Search

Motivated by recent experiments on the quantum magnet K$_{2}$Co(SeO$_{3}$)$_{2}$, we study theoretically the excitation spectrum of the nearest-neighbour triangular XXZ model in the limit of strong easy-axis anisotropy, within the up-up-down (1/3-plateau) phase. We make an expansion in $\alpha =J_{xy}/J_{zz}$ instead of $1/S$ and calculate the magnon dispersion for any value of the spin $S$ at second order in $\alpha$, with two important conclusions: (i) the 1/S expansion converges very slowly for S=1/2, making spin-wave theory quantitatively inaccurate up to very large orders; (ii) compared to the linear spin-wave predictions, our magnon dispersion presents a much better agreement with experimental results on K$_{2}$Co(SeO$_{3}$)$_{2}$, for which $\alpha \simeq 0.08$., Comment: 6 pages, 3 figures

We construct and study the ideal Poisson--Voronoi tessellation of the product of two hyperbolic planes $\mathbb{H}_{2}\times \mathbb{H}_{2}$ endowed with the $L^{1}$ norm. We prove that its law is invariant under all isometries of this space and study some geometric features of its cells. Among other things, we prove that the set of points at equal separation to any two corona points is unbounded almost surely. This is analogous to a recent result of Fr\k{a}czyk-Mellick-Wilkens for higher rank symmetric spaces., Comment: 12 pages, 4 figures. Comments welcome!

In this work, we unveil a novel tool for generating Italian crossword puzzles from text, utilizing advanced language models such as GPT-4o, Mistral-7B-Instruct-v0.3, and Llama3-8b-Instruct. Crafted specifically for educational applications, this cutting-edge generator makes use of the comprehensive Italian-Clue-Instruct dataset, which comprises over 30,000 entries including diverse text, solutions, and types of clues. This carefully assembled dataset is designed to facilitate the creation of contextually relevant clues in various styles associated with specific texts and keywords. The study delves into four distinctive styles of crossword clues: those without format constraints, those formed as definite determiner phrases, copular sentences, and bare noun phrases. Each style introduces unique linguistic structures to diversify clue presentation. Given the lack of sophisticated educational tools tailored to the Italian language, this project seeks to enhance learning experiences and cognitive development through an engaging, interactive platform. By meshing state-of-the-art AI with contemporary educational strategies, our tool can dynamically generate crossword puzzles from Italian educational materials, thereby providing an enjoyable and interactive learning environment. This technological advancement not only redefines educational paradigms but also sets a new benchmark for interactive and cognitive language learning solutions., Comment: This paper has been accepted for presentation at CLiC.it 2024

The VLT Survey Telescope Survey of Mass Assembly and Structural Hierarchy (VST-SMASH) aims to detect tidal features and remnants around very nearby galaxies, a unique and essential diagnostic of the hierarchical nature of galaxy formation. Leveraging optimal sky conditions at ESO's Paranal Observatory, combined with the VST's multi-band optical filters, VST-SMASH aims to be the definitive survey of stellar streams and tidal remnants in the Local Volume, targeting a low surface-brightness limit of $\mu \sim$ 30 mag arcsec$^{-2}$ in the g and r bands, and $\mu \sim$ 28 mag arcsec$^{-2}$ in the i band, in a volume-limited sample of local galaxies within 11 Mpc and the Euclid footprint., Comment: 4 pages, 2 figures, 1 table, published in the ESO Messenger 193

Self-supervised training of language models (LMs) has seen great success for protein sequences in learning meaningful representations and for generative drug design. Most protein LMs are based on the Transformer architecture trained on individual proteins with short context lengths. Such protein LMs cannot extrapolate to longer proteins and protein complexes well. They also fail to account for the underlying biological mechanisms carried out by biomolecular interactions and dynamics i.e., proteins often interact with other proteins, molecules, and pathways in complex biological systems. In this work, we propose LC-PLM based on an alternative protein LM architecture, BiMamba-S, built off selective structured state-space models, to learn high-quality universal protein representations at the amino acid token level using masked language modeling. We also introduce its graph-contextual variant, LC-PLM-G, which contextualizes protein-protein interaction (PPI) graphs for a second stage of training. LC-PLM demonstrates favorable neural scaling laws, better length extrapolation capability, and a 7% to 34% improvement on protein downstream tasks than Transformer-based ESM-2. LC-PLM-G further trained within the context of PPI graphs shows promising results on protein structure and function prediction tasks. Our study demonstrates the benefit of increasing the context size with computationally efficient LM architecture (e.g. structured state space models) in learning universal protein representations and incorporating molecular interaction context contained in biological graphs., Comment: 32 pages, 17 figures, 11 tables

The field of autonomous navigation for unmanned ground vehicles (UGVs) is in continuous growth and increasing levels of autonomy have been reached in the last few years. However, the task becomes more challenging when the focus is on the exploration of planet surfaces such as Mars. In those situations, UGVs are forced to navigate through unstable and rugged terrains which, inevitably, open the vehicle to more hazards, accidents, and, in extreme cases, complete mission failure. The paper addresses the challenges of autonomous navigation for unmanned ground vehicles in planetary exploration, particularly on Mars, introducing a hybrid architecture for terrain traversability analysis that combines two approaches: appearance-based and geometry-based. The appearance-based method uses semantic segmentation via deep neural networks to classify different terrain types. This is further refined by pixel-level terrain roughness classification obtained from the same RGB image, assigning different costs based on the physical properties of the soil. The geometry-based method complements the appearance-based approach by evaluating the terrain's geometrical features, identifying hazards that may not be detectable by the appearance-based side. The outputs of both methods are combined into a comprehensive hybrid cost map. The proposed architecture was trained on synthetic datasets and developed as a ROS2 application to integrate into broader autonomous navigation systems for harsh environments. Simulations have been performed in Unity, showing the ability of the method to assess online traversability analysis.

Large language models (LLMs) demonstrate surprising capabilities, but we do not understand how they are implemented. One hypothesis suggests that these capabilities are primarily executed by small subnetworks within the LLM, known as circuits. But how can we evaluate this hypothesis? In this paper, we formalize a set of criteria that a circuit is hypothesized to meet and develop a suite of hypothesis tests to evaluate how well circuits satisfy them. The criteria focus on the extent to which the LLM's behavior is preserved, the degree of localization of this behavior, and whether the circuit is minimal. We apply these tests to six circuits described in the research literature. We find that synthetic circuits -- circuits that are hard-coded in the model -- align with the idealized properties. Circuits discovered in Transformer models satisfy the criteria to varying degrees. To facilitate future empirical studies of circuits, we created the \textit{circuitry} package, a wrapper around the \textit{TransformerLens} library, which abstracts away lower-level manipulations of hooks and activations. The software is available at \url{https://github.com/blei-lab/circuitry}., Comment: Code available here: https://github.com/blei-lab/circuitry

In many industrial settings, users wish to ask questions whose answers may be found in structured data sources such as a spreadsheets, databases, APIs, or combinations thereof. Often, the user doesn't know how to identify or access the right data source. This problem is compounded even further if multiple (and potentially siloed) data sources must be assembled to derive the answer. Recently, various Text-to-SQL applications that leverage Large Language Models (LLMs) have addressed some of these problems by enabling users to ask questions in natural language. However, these applications remain impractical in realistic industrial settings because they fail to cope with the data source heterogeneity that typifies such environments. In this paper, we address heterogeneity by introducing the siwarex platform, which enables seamless natural language access to both databases and APIs. To demonstrate the effectiveness of siwarex, we extend the popular Spider dataset and benchmark by replacing some of its tables by data retrieval APIs. We find that siwarex does a good job of coping with data source heterogeneity. Our modified Spider benchmark will soon be available to the research community

The report explores the challenges and strategies for preserving 3D digital data in cultural heritage. It discusses the issue of technological obsolescence, emphasising the need for ustainable storage solutions and ongoing data management strategies. Key topics include understanding technological obsolescence, the lifecycle of digital content, digital continuity, data management plans (DMP), FAIR principles, and the use of public repositories. The report also covers the importance of metadata in long-term digital preservation, including types of metadata and strategies for building valuable metadata. It examines the evolving standards and interoperability in 3D format preservation and the importance of managing metadata and paradata. The document provides a comprehensive overview of the challenges and solutions for preserving 3D cultural heritage data in the long term.

We characterize the self-assembly and phase behavior of Janus rods over a broad range of temperatures and volume fractions, using Langevin dynamics simulations and free energy calculations. The Janus rods consist of a line of fused overlapping spheres that interact via a soft-core repulsive potential, with the addition of an attractive pseudo-square-well tail to a fraction of the spheres (the coverage) ranging from 5\% to 100\% of sites. Competition between the stability of liquid crystal phases originating from shape anisotropy and assembly driven by directional interactions gives rise to a rich polymorphism that depends on the coverage. At low density near the Boyle temperature, we observe the formation of spherical and tubular micelles at low coverage, while at higher coverage randomly oriented monolayers form as the attractive parts of the rods overlap. At higher density, bilayer structures appear and merge to form smectic and crystalline lamellar phases. All of these structures gradually become unstable as the temperature is increased until eventually regular nematic and smectic phases appear, consistent with the hard rod limit. Our results indicate that the intermediate regime where shape-entropic effects compete with anisotropic attractions provided by site specificity is rich in structural possibilities, and should help guide the design of rod-like colloids for specific applications., Comment: 13 pages, 13 figures, SI not included

We report on extensive molecular dynamics atomistic simulations of a \textit{meta}-substituted \textit{poly}-phenylacetylene (pPA) foldamer dispersed in three solvents, water \ce{H2O}, cyclohexane \ce{cC6H12}, and \textit{n}-hexane \ce{nC6H14}, and for three oligomer lengths \textit{12mer}, \textit{16mer} and \textit{20mer}. At room temperature, we find a tendency of the pPA foldamer to collapse into a helical structure in all three solvents but with rather different stability character, stable in water, marginally stable in n-hexane, unstable in cyclohexane. In the case of water, the initial and final number of hydrogen bonds of the foldamer with water molecules is found to be unchanged, with no formation of intrachain hydrogen bonding, thus indicating that hydrogen bonding plays no role in the folding process. In all three solvents, the folding is found to be mainly driven by electrostatics, nearly identical in the three cases, and largely dominant compared to van der Waals interactions that are different in the three cases. This scenario is also supported by the analysis of distribution of the bond and dihedral angles and by a direct calculation of the solvation and transfer free energies via thermodynamic integration. The different stability in the case of cyclohexane and n-hexane notwithstanding their rather similar chemical composition can be traced back to the different entropy-enthalpy compensation that is found similar for water and n-hexane, and very different for cyclohexane. A comparison with the same properties for \textit{poly}-phenylalanine oligomers underscores the crucial differences between pPA and peptides. To highlight how these findings can hardly be interpreted in terms of a simple "good" and "poor" solvent picture, a molecular dynamics study of a bead-spring polymer chain in a Lennard-Jones fluid is also included., Comment: 24 pages, 16 Figures, to appear in Soft Matter. Supplementary Material not included

Affective computing has recently gained popularity, especially in the field of human-computer interaction systems, where effectively evoking and detecting emotions is of paramount importance to enhance users experience. However, several issues are hindering progress in the field. In fact, the complexity of emotions makes it difficult to understand their triggers and control their elicitation. Additionally, effective emotion recognition requires analyzing multiple sensor data, such as facial expressions and physiological signals. These factors combined make it hard to collect high-quality datasets that can be used for research purposes (e.g., development of emotion recognition algorithms). Despite these challenges, Virtual Reality (VR) holds promise as a solution. By providing a controlled and immersive environment, VR enables the replication of real-world emotional experiences and facilitates the tracking of signals indicative of emotional states. However, controlling emotion elicitation remains a challenging task also within VR. This research paper introduces the Magic Xroom, a VR platform designed to enhance control over emotion elicitation by leveraging the theory of flow. This theory establishes a mapping between an individuals skill levels, task difficulty, and perceived emotions. In the Magic Xroom, the users skill level is continuously assessed, and task difficulty is adjusted accordingly to evoke specific emotions. Furthermore, user signals are collected using sensors, and virtual panels are utilized to determine the ground truth emotional states, making the Magic Xroom an ideal platform for collecting extensive datasets. The paper provides detailed implementation information, highlights the main properties of the Magic Xroom, and presents examples of virtual scenarios to illustrate its abilities and capabilities., Comment: Proceedings of the 25th International Conference on Mobile Human-Computer Interaction

We describe a basic correspondence between linear algebraic structures within vector embeddings in artificial neural networks and conditional independence constraints on the probability distributions modeled by these networks. Our framework aims to shed light on the emergence of structural patterns in data representations, a phenomenon widely acknowledged but arguably still lacking a solid formal grounding. Specifically, we introduce a characterization of compositional structures in terms of "interaction decompositions," and we establish necessary and sufficient conditions for the presence of such structures within the representations of a model., Comment: 15 pages, 3 figures

We present the most comprehensive catalog to date of Type I Superluminous Supernovae (SLSNe), a class of stripped envelope supernovae (SNe) characterized by exceptionally high luminosities. We have compiled a sample of 262 SLSNe reported through 2022 December 31. We verified the spectroscopic classification of each SLSN and collated an exhaustive data set of UV, optical and IR photometry from both publicly available data and our own FLEET observational follow-up program, totaling over 30,000 photometric detections. Using these data we derive observational parameters such as the peak absolute magnitudes, rise and decline timescales, as well as bolometric luminosities, temperature and photospheric radius evolution for all SLSNe. Additionally, we model all light curves using a hybrid model that includes contributions from both a magnetar central engine and the radioactive decay of $^{56}$Ni. We explore correlations among various physical and observational parameters, and recover the previously found relation between ejecta mass and magnetar spin, as well as the overall progenitor pre-explosion mass distribution with a peak at $\approx 6.5$ M$_\odot$. We find no significant redshift dependence for any parameter, and no evidence for distinct sub-types of SLSNe. We find that $< 3$\% of SLSNe are best fit with a significant contribution from radioactive decay $\gtrsim 50$\%, representing a set of relatively dim and slowly declining SNe. We provide several analytical tools designed to simulate typical SLSN light curves across a broad range of wavelengths and phases, enabling accurate K-corrections, bolometric scaling calculations, and inclusion of SLSNe in survey simulations or future comparison works. The complete catalog, including all of the photometry, models, and derived parameters, is made available as an open-source resource on GitHub., Comment: 59 pages, 22 Figures, Submitted to MNRAS

The paper concerns the extension of the Heritage Digital Twin Ontology introduced in previous work to describe the reactivity of digital twins used for cultural heritage documentation by including the semantic description of sensors and activators and all the process of interacting with the real world. After analysing previous work on the use of digital twins in cultural heritage, a summary description of the Heritage Digital Twin Ontology is provided, and the existing applications of digital twins to cultural heritage are overviewed, with references to reviews summarizing the large production of scientific contributions on the topic. Then a novel ontology, named Reactive Digital Twin Ontology is described, in which sensors, activators and the decision processes are also semantically described, turning the previous synchronic approach to cultural heritage documentation into a diachronic one. Some case studies exemplify this theory., Comment: Submitted to Sensors (Open Access Journal from MDPI). 17 pages, 3 figures

We describe a family of architectures to support transductive inference by allowing memory to grow to a finite but a-priori unknown bound while making efficient use of finite resources for inference. Current architectures use such resources to represent data either eidetically over a finite span ("context" in Transformers), or fading over an infinite span (in State Space Models, or SSMs). Recent hybrid architectures have combined eidetic and fading memory, but with limitations that do not allow the designer or the learning process to seamlessly modulate the two, nor to extend the eidetic memory span. We leverage ideas from Stochastic Realization Theory to develop a class of models called B'MOJO to seamlessly combine eidetic and fading memory within an elementary composable module. The overall architecture can be used to implement models that can access short-term eidetic memory "in-context," permanent structural memory "in-weights," fading memory "in-state," and long-term eidetic memory "in-storage" by natively incorporating retrieval from an asynchronously updated memory. We show that Transformers, existing SSMs such as Mamba, and hybrid architectures such as Jamba are special cases of B'MOJO and describe a basic implementation, to be open sourced, that can be stacked and scaled efficiently in hardware. We test B'MOJO on transductive inference tasks, such as associative recall, where it outperforms existing SSMs and Hybrid models; as a baseline, we test ordinary language modeling where B'MOJO achieves perplexity comparable to similarly-sized Transformers and SSMs up to 1.4B parameters, while being up to 10% faster to train. Finally, we show that B'MOJO's ability to modulate eidetic and fading memory results in better inference on longer sequences tested up to 32K tokens, four-fold the length of the longest sequences seen during training.

We present a lattice model for polymer solutions, explicitly incorporating interactions with a bath of solvent and cosolvent molecules. By exploiting the well-known analogy between polymer systems and the $O(n)$-vector spin model in the limit $n \to 0$, we derive an exact field-theoretic expression for the partition function of the system. The latter is then evaluated at the saddle point, providing a mean-field estimate of the free energy. The resulting expression, which conforms to the Flory-Huggins type, is then used to analyze the system's stability with respect to phase separation, complemented by a numerical approach based on convex hull evaluation. We demonstrate that this simple lattice model can effectively explain the behavior of a variety of seemingly unrelated polymer systems, which have been predominantly investigated in the past only through numerical simulations. This includes both, single-chain and multi-chain, solutions. Our findings emphasize the fundamental, mutually competing, roles of solvent and cosolvent in polymer systems., Comment: 12 pages, 6 figures (main); 9 pages, 5 figures (suppl info); Physical Review Materials, accepted for publication

We present Diffusion Soup, a compartmentalization method for Text-to-Image Generation that averages the weights of diffusion models trained on sharded data. By construction, our approach enables training-free continual learning and unlearning with no additional memory or inference costs, since models corresponding to data shards can be added or removed by re-averaging. We show that Diffusion Soup samples from a point in weight space that approximates the geometric mean of the distributions of constituent datasets, which offers anti-memorization guarantees and enables zero-shot style mixing. Empirically, Diffusion Soup outperforms a paragon model trained on the union of all data shards and achieves a 30% improvement in Image Reward (.34 $\to$ .44) on domain sharded data, and a 59% improvement in IR (.37 $\to$ .59) on aesthetic data. In both cases, souping also prevails in TIFA score (respectively, 85.5 $\to$ 86.5 and 85.6 $\to$ 86.8). We demonstrate robust unlearning -- removing any individual domain shard only lowers performance by 1% in IR (.45 $\to$ .44) -- and validate our theoretical insights on anti-memorization using real data. Finally, we showcase Diffusion Soup's ability to blend the distinct styles of models finetuned on different shards, resulting in the zero-shot generation of hybrid styles.

Probabilistic prediction aims to compute predictive distributions rather than single point predictions. These distributions enable practitioners to quantify uncertainty, compute risk, and detect outliers. However, most probabilistic methods assume parametric responses, such as Gaussian or Poisson distributions. When these assumptions fail, such models lead to bad predictions and poorly calibrated uncertainty. In this paper, we propose Treeffuser, an easy-to-use method for probabilistic prediction on tabular data. The idea is to learn a conditional diffusion model where the score function is estimated using gradient-boosted trees. The conditional diffusion model makes Treeffuser flexible and non-parametric, while the gradient-boosted trees make it robust and easy to train on CPUs. Treeffuser learns well-calibrated predictive distributions and can handle a wide range of regression tasks -- including those with multivariate, multimodal, and skewed responses. We study Treeffuser on synthetic and real data and show that it outperforms existing methods, providing better calibrated probabilistic predictions. We further demonstrate its versatility with an application to inventory allocation under uncertainty using sales data from Walmart. We implement Treeffuser in https://github.com/blei-lab/treeffuser., Comment: NeurIPS 2024

Protecting digital identities of human face from various attack vectors is paramount, and face anti-spoofing plays a crucial role in this endeavor. Current approaches primarily focus on detecting spoofing attempts within individual frames to detect presentation attacks. However, the emergence of hyper-realistic generative models capable of real-time operation has heightened the risk of digitally generated attacks. In light of these evolving threats, this paper aims to address two key aspects. First, it sheds light on the vulnerabilities of state-of-the-art face anti-spoofing methods against digital attacks. Second, it presents a comprehensive taxonomy of common threats encountered in face anti-spoofing systems. Through a series of experiments, we demonstrate the limitations of current face anti-spoofing detection techniques and their failure to generalize to novel digital attack scenarios. Notably, the existing models struggle with digital injection attacks including adversarial noise, realistic deepfake attacks, and digital replay attacks. To aid in the design and implementation of robust face anti-spoofing systems resilient to these emerging vulnerabilities, the paper proposes key design principles from model accuracy and robustness to pipeline robustness and even platform robustness. Especially, we suggest to implement the proactive face anti-spoofing system using active sensors to significant reduce the risks for unseen attack vectors and improve the user experience., Comment: Under review

Using Langevin dynamics complemented by Wang-Landau Monte Carlo simulations, we study the phase behavior of single and multiple semiflexible polymer chains in solution under poor-solvent conditions. In the case of a single chain, we obtain the full phase diagram in the temperature-bending rigidity (stiffness) plane and we provide connections with a classical mean field result on a lattice as well as with past results on the same model. At low bending rigidity and upon cooling, we find a second order coil-globule transition, followed by a subsequent first order globule-crystal transition at lower temperatures. The obtained crystals have the shape of a twisted rod whose length increases with the increase of the stiffness of the chain. Above a critical value of the stiffness, we also find a direct first order globule-crystal transition, with the crystal having the form of a twisted toroid. Close to the triple point, we find a region with isoenergetic structures with frequent switching from rods to toroids, with the toroid eventually becoming the only observed stable phase at higher stiffness. The model is then extended to many thermally equilibrated chains in a box and the analogous phase diagram is deduced where the chains are observed to first fold into a globule bundle at low stiffness upon cooling, and then rearrange into a nematic bundle via a nucleation process involving an isotropic-nematic transition. As in the single chain counterpart, above a critical stiffness the chains are observed to undergo a direct transition from a gas of isotropically distributed chains to a nematic bundle as the temperature decreases in agreement with recent suggestions from mean field theory. The consequences of these findings for self-assembly of biopolymers in solutions are discussed., Comment: 59 pages, supporting information; published in Macromolecules (2024)

While machine learning algorithms hold promise for personalised medicine, their clinical adoption remains limited, partly due to biases that can compromise the reliability of predictions. In this paper, we focus on sample selection bias (SSB), a specific type of bias where the study population is less representative of the target population, leading to biased and potentially harmful decisions. Despite being well-known in the literature, SSB remains scarcely studied in machine learning for healthcare. Moreover, the existing machine learning techniques try to correct the bias mostly by balancing distributions between the study and the target populations, which may result in a loss of predictive performance. To address these problems, our study illustrates the potential risks associated with SSB by examining SSB's impact on the performance of machine learning algorithms. Most importantly, we propose a new research direction for addressing SSB, based on the target population identification rather than the bias correction. Specifically, we propose two independent networks(T-Net) and a multitasking network (MT-Net) for addressing SSB, where one network/task identifies the target subpopulation which is representative of the study population and the second makes predictions for the identified subpopulation. Our empirical results with synthetic and semi-synthetic datasets highlight that SSB can lead to a large drop in the performance of an algorithm for the target population as compared with the study population, as well as a substantial difference in the performance for the target subpopulations that are representative of the selected and the non-selected patients from the study population. Furthermore, our proposed techniques demonstrate robustness across various settings, including different dataset sizes, event rates, and selection rates, outperforming the existing bias correction techniques., Comment: 21 pages and 11 figures (under review)

We propose NeRF-Insert, a NeRF editing framework that allows users to make high-quality local edits with a flexible level of control. Unlike previous work that relied on image-to-image models, we cast scene editing as an in-painting problem, which encourages the global structure of the scene to be preserved. Moreover, while most existing methods use only textual prompts to condition edits, our framework accepts a combination of inputs of different modalities as reference. More precisely, a user may provide a combination of textual and visual inputs including images, CAD models, and binary image masks for specifying a 3D region. We use generic image generation models to in-paint the scene from multiple viewpoints, and lift the local edits to a 3D-consistent NeRF edit. Compared to previous methods, our results show better visual quality and also maintain stronger consistency with the original NeRF.

Retrieval Augmented Generation (RAG) is emerging as a flexible and robust technique to adapt models to private users data without training, to handle credit attribution, and to allow efficient machine unlearning at scale. However, RAG techniques for image generation may lead to parts of the retrieved samples being copied in the model's output. To reduce risks of leaking private information contained in the retrieved set, we introduce Copy-Protected generation with Retrieval (CPR), a new method for RAG with strong copyright protection guarantees in a mixed-private setting for diffusion models.CPR allows to condition the output of diffusion models on a set of retrieved images, while also guaranteeing that unique identifiable information about those example is not exposed in the generated outputs. In particular, it does so by sampling from a mixture of public (safe) distribution and private (user) distribution by merging their diffusion scores at inference. We prove that CPR satisfies Near Access Freeness (NAF) which bounds the amount of information an attacker may be able to extract from the generated images. We provide two algorithms for copyright protection, CPR-KL and CPR-Choose. Unlike previously proposed rejection-sampling-based NAF methods, our methods enable efficient copyright-protected sampling with a single run of backward diffusion. We show that our method can be applied to any pre-trained conditional diffusion model, such as Stable Diffusion or unCLIP. In particular, we empirically show that applying CPR on top of unCLIP improves quality and text-to-image alignment of the generated results (81.4 to 83.17 on TIFA benchmark), while enabling credit attribution, copy-right protection, and deterministic, constant time, unlearning., Comment: CVPR 2024

Generative Vision-Language Models (VLMs) are prone to generate plausible-sounding textual answers that, however, are not always grounded in the input image. We investigate this phenomenon, usually referred to as "hallucination" and show that it stems from an excessive reliance on the language prior. In particular, we show that as more tokens are generated, the reliance on the visual prompt decreases, and this behavior strongly correlates with the emergence of hallucinations. To reduce hallucinations, we introduce Multi-Modal Mutual-Information Decoding (M3ID), a new sampling method for prompt amplification. M3ID amplifies the influence of the reference image over the language prior, hence favoring the generation of tokens with higher mutual information with the visual prompt. M3ID can be applied to any pre-trained autoregressive VLM at inference time without necessitating further training and with minimal computational overhead. If training is an option, we show that M3ID can be paired with Direct Preference Optimization (DPO) to improve the model's reliance on the prompt image without requiring any labels. Our empirical findings show that our algorithms maintain the fluency and linguistic capabilities of pre-trained VLMs while reducing hallucinations by mitigating visually ungrounded answers. Specifically, for the LLaVA 13B model, M3ID and M3ID+DPO reduce the percentage of hallucinated objects in captioning tasks by 25% and 28%, respectively, and improve the accuracy on VQA benchmarks such as POPE by 21% and 24%.

We identify the local limit of massive spanning forests on the complete graph. This generalizes a well-known theorem of Grimmett on the local limit of uniform spanning trees on the complete graph., Comment: 10 pages, 2 figures

Supervised models for Word Sense Disambiguation (WSD) currently yield to state-of-the-art results in the most popular benchmarks. Despite the recent introduction of Word Embeddings and Recurrent Neural Networks to design powerful context-related features, the interest in improving WSD models using Semantic Lexical Resources (SLRs) is mostly restricted to knowledge-based approaches. In this paper, we enhance "modern" supervised WSD models exploiting two popular SLRs: WordNet and WordNet Domains. We propose an effective way to introduce semantic features into the classifiers, and we consider using the SLR structure to augment the training data. We study the effect of different types of semantic features, investigating their interaction with local contexts encoded by means of mixtures of Word Embeddings or Recurrent Neural Networks, and we extend the proposed model into a novel multi-layer architecture for WSD. A detailed experimental comparison in the recent Unified Evaluation Framework (Raganato et al., 2017) shows that the proposed approach leads to supervised models that compare favourably with the state-of-the art., Comment: The 11th International Conference on Language Resources and Evaluation (LREC 2018)

We define bipartite and monopartite relational networks of chemical elements and compounds using two different datasets of inorganic chemical and material compounds, as well as study their topology. We discover that the connectivity between elements and compounds is distributed exponentially for materials, and with a fat tail for chemicals. Compounds networks show similar distribution of degrees, and feature a highly-connected club due to oxygen. Chemical compounds networks appear more modular than material ones, while the communities detected reveal different dominant elements specific to the topology. We successfully reproduce the connectivity of the empirical chemicals and materials networks by using a family of fitness models, where the fitness values are derived from the abundances of the elements in the aggregate compound data. Our results pave the way towards a relational network-based understanding of the inherent complexity of the vast chemical knowledge atlas, and our methodology can be applied to other systems with the ingredient-composite structure., Comment: main manuscript with 3 Figures, and supplementary information with 8 figures

Paraphrasing is the task of re-writing an input text using other words, without altering the meaning of the original content. Conversational systems can exploit automatic paraphrasing to make the conversation more natural, e.g., talking about a certain topic using different paraphrases in different time instants. Recently, the task of automatically generating paraphrases has been approached in the context of Natural Language Generation (NLG). While many existing systems simply consist in rule-based models, the recent success of the Deep Neural Networks in several NLG tasks naturally suggests the possibility of exploiting such networks for generating paraphrases. However, the main obstacle toward neural-network-based paraphrasing is the lack of large datasets with aligned pairs of sentences and paraphrases, that are needed to efficiently train the neural models. In this paper we present a method for the automatic generation of large aligned corpora, that is based on the assumption that news and blog websites talk about the same events using different narrative styles. We propose a similarity search procedure with linguistic constraints that, given a reference sentence, is able to locate the most similar candidate paraphrases out from millions of indexed sentences. The data generation process is evaluated in the case of the Italian language, performing experiments using pointer-based deep neural architectures., Comment: The 6th International Conference on Social Networks Analysis, Management and Security (SNAMS 2019)

From robots that replace workers to robots that serve as helpful colleagues, the field of robotic automation is experiencing a new trend that represents a huge challenge for component manufacturers. The contribution starts from an innovative vision that sees an ever closer collaboration between Cobot, able to do a specific physical job with precision, the AI world, able to analyze information and support the decision-making process, and the man able to have a strategic vision of the future.

Quantifying the degree of similarity between images is a key copyright issue for image-based machine learning. In legal doctrine however, determining the degree of similarity between works requires subjective analysis, and fact-finders (judges and juries) can demonstrate considerable variability in these subjective judgement calls. Images that are structurally similar can be deemed dissimilar, whereas images of completely different scenes can be deemed similar enough to support a claim of copying. We seek to define and compute a notion of "conceptual similarity" among images that captures high-level relations even among images that do not share repeated elements or visually similar components. The idea is to use a base multi-modal model to generate "explanations" (captions) of visual data at increasing levels of complexity. Then, similarity can be measured by the length of the caption needed to discriminate between the two images: Two highly dissimilar images can be discriminated early in their description, whereas conceptually dissimilar ones will need more detail to be distinguished. We operationalize this definition and show that it correlates with subjective (averaged human evaluation) assessment, and beats existing baselines on both image-to-image and text-to-text similarity benchmarks. Beyond just providing a number, our method also offers interpretability by pointing to the specific level of granularity of the description where the source data are differentiated.