produced by a focused helium ion beam. The single photon emitters can be understood as single sulfur vacancies. The corresponding far-field luminescence spectra reveal several photoluminescence lines below the dominating luminescence of the exciton in MoS2. These sub-bandgap emission lines were predicted by ab initio theory, but they have never been resolved in luminescence experiments because of their small amplitude. We reveal the lines by their dependence as a function of the photon energy and momentum as measured in the back focal plane of the optical circuitry. The agreement between theory and experiment suggests that the defect states interact strongly within the Brillouin zone.

The training process of foundation models as for other classes of deep learning systems is based on minimizing the reconstruction error over a training set. For this reason, they are susceptible to the memorization and subsequent reproduction of training samples. In this paper, we introduce a training-as-compressing perspective, wherein the model's weights embody a compressed representation of the training data. From a copyright standpoint, this point of view implies that the weights could be considered a reproduction or a derivative work of a potentially protected set of works. We investigate the technical and legal challenges that emerge from this framing of the copyright of outputs generated by foundation models, including their implications for practitioners and researchers. We demonstrate that adopting an information-centric approach to the problem presents a promising pathway for tackling these emerging complex legal issues., Comment: Spotlight presentation at GenLaw'24, see https://www.genlaw.org/2024-icml-papers#training-foundation-models-as-data-compression-on-information-model-weights-and-copyright-law

How can a scientist use a Reinforcement Learning (RL) algorithm to design experiments over a dynamical system's state space? In the case of finite and Markovian systems, an area called Active Exploration (AE) relaxes the optimization problem of experiments design into Convex RL, a generalization of RL admitting a wider notion of reward. Unfortunately, this framework is currently not scalable and the potential of AE is hindered by the vastness of experiment spaces typical of scientific discovery applications. However, these spaces are often endowed with natural geometries, e.g., permutation invariance in molecular design, that an agent could leverage to improve the statistical and computational efficiency of AE. To achieve this, we bridge AE and MDP homomorphisms, which offer a way to exploit known geometric structures via abstraction. Towards this goal, we make two fundamental contributions: we extend MDP homomorphisms formalism to Convex RL, and we present, to the best of our knowledge, the first analysis that formally captures the benefit of abstraction via homomorphisms on sample efficiency. Ultimately, we propose the Geometric Active Exploration (GAE) algorithm, which we analyse theoretically and experimentally in environments motivated by problems in scientific discovery., Comment: ICML 2024

Assessing an athlete's performance in canoe sprint is often established by measuring a variety of kinematic parameters during training sessions. Many of these parameters are related to single or multiple paddle stroke cycles. Determining on- and offset of these cycles in force sensor signals is usually not straightforward and requires human interaction. This paper explores convolutional neural networks (CNNs) and recurrent neural networks (RNNs) in terms of their ability to automatically predict these events. In addition, our work proposes an extension to the recently published SoftED metric for event detection in order to properly assess the model performance on time windows. In our results, an RNN based on bidirectional gated recurrent units (BGRUs) turned out to be the most suitable model for paddle stroke detection.

Recommendation systems and assistants (in short, recommenders) are ubiquitous in online platforms and influence most actions of our day-to-day lives, suggesting items or providing solutions based on users' preferences or requests. This survey analyses the impact of recommenders in four human-AI ecosystems: social media, online retail, urban mapping and generative AI ecosystems. Its scope is to systematise a fast-growing field in which terminologies employed to classify methodologies and outcomes are fragmented and unsystematic. We follow the customary steps of qualitative systematic review, gathering 144 articles from different disciplines to develop a parsimonious taxonomy of: methodologies employed (empirical, simulation, observational, controlled), outcomes observed (concentration, model collapse, diversity, echo chamber, filter bubble, inequality, polarisation, radicalisation, volume), and their level of analysis (individual, item, model, and systemic). We systematically discuss all findings of our survey substantively and methodologically, highlighting also potential avenues for future research. This survey is addressed to scholars and practitioners interested in different human-AI ecosystems, policymakers and institutional stakeholders who want to understand better the measurable outcomes of recommenders, and tech companies who wish to obtain a systematic view of the impact of their recommenders.

SpeechBrain is an open-source Conversational AI toolkit based on PyTorch, focused particularly on speech processing tasks such as speech recognition, speech enhancement, speaker recognition, text-to-speech, and much more. It promotes transparency and replicability by releasing both the pre-trained models and the complete "recipes" of code and algorithms required for training them. This paper presents SpeechBrain 1.0, a significant milestone in the evolution of the toolkit, which now has over 200 recipes for speech, audio, and language processing tasks, and more than 100 models available on Hugging Face. SpeechBrain 1.0 introduces new technologies to support diverse learning modalities, Large Language Model (LLM) integration, and advanced decoding strategies, along with novel models, tasks, and modalities. It also includes a new benchmark repository, offering researchers a unified platform for evaluating models across diverse tasks., Comment: Accepted to the Journal of Machine Learning research (JMLR), Machine Learning Open Source Software

Consumer IP cameras are now the most widely adopted solution for remote monitoring in various contexts, such as private homes or small offices. While the security of these devices has been scrutinized, most approaches are limited to relatively shallow network-based analyses. In this paper, we discuss a methodology for the security analysis and identification of remotely exploitable vulnerabilities in IP cameras, which includes static and dynamic analyses of executables extracted from IP camera firmware. Compared to existing methodologies, our approach leverages the context of the target device to focus on the identification of malicious invocation sequences that could lead to exploitable vulnerabilities. We demonstrate the application of our methodology by using the Tenda CP3 IP camera as a case study. We identified five novel CVEs, with CVSS scores ranging from 7.5 to 9.8. To partially automate our analysis, we also developed a custom tool based on Ghidra and rhabdomancer., Comment: 8 pages, 3 figures, accepted for publication at the 19th International Workshop on Security (IWSEC 2024)

Discrete audio tokens have recently gained considerable attention for their potential to connect audio and language processing, enabling the creation of modern multimodal large language models. Ideal audio tokens must effectively preserve phonetic and semantic content along with paralinguistic information, speaker identity, and other details. While several types of audio tokens have been recently proposed, identifying the optimal tokenizer for various tasks is challenging due to the inconsistent evaluation settings in existing studies. To address this gap, we release the Discrete Audio and Speech Benchmark (DASB), a comprehensive leaderboard for benchmarking discrete audio tokens across a wide range of discriminative tasks, including speech recognition, speaker identification and verification, emotion recognition, keyword spotting, and intent classification, as well as generative tasks such as speech enhancement, separation, and text-to-speech. Our results show that, on average, semantic tokens outperform compression tokens across most discriminative and generative tasks. However, the performance gap between semantic tokens and standard continuous representations remains substantial, highlighting the need for further research in this field., Comment: 9 pages, 5 tables

The problem of pure exploration in Markov decision processes has been cast as maximizing the entropy over the state distribution induced by the agent's policy, an objective that has been extensively studied. However, little attention has been dedicated to state entropy maximization under partial observability, despite the latter being ubiquitous in applications, e.g., finance and robotics, in which the agent only receives noisy observations of the true state governing the system's dynamics. How can we address state entropy maximization in those domains? In this paper, we study the simple approach of maximizing the entropy over observations in place of true latent states. First, we provide lower and upper bounds to the approximation of the true state entropy that only depends on some properties of the observation function. Then, we show how knowledge of the latter can be exploited to compute a principled regularization of the observation entropy to improve performance. With this work, we provide both a flexible approach to bring advances in state entropy maximization to the POMDP setting and a theoretical characterization of its intrinsic limits.

Discrete audio tokens have recently gained attention for their potential to bridge the gap between audio and language processing. Ideal audio tokens must preserve content, paralinguistic elements, speaker identity, and many other audio details. Current audio tokenization methods fall into two categories: Semantic tokens, acquired through quantization of Self-Supervised Learning (SSL) models, and Neural compression-based tokens (codecs). Although previous studies have benchmarked codec models to identify optimal configurations, the ideal setup for quantizing pretrained SSL models remains unclear. This paper explores the optimal configuration of semantic tokens across discriminative and generative tasks. We propose a scalable solution to train a universal vocoder across multiple SSL layers. Furthermore, an attention mechanism is employed to identify task-specific influential layers, enhancing the adaptability and performance of semantic tokens in diverse audio applications., Comment: 4 pages, 2 figures, 2 tables, Accepted at Interspeech 2024

In this paper, we propose Phoneme Discretized Saliency Maps (PDSM), a discretization algorithm for saliency maps that takes advantage of phoneme boundaries for explainable detection of AI-generated voice. We experimentally show with two different Text-to-Speech systems (i.e., Tacotron2 and Fastspeech2) that the proposed algorithm produces saliency maps that result in more faithful explanations compared to standard posthoc explanation methods. Moreover, by associating the saliency maps to the phoneme representations, this methodology generates explanations that tend to be more understandable than standard saliency maps on magnitude spectrograms., Comment: Proc. Interspeech 2024, 3295-3299, doi: 10.21437/Interspeech.2024-632

Polarons play a major role in the description of optical, electrical and dielectrical properties of several ferroelectric oxides. The motion of those particles occur by elementary hops among the material lattice sites. In order to compute macroscopic transport parameters such as charge mobility, normal diffusion laws are generally assumed. In this paper we show that when defect states able to trap the polarons for long times are considered, significant deviations from the normal diffusion behaviour arise. As an example of this behavior, we consider here the case of lithium niobate (LN), a prototypical system, having interacting polaron types. Our analysis considers the case of a stoichiometric LN containing a certain concentration of small electron polarons hopping on regular Nb sites and compares it to the material in congruent composition, which is characterized by a large concentration of antisite defects. While in the first case the charge carriers are free polarons hopping on a regular Nb sublattice, in the second case a fraction of polarons is trapped on antisite defects. Thus a range of different hopping possibilities arises, depending on the type of starting and destination sites. We develop a formalism encompassing all these microscopic processes in the framework of a switching diffusion model which can be well approximated by a mobile-immobile transport model providing explicit expressions for the polaron mobility. Starting from the Marcus-Holstein model for the polaron hopping frequency we verify by means of a Monte Carlo approach the diffusion/mobility of the different polarons showing that, while free polarons obey the laws for normal diffusion as expected, bound polarons follow an anomalous diffusion behaviour and that in the case of the congruent crystal where mixed free and bound polaron transport is involved, our expressions indeed provide a satisfactory description.

This paper introduces the Precision-Timed Virtual Machine (PretVM), an intermediate platform facilitating the execution of quasi-static schedules compiled from a subset of programs written in the Lingua Franca (LF) coordination language. The subset consists of those programs that in principle should have statically verifiable and predictable timing behavior. The PretVM provides a schedule with well-defined worst-case timing bounds. The PretVM provides a clean separation between application logic and coordination logic, yielding more analyzable program executions. Experiments compare the PretVM against the default (more dynamic) LF scheduler and show that it delivers time-accurate deterministic execution.

In online Inverse Reinforcement Learning (IRL), the learner can collect samples about the dynamics of the environment to improve its estimate of the reward function. Since IRL suffers from identifiability issues, many theoretical works on online IRL focus on estimating the entire set of rewards that explain the demonstrations, named the feasible reward set. However, none of the algorithms available in the literature can scale to problems with large state spaces. In this paper, we focus on the online IRL problem in Linear Markov Decision Processes (MDPs). We show that the structure offered by Linear MDPs is not sufficient for efficiently estimating the feasible set when the state space is large. As a consequence, we introduce the novel framework of rewards compatibility, which generalizes the notion of feasible set, and we develop CATY-IRL, a sample efficient algorithm whose complexity is independent of the cardinality of the state space in Linear MDPs. When restricted to the tabular setting, we demonstrate that CATY-IRL is minimax optimal up to logarithmic factors. As a by-product, we show that Reward-Free Exploration (RFE) enjoys the same worst-case rate, improving over the state-of-the-art lower bound. Finally, we devise a unifying framework for IRL and RFE that may be of independent interest., Comment: Advances in Neural Information Processing Systems 38 (NeurIPS 2024)

Recent works have studied *state entropy maximization* in reinforcement learning, in which the agent's objective is to learn a policy inducing high entropy over states visitation (Hazan et al., 2019). They typically assume full observability of the state of the system, so that the entropy of the observations is maximized. In practice, the agent may only get *partial* observations, e.g., a robot perceiving the state of a physical space through proximity sensors and cameras. A significant mismatch between the entropy over observations and true states of the system can arise in those settings. In this paper, we address the problem of entropy maximization over the *true states* with a decision policy conditioned on partial observations *only*. The latter is a generalization of POMDPs, which is intractable in general. We develop a memory and computationally efficient *policy gradient* method to address a first-order relaxation of the objective defined on *belief* states, providing various formal characterizations of approximation gaps, the optimization landscape, and the *hallucination* problem. This paper aims to generalize state entropy maximization to more realistic domains that meet the challenges of applications.

Meta reinforcement learning sets a distribution over a set of tasks on which the agent can train at will, then is asked to learn an optimal policy for any test task efficiently. In this paper, we consider a finite set of tasks modeled through Markov decision processes with various dynamics. We assume to have endured a long training phase, from which the set of tasks is perfectly recovered, and we focus on regret minimization against the optimal policy in the unknown test task. Under a separation condition that states the existence of a state-action pair revealing a task against another, Chen et al. (2022) show that $O(M^2 \log(H))$ regret can be achieved, where $M, H$ are the number of tasks in the set and test episodes, respectively. In our first contribution, we demonstrate that the latter rate is nearly optimal by developing a novel lower bound for test-time regret minimization under separation, showing that a linear dependence with $M$ is unavoidable. Then, we present a family of stronger yet reasonable assumptions beyond separation, which we call strong identifiability, enabling algorithms achieving fast rates $\log (H)$ and sublinear dependence with $M$ simultaneously. Our paper provides a new understanding of the statistical barriers of test-time regret minimization and when fast rates can be achieved.

In this paper, we introduce Partial Information Decomposition of Features (PIDF), a new paradigm for simultaneous data interpretability and feature selection. Contrary to traditional methods that assign a single importance value, our approach is based on three metrics per feature: the mutual information shared with the target variable, the feature's contribution to synergistic information, and the amount of this information that is redundant. In particular, we develop a novel procedure based on these three metrics, which reveals not only how features are correlated with the target but also the additional and overlapping information provided by considering them in combination with other features. We extensively evaluate PIDF using both synthetic and real-world data, demonstrating its potential applications and effectiveness, by considering case studies from genetics and neuroscience.

We present for the first time a supermartingale certificate for $\omega$-regular specifications. We leverage the Robbins & Siegmund convergence theorem to characterize supermartingale certificates for the almost-sure acceptance of Streett conditions on general stochastic processes, which we call Streett supermartingales. This enables effective verification and control of discrete-time stochastic dynamical models with infinite state space under $\omega$-regular and linear temporal logic specifications. Our result generalises reachability, safety, reach-avoid, persistence and recurrence specifications; our contribution applies to discrete-time stochastic dynamical models and probabilistic programs with discrete and continuous state spaces and distributions, and carries over to deterministic models and programs. We provide a synthesis algorithm for control policies and Streett supermartingales as proof certificates for $\omega$-regular objectives, which is sound and complete for supermartingales and control policies with polynomial templates and any stochastic dynamical model whose post-expectation is expressible as a polynomial. We additionally provide an optimisation of our algorithm that reduces the problem to satisfiability modulo theories, under the assumption that templates and post-expectation are in piecewise linear form. We have built a prototype and have demonstrated the efficacy of our approach on several exemplar $\omega$-regular verification and control synthesis problems., Comment: The conference version of this manuscript appeared at CAV'24

Interpreting the decisions of deep learning models, including audio classifiers, is crucial for ensuring the transparency and trustworthiness of this technology. In this paper, we introduce LMAC-ZS (Listenable Maps for Audio Classifiers in the Zero-Shot context), which, to the best of our knowledge, is the first decoder-based post-hoc interpretation method for explaining the decisions of zero-shot audio classifiers. The proposed method utilizes a novel loss function that maximizes the faithfulness to the original similarity between a given text-and-audio pair. We provide an extensive evaluation using the Contrastive Language-Audio Pretraining (CLAP) model to showcase that our interpreter remains faithful to the decisions in a zero-shot classification context. Moreover, we qualitatively show that our method produces meaningful explanations that correlate well with different text prompts.

Burst effect of Silicon Photomultiplier (SiPM) at cryogenic temperatures have been discovered few years ago looking at the dark count rate of SiPMs at liquid nitrogen temperatures. Bursts are trains of single signals that happen randomly and are clearly distinguishable from the primary DCR and correlated noise because of their particular time distribution. In this article we describe a detailed study related to both the external causes that triggers bursts and to the phenomenon, internal to the sensor, that produces this dark signals. We related the burst occurrence to the luminescence produced by some trapping centers in the SiPMs when they are excited by ionizing radiation that impinges on the sensor., Comment: 10 pages, 4 figures

We introduce a data-driven approach to computing finite bisimulations for state transition systems with very large, possibly infinite state space. Our novel technique computes stutter-insensitive bisimulations of deterministic systems, which we characterize as the problem of learning a state classifier together with a ranking function for each class. Our procedure learns a candidate state classifier and candidate ranking functions from a finite dataset of sample states; then, it checks whether these generalise to the entire state space using satisfiability modulo theory solving. Upon the affirmative answer, the procedure concludes that the classifier constitutes a valid stutter-insensitive bisimulation of the system. Upon a negative answer, the solver produces a counterexample state for which the classifier violates the claim, adds it to the dataset, and repeats learning and checking in a counterexample-guided inductive synthesis loop until a valid bisimulation is found. We demonstrate on a range of benchmarks from reactive verification and software model checking that our method yields faster verification results than alternative state-of-the-art tools in practice. Our method produces succinct abstractions that enable an effective verification of linear temporal logic without next operator, and are interpretable for system diagnostics.

Causal structure discovery from observations can be improved by integrating background knowledge provided by an expert to reduce the hypothesis space. Recently, Large Language Models (LLMs) have begun to be considered as sources of prior information given the low cost of querying them relative to a human expert. In this work, firstly, we propose a set of metrics for assessing LLM judgments for causal graph discovery independently of the downstream algorithm. Secondly, we systematically study a set of prompting designs that allows the model to specify priors about the structure of the causal graph. Finally, we present a general methodology for the integration of LLM priors in graph discovery algorithms, finding that they help improve performance on common-sense benchmarks and especially when used for assessing edge directionality. Our work highlights the potential as well as the shortcomings of the use of LLMs in this problem space.

The smart grid represents a pivotal innovation in modernizing the electricity sector, offering an intelligent, digitalized energy network capable of optimizing energy delivery from source to consumer. It hence represents the backbone of the energy sector of a nation. Due to its central role, the availability of the smart grid is paramount and is hence necessary to have in-depth control of its operations and safety. To this aim, researchers developed multiple solutions to assess the smart grid's stability and guarantee that it operates in a safe state. Artificial intelligence and Machine learning algorithms have proven to be effective measures to accurately predict the smart grid's stability. Despite the presence of known adversarial attacks and potential solutions, currently, there exists no standardized measure to protect smart grids against this threat, leaving them open to new adversarial attacks. In this paper, we propose GAN-GRID a novel adversarial attack targeting the stability prediction system of a smart grid tailored to real-world constraints. Our findings reveal that an adversary armed solely with the stability model's output, devoid of data or model knowledge, can craft data classified as stable with an Attack Success Rate (ASR) of 0.99. Also by manipulating authentic data and sensor values, the attacker can amplify grid issues, potentially undetected due to a compromised stability prediction system. These results underscore the imperative of fortifying smart grid security mechanisms against adversarial manipulation to uphold system stability and reliability.

Traditional recommendation proposals, including content-based and collaborative filtering, usually focus on similarity between items or users. Existing approaches lack ways of introducing unexpectedness into recommendations, prioritizing globally popular items over exposing users to unforeseen items. This investigation aims to design and evaluate a novel layer on top of recommender systems suited to incorporate relational information and suggest items with a user-defined degree of surprise. We propose a Knowledge Graph (KG) based recommender system by encoding user interactions on item catalogs. Our study explores whether network-level metrics on KGs can influence the degree of surprise in recommendations. We hypothesize that surprisingness correlates with certain network metrics, treating user profiles as subgraphs within a larger catalog KG. The achieved solution reranks recommendations based on their impact on structural graph metrics. Our research contributes to optimizing recommendations to reflect the metrics. We experimentally evaluate our approach on two datasets of LastFM listening histories and synthetic Netflix viewing profiles. We find that reranking items based on complex network metrics leads to a more unexpected and surprising composition of recommendation lists.

In this paper, we introduce the design of HackCar, a testing platform for replicating attacks and defenses on a generic automotive system without requiring access to a complete vehicle. This platform empowers security researchers to illustrate the consequences of attacks targeting an automotive system on a realistic platform, facilitating the development and testing of security countermeasures against both existing and novel attacks. The HackCar platform is built upon an F1-10th model, to which various automotive-grade microcontrollers are connected through automotive communication protocols. This solution is crafted to be entirely modular, allowing for the creation of diverse test scenarios. Researchers and practitioners can thus develop innovative security solutions while adhering to the constraints of automotive-grade microcontrollers. We showcase our design by comparing it with a real, licensed, and unmodified vehicle. Additionally, we analyze the behavior of the HackCar in both an attack-free scenario and a scenario where an attack on in-vehicle communication is deployed., Comment: 7 pages, 2 figures, accepted for publication at the 35th IEEE Intelligent Vehicles Symposium (IEEE IV 2024)

Large language models are revolutionizing several areas, including artificial creativity. However, the process of generation in machines profoundly diverges from that observed in humans. In particular, machine generation is characterized by a lack of intentionality and an underlying creative process. We propose a method called Creative Beam Search that uses Diverse Beam Search and LLM-as-a-Judge to perform response generation and response validation. The results of a qualitative experiment show how our approach can provide better output than standard sampling techniques. We also show that the response validation step is a necessary complement to the response generation step., Comment: Presented as a short paper at the 15th International Conference on Computational Creativity (ICCC'24)

The widespread exchange of digital documents in various domains has resulted in abundant private information being shared. This proliferation necessitates redaction techniques to protect sensitive content and user privacy. While numerous redaction methods exist, their effectiveness varies, with some proving more robust than others. As such, the literature proposes several deanonymization techniques, raising awareness of potential privacy threats. However, while none of these methods are successful against the most effective redaction techniques, these attacks only focus on the anonymized tokens and ignore the sentence context. In this paper, we propose RedactBuster, the first deanonymization model using sentence context to perform Named Entity Recognition on reacted text. Our methodology leverages fine-tuned state-of-the-art Transformers and Deep Learning models to determine the anonymized entity types in a document. We test RedactBuster against the most effective redaction technique and evaluate it using the publicly available Text Anonymization Benchmark (TAB). Our results show accuracy values up to 0.985 regardless of the document nature or entity type. In raising awareness of this privacy issue, we propose a countermeasure we call character evasion that helps strengthen the secrecy of sensitive information. Furthermore, we make our model and testbed open-source to aid researchers and practitioners in evaluating the resilience of novel redaction techniques and enhancing document privacy.

Graphs are a natural representation for systems based on relations between connected entities. Combinatorial optimization problems, which arise when considering an objective function related to a process of interest on discrete structures, are often challenging due to the rapid growth of the solution space. The trial-and-error paradigm of Reinforcement Learning has recently emerged as a promising alternative to traditional methods, such as exact algorithms and (meta)heuristics, for discovering better decision-making strategies in a variety of disciplines including chemistry, computer science, and statistics. Despite the fact that they arose in markedly different fields, these techniques share significant commonalities. Therefore, we set out to synthesize this work in a unifying perspective that we term Graph Reinforcement Learning, interpreting it as a constructive decision-making method for graph problems. After covering the relevant technical background, we review works along the dividing line of whether the goal is to optimize graph structure given a process of interest, or to optimize the outcome of the process itself under fixed graph structure. Finally, we discuss the common challenges facing the field and open research questions. In contrast with other surveys, the present work focuses on non-canonical graph problems for which performant algorithms are typically not known and Reinforcement Learning is able to provide efficient and effective solutions., Comment: To appear in Transactions on Machine Learning Research (TMLR)

At the extreme densities reached in the core of neutron stars, it is possible that quark deconfined matter is produced. The formation of this new phase of strongly interacting matter is likely to occur via a first-order phase transition for the typical temperatures reached in astrophysical processes. The first seeds of quark matter would then form through a process of nucleation within the metastable hadronic phase. Here we address the role of the thermal fluctuations in the hadronic composition on the nucleation of two-flavour quark matter. At finite temperature, thermodynamic quantities in a system fluctuate around average values. Being nucleation a local process, it is possible that it occurs in a subsystem whose composition makes the nucleation easier. We will consider the total probability of the nucleation as the product between the probability that a subsystem has a certain hadronic composition different from the average in the bulk, and the nucleation probability in that subsystem. We will show how those fluctuations of the hadronic composition can increase the efficiency of nucleation already for temperatures $\sim (0.1-1)$ keV. However, for temperatures $\lesssim (1-10)$ MeV, the needed overpressure exceeds the maximum pressure reached in compact stars. Finally, for even larger temperatures the process of nucleation can take place, even taking into account finite size effects., Comment: 16 pages, 9 figures

LiNb$_{1-x}$Ta$_x$O$_3$ solid solutions are investigated from first principles and by optical spectroscopy. The ground- and excited-state properties of the solid solutions are modelled within density functional theory as a function of the Ta concentration using special quasirandom structures spanning the entire composition range between LiNbO$_3$ and LiTaO$_3$. Deviations from a Vegard behavior are predicted for the lattice parameters, the heat capacity, the electronic bandgap, and consequently the absorption edge. The latter is measured for crystals of different composition by low temperature optical spectroscopy, qualitatively confirming the theoretical predictions. The LiNb$_{0.11}$Ta$_{0.89}$O$_3$ composition is found to be a highly unusual crystal with a permanent macroscopic electric polarization and nonetheless zero birefringence.

Predicting and classifying faults in electricity networks is crucial for uninterrupted provision and keeping maintenance costs at a minimum. Thanks to the advancements in the field provided by the smart grid, several data-driven approaches have been proposed in the literature to tackle fault prediction tasks. Implementing these systems brought several improvements, such as optimal energy consumption and quick restoration. Thus, they have become an essential component of the smart grid. However, the robustness and security of these systems against adversarial attacks have not yet been extensively investigated. These attacks can impair the whole grid and cause additional damage to the infrastructure, deceiving fault detection systems and disrupting restoration. In this paper, we present FaultGuard, the first framework for fault type and zone classification resilient to adversarial attacks. To ensure the security of our system, we employ an Anomaly Detection System (ADS) leveraging a novel Generative Adversarial Network training layer to identify attacks. Furthermore, we propose a low-complexity fault prediction model and an online adversarial training technique to enhance robustness. We comprehensively evaluate the framework's performance against various adversarial attacks using the IEEE13-AdvAttack dataset, which constitutes the state-of-the-art for resilient fault prediction benchmarking. Our model outclasses the state-of-the-art even without considering adversaries, with an accuracy of up to 0.958. Furthermore, our ADS shows attack detection capabilities with an accuracy of up to 1.000. Finally, we demonstrate how our novel training layers drastically increase performances across the whole framework, with a mean increase of 154% in ADS accuracy and 118% in model accuracy.

Femto-/nanosecond pulse-induced, red and near-infrared absorption is studied in $\mathrm{LiNb}_{1-x}\mathrm{Ta}_{x}\mathrm{O}_3$ (LNT) solid solutions with the goal to probe the intrinsic defect structure via the formation, transport and recombination of optically generated small bound electron polarons with strong coupling to the lattice. As a result, long-lived transients are uncovered for LNT which exceed lifetimes of LN and LT by a factor of up to 100 over the entire range of investigated compositions. At the same time, the starting amplitude varies in the range of $\alpha_\mathrm{li}^0\approx10-100\,\mathrm{m}^{-1}$ as a function of $x$ and exceed the ones of LN and LT by a factor of up to ten. The results are interpreted in the model of three-dimensional small polaron hopping transport considering the simultaneous presence of three different types of small bound polarons, in particular of small electron $\mathrm{Nb}_\mathrm{Li}^{4+}$ and $\mathrm{Ta}_\mathrm{Li}^{4+}$ antisite polarons, and of small electron $\mathrm{Ta}_\mathrm{V}^{4+}$ interstitial polarons. We conclude that the differences between LNT, LN, and LT may point to model systems that consist of one (LN), two (LT) and three (LNT) intrinsic defect centers for electron localization.

Despite the impressive performance of deep learning models across diverse tasks, their complexity poses challenges for interpretation. This challenge is particularly evident for audio signals, where conveying interpretations becomes inherently difficult. To address this issue, we introduce Listenable Maps for Audio Classifiers (L-MAC), a posthoc interpretation method that generates faithful and listenable interpretations. L-MAC utilizes a decoder on top of a pretrained classifier to generate binary masks that highlight relevant portions of the input audio. We train the decoder with a loss function that maximizes the confidence of the classifier decision on the masked-in portion of the audio while minimizing the probability of model output for the masked-out portion. Quantitative evaluations on both in-domain and out-of-domain data demonstrate that L-MAC consistently produces more faithful interpretations than several gradient and masking-based methodologies. Furthermore, a user study confirms that, on average, users prefer the interpretations generated by the proposed technique., Comment: Accepted to ICML 2024 (Oral)