Fabrizio Silvestri

Antonio Purificato, Giulia Cassarà, Federico Siciliano et al.

Recent advancements in Graph Neural Networks (GNN) have facilitated their widespread adoption in various applications, including recommendation systems. GNNs have proven to be effective in addressing the challenges posed by recommendation systems by efficiently modeling graphs in which nodes represent users or items and edges denote preference relationships. However, current GNN techniques represent nodes by means of a single static vector, which may inadequately capture the intricate complexities of users and items. To overcome these limitations, we propose a solution integrating a cutting-edge model inspired by category theory: Sheaf4Rec. Unlike single vector representations, Sheaf Neural Networks and their corresponding Laplacians represent each node (and edge) using a vector space. Our approach takes advantage from this theory and results in a more comprehensive representation that can be effectively exploited during inference, providing a versatile method applicable to a wide range of graph-related tasks and demonstrating unparalleled performance. Our proposed model exhibits a noteworthy relative improvement of up to 8.53% on F1-Score@10 and an impressive increase of up to 11.29% on NDCG@10, outperforming existing state-of-the-art models such as Neural Graph Collaborative Filtering (NGCF), KGTORe and other recently developed GNN-based models. In addition to its superior predictive capabilities, Sheaf4Rec shows remarkable improvements in terms of efficiency: we observe substantial runtime improvements ranging from 2.5% up to 37% when compared to other GNN-based competitor models, indicating a more efficient way of handling information while achieving better performance. Code is available at https://github.com/antoniopurificato/Sheaf4Rec.

Shivam Sharma, Firoj Alam, Md. Shad Akhtar et al.

The automatic identification of harmful content online is of major concern for social media platforms, policymakers, and society. Researchers have studied textual, visual, and audio content, but typically in isolation. Yet, harmful content often combines multiple modalities, as in the case of memes, which are of particular interest due to their viral nature. With this in mind, here we offer a comprehensive survey with a focus on harmful memes. Based on a systematic analysis of recent literature, we first propose a new typology of harmful memes, and then we highlight and summarize the relevant state of the art. One interesting finding is that many types of harmful memes are not really studied, e.g., such featuring self-harm and extremism, partly due to the lack of suitable datasets. We further find that existing datasets mostly capture multi-class scenarios, which are not inclusive of the affective spectrum that memes can represent. Another observation is that memes can propagate globally through repackaging in different languages and that they can also be multilingual, blending different cultures. We conclude by highlighting several challenges related to multimodal semiotics, technological constraints, and non-trivial social engagement, and we present several open-ended aspects such as delineating online harm and empirically examining related frameworks and assistive interventions, which we believe will motivate and drive future research.

Valeria Ruscio, Umberto Nanni, Fabrizio Silvestri

RoPE-trained transformers distinguish absolute position in their attention patterns, even though RoPE encodes only relative offsets in the inner product. We trace this leakage to two architectural components, The causal mask is responsible for the first: its per-query softmax denominator depends on the absolute query position by construction. The residual stream supplies the second. Under causal attention the activation at position $0$ attends only to itself and runs as a closed dynamical system from the embedding of the token at that position; downstream attention reads this trajectory through sink-reading heads. Both components appear in all three architectures we study, in architecturally specific balance: NTK scaling suppresses the residual-stream component, sliding-window attention allows it to accumulate with depth, and standard RoPE sits between. Replacing the \texttt{BOS} embedding before the forward pass removes $40\%$ of the residual-stream component at early queries. Attention sinks are token-anchored stabilizers that pass forward a deterministic fingerprint of the token at position $0$, constant across inputs when that token is the auto-prepended \texttt{BOS} and varying with it otherwise.

Andrea Bacciu, Giovanni Trappolini, Andrea Santilli et al.

This paper presents Fauno, the first and largest open-source Italian conversational Large Language Model (LLM). Our goal with Fauno is to democratize the study of LLMs in Italian, demonstrating that obtaining a fine-tuned conversational bot with a single GPU is possible. In addition, we release a collection of datasets for conversational AI in Italian. The datasets on which we fine-tuned Fauno include various topics such as general question answering, computer science, and medical questions. We release our code and datasets on \url{https://github.com/RSTLess-research/Fauno-Italian-LLM}

Filippo Betello, Federico Siciliano, Pushkar Mishra et al. · meta-ai

Sequential Recommender Systems (SRSs) are widely employed to model user behavior over time. However, their robustness in the face of perturbations in training data remains a largely understudied yet critical issue. A fundamental challenge emerges in previous studies aimed at assessing the robustness of SRSs: the Rank-Biased Overlap (RBO) similarity is not particularly suited for this task as it is designed for infinite rankings of items and thus shows limitations in real-world scenarios. For instance, it fails to achieve a perfect score of 1 for two identical finite-length rankings. To address this challenge, we introduce a novel contribution: Finite Rank-Biased Overlap (FRBO), an enhanced similarity tailored explicitly for finite rankings. This innovation facilitates a more intuitive evaluation in practical settings. In pursuit of our goal, we empirically investigate the impact of removing items at different positions within a temporally ordered sequence. We evaluate two distinct SRS models across multiple datasets, measuring their performance using metrics such as Normalized Discounted Cumulative Gain (NDCG) and Rank List Sensitivity. Our results demonstrate that removing items at the end of the sequence has a statistically significant impact on performance, with NDCG decreasing up to 60%. Conversely, removing items from the beginning or middle has no significant effect. These findings underscore the criticality of the position of perturbed items in the training data. As we spotlight the vulnerabilities inherent in current SRSs, we fervently advocate for intensified research efforts to fortify their robustness against adversarial perturbations.

Filippo Betello, Antonio Purificato, Federico Siciliano et al.

Sequential Recommender Systems (SRSs) have emerged as a highly efficient approach to recommendation systems. By leveraging sequential data, SRSs can identify temporal patterns in user behaviour, significantly improving recommendation accuracy and relevance.Ensuring the reproducibility of these models is paramount for advancing research and facilitating comparisons between them. Existing works exhibit shortcomings in reproducibility and replicability of results, leading to inconsistent statements across papers. Our work fills these gaps by standardising data pre-processing and model implementations, providing a comprehensive code resource, including a framework for developing SRSs and establishing a foundation for consistent and reproducible experimentation. We conduct extensive experiments on several benchmark datasets, comparing various SRSs implemented in our resource. We challenge prevailing performance benchmarks, offering new insights into the SR domain. For instance, SASRec does not consistently outperform GRU4Rec. On the contrary, when the number of model parameters becomes substantial, SASRec starts to clearly dominate all the other SRSs. This discrepancy underscores the significant impact that experimental configuration has on the outcomes and the importance of setting it up to ensure precise and comprehensive results. Failure to do so can lead to significantly flawed conclusions, highlighting the need for rigorous experimental design and analysis in SRS research. Our code is available at https://github.com/antoniopurificato/recsys_repro_conf.

Ziheng Chen, Fabrizio Silvestri, Jia Wang et al.

Recently, graph neural networks (GNNs) have been widely used to develop successful recommender systems. Although powerful, it is very difficult for a GNN-based recommender system to attach tangible explanations of why a specific item ends up in the list of suggestions for a given user. Indeed, explaining GNN-based recommendations is unique, and existing GNN explanation methods are inappropriate for two reasons. First, traditional GNN explanation methods are designed for node, edge, or graph classification tasks rather than ranking, as in recommender systems. Second, standard machine learning explanations are usually intended to support skilled decision-makers. Instead, recommendations are designed for any end-user, and thus their explanations should be provided in user-understandable ways. In this work, we propose GREASE, a novel method for explaining the suggestions provided by any black-box GNN-based recommender system. Specifically, GREASE first trains a surrogate model on a target user-item pair and its $l$-hop neighborhood. Then, it generates both factual and counterfactual explanations by finding optimal adjacency matrix perturbations to capture the sufficient and necessary conditions for an item to be recommended, respectively. Experimental results conducted on real-world datasets demonstrate that GREASE can generate concise and effective explanations for popular GNN-based recommender models.

Lucie Charlotte Magister, Pietro Barbiero, Dmitry Kazhdan et al.

The opaque reasoning of Graph Neural Networks induces a lack of human trust. Existing graph network explainers attempt to address this issue by providing post-hoc explanations, however, they fail to make the model itself more interpretable. To fill this gap, we introduce the Concept Encoder Module, the first differentiable concept-discovery approach for graph networks. The proposed approach makes graph networks explainable by design by first discovering graph concepts and then using these to solve the task. Our results demonstrate that this approach allows graph networks to: (i) attain model accuracy comparable with their equivalent vanilla versions, (ii) discover meaningful concepts that achieve high concept completeness and purity scores, (iii) provide high-quality concept-based logic explanations for their prediction, and (iv) support effective interventions at test time: these can increase human trust as well as significantly improve model performance.

Farooq Ahmad Wani, Maria Sofia Bucarelli, Fabrizio Silvestri

We introduce a novel method for training machine learning models in the presence of noisy labels, which are prevalent in domains such as medical diagnosis and autonomous driving and have the potential to degrade a model's generalization performance. Inspired by established literature that highlights how deep learning models are prone to overfitting to noisy samples in the later epochs of training, we propose a strategic approach. This strategy leverages the distance to class centroids in the latent space and incorporates a discounting mechanism, aiming to diminish the influence of samples that lie distant from all class centroids. By doing so, we effectively counteract the adverse effects of noisy labels. The foundational premise of our approach is the assumption that samples situated further from their respective class centroid in the initial stages of training are more likely to be associated with noise. Our methodology is grounded in robust theoretical principles and has been validated empirically through extensive experiments on several benchmark datasets. Our results show that our method consistently outperforms the existing state-of-the-art techniques, achieving significant improvements in classification accuracy in the presence of noisy labels. The code for our proposed loss function and supplementary materials is available at https://github.com/wanifarooq/NCOD

Giovanni Trappolini, Valentino Maiorca, Silvio Severino et al.

Graph Neural Networks (GNNs) have proven to be successful in several predictive modeling tasks for graph-structured data. Amongst those tasks, link prediction is one of the fundamental problems for many real-world applications, such as recommender systems. However, GNNs are not immune to adversarial attacks, i.e., carefully crafted malicious examples that are designed to fool the predictive model. In this work, we focus on a specific, white-box attack to GNN-based link prediction models, where a malicious node aims to appear in the list of recommended nodes for a given target victim. To achieve this goal, the attacker node may also count on the cooperation of other existing peers that it directly controls, namely on the ability to inject a number of ``vicious'' nodes in the network. Specifically, all these malicious nodes can add new edges or remove existing ones, thereby perturbing the original graph. Thus, we propose SAVAGE, a novel framework and a method to mount this type of link prediction attacks. SAVAGE formulates the adversary's goal as an optimization task, striking the balance between the effectiveness of the attack and the sparsity of malicious resources required. Extensive experiments conducted on real-world and synthetic datasets demonstrate that adversarial attacks implemented through SAVAGE indeed achieve high attack success rate yet using a small amount of vicious nodes. Finally, despite those attacks require full knowledge of the target model, we show that they are successfully transferable to other black-box methods for link prediction.

Valeria Ruscio, Valentino Maiorca, Fabrizio Silvestri

We analyze how large language models (LLMs) represent out-of-context words, investigating their reliance on the given context to capture their semantics. Our likelihood-guided text perturbations reveal a correlation between token likelihood and attention values in transformer-based language models. Extensive experiments reveal that unexpected tokens cause the model to attend less to the information coming from themselves to compute their representations, particularly at higher layers. These findings have valuable implications for assessing the robustness of LLMs in real-world scenarios. Fully reproducible codebase at https://github.com/Flegyas/AttentionLikelihood.

Andrea Bacciu, Florin Cuconasu, Federico Siciliano et al.

The emergence of large language models (LLMs) has revolutionized machine learning and related fields, showcasing remarkable abilities in comprehending, generating, and manipulating human language. However, their conventional usage through API-based text prompt submissions imposes certain limitations in terms of context constraints and external source availability. To address these challenges, we propose a novel framework called Reinforced Retrieval Augmented Machine Learning (RRAML). RRAML integrates the reasoning capabilities of LLMs with supporting information retrieved by a purpose-built retriever from a vast user-provided database. By leveraging recent advancements in reinforcement learning, our method effectively addresses several critical challenges. Firstly, it circumvents the need for accessing LLM gradients. Secondly, our method alleviates the burden of retraining LLMs for specific tasks, as it is often impractical or impossible due to restricted access to the model and the computational intensity involved. Additionally we seamlessly link the retriever's task with the reasoner, mitigating hallucinations and reducing irrelevant, and potentially damaging retrieved documents. We believe that the research agenda outlined in this paper has the potential to profoundly impact the field of AI, democratizing access to and utilization of LLMs for a wide range of entities.

Gabriele Tolomei, Cesare Campagnano, Fabrizio Silvestri et al.

In this paper, we present a groundbreaking paradigm for human-computer interaction that revolutionizes the traditional notion of an operating system. Within this innovative framework, user requests issued to the machine are handled by an interconnected ecosystem of generative AI models that seamlessly integrate with or even replace traditional software applications. At the core of this paradigm shift are large generative models, such as language and diffusion models, which serve as the central interface between users and computers. This pioneering approach leverages the abilities of advanced language models, empowering users to engage in natural language conversations with their computing devices. Users can articulate their intentions, tasks, and inquiries directly to the system, eliminating the need for explicit commands or complex navigation. The language model comprehends and interprets the user's prompts, generating and displaying contextual and meaningful responses that facilitate seamless and intuitive interactions. This paradigm shift not only streamlines user interactions but also opens up new possibilities for personalized experiences. Generative models can adapt to individual preferences, learning from user input and continuously improving their understanding and response generation. Furthermore, it enables enhanced accessibility, as users can interact with the system using speech or text, accommodating diverse communication preferences. However, this visionary concept raises significant challenges, including privacy, security, trustability, and the ethical use of generative models. Robust safeguards must be in place to protect user data and prevent potential misuse or manipulation of the language model. While the full realization of this paradigm is still far from being achieved, this paper serves as a starting point for envisioning this transformative potential.

Andrea Bernini, Fabrizio Silvestri, Gabriele Tolomei

Community detection techniques are useful for social media platforms to discover tightly connected groups of users who share common interests. However, this functionality often comes at the expense of potentially exposing individuals to privacy breaches by inadvertently revealing their tastes or preferences. Therefore, some users may wish to preserve their anonymity and opt out of community detection for various reasons, such as affiliation with political or religious organizations, without leaving the platform. In this study, we address the challenge of community membership hiding, which involves strategically altering the structural properties of a network graph to prevent one or more nodes from being identified by a given community detection algorithm. We tackle this problem by formulating it as a constrained counterfactual graph objective, and we solve it via deep reinforcement learning. Extensive experiments demonstrate that our method outperforms existing baselines, striking the best balance between accuracy and cost.

Iulia Duta, Giulia Cassarà, Fabrizio Silvestri et al.

Higher-order relations are widespread in nature, with numerous phenomena involving complex interactions that extend beyond simple pairwise connections. As a result, advancements in higher-order processing can accelerate the growth of various fields requiring structured data. Current approaches typically represent these interactions using hypergraphs. We enhance this representation by introducing cellular sheaves for hypergraphs, a mathematical construction that adds extra structure to the conventional hypergraph while maintaining their local, higherorder connectivity. Drawing inspiration from existing Laplacians in the literature, we develop two unique formulations of sheaf hypergraph Laplacians: linear and non-linear. Our theoretical analysis demonstrates that incorporating sheaves into the hypergraph Laplacian provides a more expressive inductive bias than standard hypergraph diffusion, creating a powerful instrument for effectively modelling complex data structures. We employ these sheaf hypergraph Laplacians to design two categories of models: Sheaf Hypergraph Neural Networks and Sheaf Hypergraph Convolutional Networks. These models generalize classical Hypergraph Networks often found in the literature. Through extensive experimentation, we show that this generalization significantly improves performance, achieving top results on multiple benchmark datasets for hypergraph node classification.

Marco Salmè, Federico Siciliano, Fabrizio Silvestri et al.

Radiology Report Generation (RRG) through Vision-Language Models (VLMs) promises to reduce documentation burden, improve reporting consistency, and accelerate clinical workflows. However, their clinical adoption remains limited by the lack of interpretability and the tendency to hallucinate findings misaligned with imaging evidence. Existing research typically treats interpretability and accuracy as separate objectives, with concept-based explainability techniques focusing primarily on transparency, while Retrieval-Augmented Generation (RAG) methods targeting factual grounding through external retrieval. We present Concept-Enhanced Multimodal RAG (CEMRAG), a unified framework that decomposes visual representations into interpretable clinical concepts and integrates them with multimodal RAG. This approach exploits enriched contextual prompts for RRG, improving both interpretability and factual accuracy. Experiments on MIMIC-CXR and IU X-Ray across multiple VLM architectures, training regimes, and retrieval configurations demonstrate consistent improvements over both conventional RAG and concept-only baselines on clinical accuracy metrics and standard NLP measures. These results challenge the assumed trade-off between interpretability and performance, showing that transparent visual concepts can enhance rather than compromise diagnostic accuracy in medical VLMs. Our modular design decomposes interpretability into visual transparency and structured language model conditioning, providing a principled pathway toward clinically trustworthy AI-assisted radiology.

Giulia Di Teodoro, Federico Siciliano, Valerio Guarrasi et al.

Predicting the outcome of antiretroviral therapies (ART) for HIV-1 is a pressing clinical challenge, especially when the ART includes drugs with limited effectiveness data. This scarcity of data can arise either due to the introduction of a new drug to the market or due to limited use in clinical settings, resulting in clinical dataset with highly unbalanced therapy representation. To tackle this issue, we introduce a novel joint fusion model, which combines features from a Fully Connected (FC) Neural Network and a Graph Neural Network (GNN) in a multi-modality fashion. Our model uses both tabular data about genetic sequences and a knowledge base derived from Stanford drug-resistance mutation tables, which serve as benchmark references for deducing in-vivo treatment efficacy based on the viral genetic sequence. By leveraging this knowledge base structured as a graph, the GNN component enables our model to adapt to imbalanced data distributions and account for Out-of-Distribution (OoD) drugs. We evaluated these models' robustness against OoD drugs in the test set. Our comprehensive analysis demonstrates that the proposed model consistently outperforms the FC model. These results underscore the advantage of integrating Stanford scores in the model, thereby enhancing its generalizability and robustness, but also extending its utility in contributing in more informed clinical decisions with limited data availability. The source code is available at https://github.com/federicosiciliano/graph-ood-hiv

Farooq Ahmad Wani, Alessandro Suglia, Rohit Saxena et al.

The robustness of Vision Language Models (VLMs) is commonly assessed through output-level invariance, implicitly assuming that stable predictions reflect stable multimodal processing. In this work, we argue that this assumption is insufficient. We introduce a representation-aware and frequency-aware evaluation framework that measures internal embedding drift, spectral sensitivity, and structural smoothness (spatial consistency of vision tokens), alongside standard label-based metrics. Applying this framework to modern VLMs across the SEEDBench, MMMU, and POPE datasets reveals three distinct failure modes. First, models frequently preserve predicted answers while undergoing substantial internal representation drift; for perturbations such as text overlays, this drift approaches the magnitude of inter-image variability, indicating that representations move to regions typically occupied by unrelated inputs despite unchanged outputs. Second, robustness does not improve with scale; larger models achieve higher accuracy but exhibit equal or greater sensitivity, consistent with sharper yet more fragile decision boundaries. Third, we find that perturbations affect tasks differently: they harm reasoning when they disrupt how models combine coarse and fine visual cues, but on the hallucination benchmarks, they can reduce false positives by making models generate more conservative answers.

Antonio Purificato, Fabrizio Silvestri

Recommender systems play a crucial role in alleviating information overload by providing personalized recommendations tailored to users' preferences and interests. Recently, Graph Neural Networks (GNNs) have emerged as a promising approach for recommender systems, leveraging their ability to effectively capture complex relationships and dependencies between users and items by representing them as nodes in a graph structure. In this study, we investigate the environmental impact of GNN-based recommender systems, an aspect that has been largely overlooked in the literature. Specifically, we conduct a comprehensive analysis of the carbon emissions associated with training and deploying GNN models for recommendation tasks. We evaluate the energy consumption and carbon footprint of different GNN architectures and configurations, considering factors such as model complexity, training duration, hardware specifications and embedding size. By addressing the environmental impact of resource-intensive algorithms in recommender systems, this study contributes to the ongoing efforts towards sustainable and responsible artificial intelligence, promoting the development of eco-friendly recommendation technologies that balance performance and environmental considerations. Code is available at: https://github.com/antoniopurificato/gnn_recommendation_and_environment.

Francesco Caso, Giovanni Trappolini, Andrea Bacciu et al.

Graph neural networks process information on graphs represented at a given resolution scale. We analyze the effect of using different coarse-grained graph resolutions, obtained through the Laplacian renormalization group theory, on node classification tasks. At the theory's core is grouping nodes connected by significant information flow at a given time scale. Representations of the graph at different scales encode interaction information at different ranges. We specifically experiment using representations at the characteristic scale of the graph's mesoscopic structures. We provide the models with the original graph and the graph represented at the characteristic resolution scale and compare them to models that can only access the original graph. Our results showed that models with access to both the original graph and the characteristic scale graph can achieve statistically significant improvements in test accuracy.

Antonio Purificato, Maria Sofia Bucarelli, Andrea Bacciu et al.

Human annotation cost and time remain significant bottlenecks in Natural Language Processing (NLP), with test data annotation being particularly expensive due to the stringent requirement for low-error and high-quality labels necessary for reliable model evaluation. Traditional approaches require annotating entire test sets, leading to substantial resource requirements. Active Testing is a framework that selects the most informative test samples for annotation. Given a labeling budget, it aims to choose the subset that best estimates model performance while minimizing cost and human effort. In this work, we formalize Active Testing in NLP and we conduct an extensive benchmarking of existing approaches across 18 datasets and 4 embedding strategies spanning 4 different NLP tasks. The experiments show annotation reductions of up to 95%, with performance estimation accuracy difference from the full test set within 1%. Our analysis reveals variations in method effectiveness across different data characteristics and task types, with no single approach emerging as universally superior. Lastly, to address the limitation of requiring a predefined annotation budget in existing sample selection strategies, we introduce an adaptive stopping criterion that automatically determines the optimal number of samples.

Giulio D'Erasmo, Cesare Campagnano, Antonio Mallia et al.

High-dimensional dense embeddings have become central to modern Information Retrieval, but many dimensions are noisy or redundant. Recently proposed DIME (Dimension IMportance Estimation), provides query-dependent scores to identify informative components of embeddings. DIME relies on a costly grid search to select a priori a dimensionality for all the query corpus's embeddings. Our work provides a statistically grounded criterion that directly identifies the optimal set of dimensions for each query at inference time. Experiments confirm achieving parity of effectiveness and reduces embedding size by an average of $\sim50\%$ across different models and datasets at inference time.

Flavio Giorgi, Cesare Campagnano, Fabrizio Silvestri et al.

Explainable Artificial Intelligence (XAI) has emerged as a critical area of research to unravel the opaque inner logic of (deep) machine learning models. Among the various XAI techniques proposed in the literature, counterfactual explanations stand out as one of the most promising approaches. However, these "what-if" explanations are frequently complex and technical, making them difficult for non-experts to understand and, more broadly, challenging for humans to interpret. To bridge this gap, in this work, we exploit the power of open-source Large Language Models to generate natural language explanations when prompted with valid counterfactual instances produced by state-of-the-art explainers for graph-based models. Experiments across several graph datasets and counterfactual explainers show that our approach effectively produces accurate natural language representations of counterfactual instances, as demonstrated by key performance metrics.

Dario Loi, Elena Maria Muià, Federico Siciliano et al.

We present AutoBench, a fully automated and self-sustaining framework for evaluating Large Language Models (LLMs) through reciprocal peer assessment. This paper provides a rigorous scientific validation of the AutoBench methodology, originally developed as an open-source project by eZecute S.R.L.. Unlike static benchmarks that suffer from test-set contamination and limited adaptability, AutoBench dynamically generates novel evaluation tasks while models alternately serve as question generators, contestants, and judges across diverse domains. An iterative weighting mechanism amplifies the influence of consistently reliable evaluators, aggregating peer judgments into consensus-based rankings that reflect collective model agreement. Our experiments demonstrate strong correlations with established benchmarks including MMLU-Pro and GPQA (respectively 78\% and 63\%), validating this peer-driven evaluation paradigm. The multi-judge design significantly outperforms single-judge baselines, confirming that distributed evaluation produces more robust and human-consistent assessments. AutoBench offers a scalable, contamination-resistant alternative to static benchmarks for the continuous evaluation of evolving language models.

Andrea Giuseppe Di Francesco, Francesco Caso, Maria Sofia Bucarelli et al.

The message-passing mechanism underlying Graph Neural Networks (GNNs) is not naturally suited for heterophilic datasets, where adjacent nodes often have different labels. Most solutions to this problem remain confined to the task of node classification. In this article, we focus on the valuable task of link prediction under heterophily, an interesting problem for recommendation systems, social network analysis, and other applications. GNNs like GRAFF have improved node classification under heterophily by incorporating physics biases in the architecture. Similarly, we propose GRAFF-LP, an extension of GRAFF for link prediction. We show that GRAFF-LP effectively discriminates existing from non-existing edges by learning implicitly to separate the edge gradients. Based on this information, we propose a new readout function inspired by physics. Remarkably, this new function not only enhances the performance of GRAFF-LP but also improves that of other baseline models, leading us to reconsider how every link prediction experiment has been conducted so far. Finally, we provide evidence that even simple GNNs did not experience greater difficulty in predicting heterophilic links compared to homophilic ones. This leads us to believe in the necessity for heterophily measures specifically tailored for link prediction, distinct from those used in node classification. The code and appendix are available at https://github.com/difra100/Link_Prediction_with_PIGNN_IJCNN.

Giovanni Trappolini, Andrea Santilli, Emanuele Rodolà et al.

The rise in loosely-structured data available through text, images, and other modalities has called for new ways of querying them. Multimedia Information Retrieval has filled this gap and has witnessed exciting progress in recent years. Tasks such as search and retrieval of extensive multimedia archives have undergone massive performance improvements, driven to a large extent by recent developments in multimodal deep learning. However, methods in this field remain limited in the kinds of queries they support and, in particular, their inability to answer database-like queries. For this reason, inspired by recent work on neural databases, we propose a new framework, which we name Multimodal Neural Databases (MMNDBs). MMNDBs can answer complex database-like queries that involve reasoning over different input modalities, such as text and images, at scale. In this paper, we present the first architecture able to fulfill this set of requirements and test it with several baselines, showing the limitations of currently available models. The results show the potential of these new techniques to process unstructured data coming from different modalities, paving the way for future research in the area. Code to replicate the experiments will be released at https://github.com/GiovanniTRA/MultimodalNeuralDatabases

Mihajlo Grbovic, Nemanja Djuric, Vladan Radosavljevic et al.

Sponsored search represents a major source of revenue for web search engines. This popular advertising model brings a unique possibility for advertisers to target users' immediate intent communicated through a search query, usually by displaying their ads alongside organic search results for queries deemed relevant to their products or services. However, due to a large number of unique queries it is challenging for advertisers to identify all such relevant queries. For this reason search engines often provide a service of advanced matching, which automatically finds additional relevant queries for advertisers to bid on. We present a novel advanced matching approach based on the idea of semantic embeddings of queries and ads. The embeddings were learned using a large data set of user search sessions, consisting of search queries, clicked ads and search links, while utilizing contextual information such as dwell time and skipped ads. To address the large-scale nature of our problem, both in terms of data and vocabulary size, we propose a novel distributed algorithm for training of the embeddings. Finally, we present an approach for overcoming a cold-start problem associated with new ads and queries. We report results of editorial evaluation and online tests on actual search traffic. The results show that our approach significantly outperforms baselines in terms of relevance, coverage, and incremental revenue. Lastly, we open-source learned query embeddings to be used by researchers in computational advertising and related fields.

Antonio Andrea Gargiulo, Donato Crisostomi, Maria Sofia Bucarelli et al.

Task Arithmetic has emerged as a simple yet effective method to merge models without additional training. However, by treating entire networks as flat parameter vectors, it overlooks key structural information and is susceptible to task interference. In this paper, we study task vectors at the layer level, focusing on task layer matrices and their singular value decomposition. In particular, we concentrate on the resulting singular vectors, which we refer to as Task Singular Vectors (TSV). Recognizing that layer task matrices are often low-rank, we propose TSV-Compress (TSV-C), a simple procedure that compresses them to 10% of their original size while retaining 99% of accuracy. We further leverage this low-rank space to define a new measure of task interference based on the interaction of singular vectors from different tasks. Building on these findings, we introduce TSV-Merge (TSV-M), a novel model merging approach that combines compression with interference reduction, significantly outperforming existing methods.

Marco De Nadai, Francesco Fabbri, Paul Gigioli et al.

In the ever-evolving digital audio landscape, Spotify, well-known for its music and talk content, has recently introduced audiobooks to its vast user base. While promising, this move presents significant challenges for personalized recommendations. Unlike music and podcasts, audiobooks, initially available for a fee, cannot be easily skimmed before purchase, posing higher stakes for the relevance of recommendations. Furthermore, introducing a new content type into an existing platform confronts extreme data sparsity, as most users are unfamiliar with this new content type. Lastly, recommending content to millions of users requires the model to react fast and be scalable. To address these challenges, we leverage podcast and music user preferences and introduce 2T-HGNN, a scalable recommendation system comprising Heterogeneous Graph Neural Networks (HGNNs) and a Two Tower (2T) model. This novel approach uncovers nuanced item relationships while ensuring low latency and complexity. We decouple users from the HGNN graph and propose an innovative multi-link neighbor sampler. These choices, together with the 2T component, significantly reduce the complexity of the HGNN model. Empirical evaluations involving millions of users show significant improvement in the quality of personalized recommendations, resulting in a +46% increase in new audiobooks start rate and a +23% boost in streaming rates. Intriguingly, our model's impact extends beyond audiobooks, benefiting established products like podcasts.

Daniel Trippa, Cesare Campagnano, Maria Sofia Bucarelli et al.

Machine Unlearning, the process of selectively eliminating the influence of certain data examples used during a model's training, has gained significant attention as a means for practitioners to comply with recent data protection regulations. However, existing unlearning methods face critical drawbacks, including their prohibitively high cost, often associated with a large number of hyperparameters, and the limitation of forgetting only relatively small data portions. This often makes retraining the model from scratch a quicker and more effective solution. In this study, we introduce Gradient-based and Task-Agnostic machine Unlearning ($\nabla τ$), an optimization framework designed to remove the influence of a subset of training data efficiently. It applies adaptive gradient ascent to the data to be forgotten while using standard gradient descent for the remaining data. $\nabla τ$ offers multiple benefits over existing approaches. It enables the unlearning of large sections of the training dataset (up to 30%). It is versatile, supporting various unlearning tasks (such as subset forgetting or class removal) and applicable across different domains (images, text, etc.). Importantly, $\nabla τ$ requires no hyperparameter adjustments, making it a more appealing option than retraining the model from scratch. We evaluate our framework's effectiveness using a set of well-established Membership Inference Attack metrics, demonstrating up to 10% enhancements in performance compared to state-of-the-art methods without compromising the original model's accuracy.

Jacopo Lenti, Fabrizio Silvestri, Gianmarco De Francisci Morales

Despite the frequent use of agent-based models (ABMs) for studying social phenomena, parameter estimation remains a challenge, often relying on costly simulation-based heuristics. This work uses variational inference to estimate the parameters of an opinion dynamics ABM, by transforming the estimation problem into an optimization task that can be solved directly. Our proposal relies on probabilistic generative ABMs (PGABMs): we start by synthesizing a probabilistic generative model from the ABM rules. Then, we transform the inference process into an optimization problem suitable for automatic differentiation. In particular, we use the Gumbel-Softmax reparameterization for categorical agent attributes and stochastic variational inference for parameter estimation. Furthermore, we explore the trade-offs of using variational distributions with different complexity: normal distributions and normalizing flows. We validate our method on a bounded confidence model with agent roles (leaders and followers). Our approach estimates both macroscopic (bounded confidence intervals and backfire thresholds) and microscopic ($200$ categorical, agent-level roles) more accurately than simulation-based and MCMC methods. Consequently, our technique enables experts to tune and validate their ABMs against real-world observations, thus providing insights into human behavior in social systems via data-driven analysis.

Luca Zhou, Daniele Solombrino, Donato Crisostomi et al.

Model merging has emerged as a cost-efficient approximation to multitask learning. Among merging strategies, task arithmetic is notable for its simplicity and effectiveness. In this work, we provide a theoretical motivation for task vectors by highlighting that, under single-epoch full-batch gradient descent, they are equivalent to multitask gradients. This insight leads us to reinterpret model merging as a single step in an iterative procedure that Alternates between Tuning and Merging (ATM). We propose two applications of ATM: (1) as an alternative to multitask learning in scenarios where data sharing is restricted (e.g., federated settings), and (2) as a lightweight refinement step to improve existing model merging methods using a small validation set. Experiments across diverse vision tasks demonstrate the effectiveness of ATM.

Florin Cuconasu, Simone Filice, Guy Horowitz et al.

Retrieval Augmented Generation enhances LLM accuracy by adding passages retrieved from an external corpus to the LLM prompt. This paper investigates how positional bias - the tendency of LLMs to weight information differently based on its position in the prompt - affects not only the LLM's capability to capitalize on relevant passages, but also its susceptibility to distracting passages. Through extensive experiments on three benchmarks, we show how state-of-the-art retrieval pipelines, while attempting to retrieve relevant passages, systematically bring highly distracting ones to the top ranks, with over 60% of queries containing at least one highly distracting passage among the top-10 retrieved passages. As a result, the impact of the LLM positional bias, which in controlled settings is often reported as very prominent by related works, is actually marginal in real scenarios since both relevant and distracting passages are, in turn, penalized. Indeed, our findings reveal that sophisticated strategies that attempt to rearrange the passages based on LLM positional preferences do not perform better than random shuffling.

Ziheng Chen, Jia Wang, Jun Zhuang et al.

The right to be forgotten (RTBF) seeks to safeguard individuals from the enduring effects of their historical actions by implementing machine-learning techniques. These techniques facilitate the deletion of previously acquired knowledge without requiring extensive model retraining. However, they often overlook a critical issue: unlearning processes bias. This bias emerges from two main sources: (1) data-level bias, characterized by uneven data removal, and (2) algorithm-level bias, which leads to the contamination of the remaining dataset, thereby degrading model accuracy. In this work, we analyze the causal factors behind the unlearning process and mitigate biases at both data and algorithmic levels. Typically, we introduce an intervention-based approach, where knowledge to forget is erased with a debiased dataset. Besides, we guide the forgetting procedure by leveraging counterfactual examples, as they maintain semantic data consistency without hurting performance on the remaining dataset. Experimental results demonstrate that our method outperforms existing machine unlearning baselines on evaluation metrics.

Alex J. Chan, José Luis Redondo García, Fabrizio Silvestri et al.

Content moderation on a global scale must navigate a complex array of local cultural distinctions, which can hinder effective enforcement. While global policies aim for consistency and broad applicability, they often miss the subtleties of regional language interpretation, cultural beliefs, and local legislation. This work introduces a flexible framework that enhances foundation language models with cultural knowledge. Our approach involves fine-tuning encoder-decoder models on media-diet data to capture cultural nuances, and applies a continued training regime to effectively integrate these models into a content moderation pipeline. We evaluate this framework in a case study of an online podcast platform with content spanning various regions. The results show that our culturally adapted models improve the accuracy of local violation detection and offer explanations that align more closely with regional cultural norms. Our findings reinforce the need for an adaptable content moderation approach that remains flexible in response to the diverse cultural landscapes it operates in and represents a step towards a more equitable and culturally sensitive framework for content moderation, demonstrating what is achievable in this domain.

Maria Sofia Bucarelli, Giuseppe Alessio D'Inverno, Monica Bianchini et al.

In the context of deep learning models, attention has recently been paid to studying the surface of the loss function in order to better understand training with methods based on gradient descent. This search for an appropriate description, both analytical and topological, has led to numerous efforts to identify spurious minima and characterize gradient dynamics. Our work aims to contribute to this field by providing a topological measure to evaluate loss complexity in the case of multilayer neural networks. We compare deep and shallow architectures with common sigmoidal activation functions by deriving upper and lower bounds on the complexity of their loss function and revealing how that complexity is influenced by the number of hidden units, training models, and the activation function used. Additionally, we found that certain variations in the loss function or model architecture, such as adding an $\ell_2$ regularization term or implementing skip connections in a feedforward network, do not affect loss topology in specific cases.

Antonio Purificato, Maria Sofia Bucarelli, Anil Kumar Nelakanti et al.

Reliably labelling data typically requires annotations from multiple human workers. However, humans are far from being perfect. Hence, it is a common practice to aggregate labels gathered from multiple annotators to make a more confident estimate of the true label. Among many aggregation methods, the simple and well known Majority Vote (MV) selects the class label polling the highest number of votes. However, despite its importance, the optimality of MV's label aggregation has not been extensively studied. We address this gap in our work by characterising the conditions under which MV achieves the theoretically optimal lower bound on label estimation error. Our results capture the tolerable limits on annotation noise under which MV can optimally recover labels for a given class distribution. This certificate of optimality provides a more principled approach to model selection for label aggregation as an alternative to otherwise inefficient practices that sometimes include higher experts, gold labels, etc., that are all marred by the same human uncertainty despite huge time and monetary costs. Experiments on both synthetic and real world data corroborate our theoretical findings.

Ludovica Schaerf, Andrea Alfarano, Fabrizio Silvestri et al.

Recent advances in diffusion models for image generation have led to detailed examinations of several components within the U-Net architecture for image editing. While previous studies have focused on the bottleneck layer (h-space), cross-attention, self-attention, and decoding layers, the overall role of the skip connections of the U-Net itself has not been specifically addressed. We conduct thorough analyses on the role of the skip connections and find that the residual connections passed by the third encoder block carry most of the spatial information of the reconstructed image, splitting the content from the style, passed by the remaining stream in the opposed decoding layer. We show that injecting the representations from this block can be used for text-based editing, precise modifications, and style transfer. We compare our method, SkipInject, to state-of-the-art style transfer and image editing methods and demonstrate that our method obtains the best content alignment and optimal structural preservation tradeoff.

Luca Zhou, Daniele Solombrino, Donato Crisostomi et al.

Task arithmetic has emerged as a simple yet powerful technique for model merging, enabling the combination of multiple finetuned models into one. Despite its empirical success, a clear theoretical explanation of why and when it works is lacking. This paper provides a rigorous theoretical foundation for task arithmetic by establishing a connection between task vectors and gradients of the task losses. We show that under standard gradient descent, a task vector generated from one epoch of finetuning is exactly equivalent to the negative gradient of the loss, scaled by the learning rate. For the practical multi-epoch setting, we prove that this equivalence holds approximately, with a second-order error term that we explicitly bound for feed-forward networks. Our empirical analysis across seven vision benchmarks corroborates our theory, demonstrating that the first-epoch gradient dominates the finetuning trajectory in both norm and direction. A key implication is that merging models finetuned for only a single epoch often yields performance comparable to merging fully converged models. These findings reframe task arithmetic as a form of approximate multitask learning, providing a clear rationale for its effectiveness and highlighting the critical role of early training dynamics in model merging.

Filippo Betello, Antonio Purificato, Vittoria Vineis et al.

The environmental impact of Artificial Intelligence (AI) is emerging as a significant global concern, particularly regarding model training. In this paper, we introduce GREEN (Guided Recommendations of Energy-Efficient Networks), a novel, inference-time approach for recommending Pareto-optimal AI model configurations that optimize validation performance and energy consumption across diverse AI domains and tasks. Our approach directly addresses the limitations of current eco-efficient neural architecture search methods, which are often restricted to specific architectures or tasks. Central to this work is EcoTaskSet, a dataset comprising training dynamics from over 1767 experiments across computer vision, natural language processing, and recommendation systems using both widely used and cutting-edge architectures. Leveraging this dataset and a prediction model, our approach demonstrates effectiveness in selecting the best model configuration based on user preferences. Experimental results show that our method successfully identifies energy-efficient configurations while ensuring competitive performance.

Flavio Giorgi, Fabrizio Silvestri, Gabriele Tolomei

Counterfactual explanations have emerged as a powerful tool to unveil the opaque decision-making processes of graph neural networks (GNNs). However, existing techniques primarily focus on edge modifications, often overlooking the crucial role of node feature perturbations in shaping model predictions. To address this limitation, we propose COMBINEX, a novel GNN explainer that generates counterfactual explanations for both node and graph classification tasks. Unlike prior methods, which treat structural and feature-based changes independently, COMBINEX optimally balances modifications to edges and node features by jointly optimizing these perturbations. This unified approach ensures minimal yet effective changes required to flip a model's prediction, resulting in realistic and interpretable counterfactuals. Additionally, COMBINEX seamlessly handles both continuous and discrete node features, enhancing its versatility across diverse datasets and GNN architectures. Extensive experiments on real-world datasets and various GNN architectures demonstrate the effectiveness and robustness of our approach over existing baselines.

Farooq Ahmad Wani, Maria Sofia Bucarelli, Andrea Giuseppe Di Francesco et al.

Graph Neural Networks (GNNs) are powerful at solving graph classification tasks, yet applied problems often contain noisy labels. In this work, we study GNN robustness to label noise, demonstrate GNN failure modes when models struggle to generalise on low-order graphs, low label coverage, or when a model is over-parameterized. We establish both empirical and theoretical links between GNN robustness and the reduction of the total Dirichlet Energy of learned node representations, which encapsulates the hypothesized GNN smoothness inductive bias. Finally, we introduce two training strategies to enhance GNN robustness: (1) by incorporating a novel inductive bias in the weight matrices through the removal of negative eigenvalues, connected to Dirichlet Energy minimization; (2) by extending to GNNs a loss penalty that promotes learned smoothness. Importantly, neither approach negatively impacts performance in noise-free settings, supporting our hypothesis that the source of GNNs robustness is their smoothness inductive bias.

Valeria Ruscio, Umberto Nanni, Fabrizio Silvestri

This paper studies how Transformer models with Rotary Position Embeddings (RoPE) develop emergent, wavelet-like properties that compensate for the positional encoding's theoretical limitations. Through an analysis spanning model scales, architectures, and training checkpoints, we show that attention heads evolve to implement multi-resolution processing analogous to wavelet transforms. We demonstrate that this scale-invariant behavior is unique to RoPE, emerges through distinct evolutionary phases during training, and statistically adheres to the fundamental uncertainty principle. Our findings suggest that the effectiveness of modern Transformers stems from their remarkable ability to spontaneously develop optimal, multi-resolution decompositions to address inherent architectural constraints.

Alessio Borgi, Fabrizio Silvestri, Pietro Liò

Sheaf Neural Networks equip graph structures with a cellular sheaf: a geometric structure which assigns local vector spaces (stalks) and a linear learnable restriction/transport maps to nodes and edges, yielding an edge-aware inductive bias that handles heterophily and limits oversmoothing. However, common Neural Sheaf Diffusion implementations rely on SVD-based sheaf normalization and dense per-edge restriction maps, which scale with stalk dimension, require frequent Laplacian rebuilds, and yield brittle gradients. To address these limitations, we introduce Polynomial Neural Sheaf Diffusion (PolyNSD), a new sheaf diffusion approach whose propagation operator is a degree-K polynomial in a normalised sheaf Laplacian, evaluated via a stable three-term recurrence on a spectrally rescaled operator. This provides an explicit K-hop receptive field in a single layer (independently of the stalk dimension), with a trainable spectral response obtained as a convex mixture of K+1 orthogonal polynomial basis responses. PolyNSD enforces stability via convex mixtures, spectral rescaling, and residual/gated paths, reaching new state-of-the-art results on both homophilic and heterophilic benchmarks, inverting the Neural Sheaf Diffusion trend by obtaining these results with just diagonal restriction maps, decoupling performance from large stalk dimension, while reducing runtime and memory requirements.

Ali Ghasemi, Farooq Ahmad Wani, Maria Sofia Bucarelli et al.

Recent work has extended the invariance principle for out-of-distribution (OOD) generalization from Euclidean to graph data, where challenges arise due to complex structures and diverse distribution shifts in node attributes and topology. To handle these, Chen et al. proposed CIGA (Chen et al., 2022b), which uses causal modeling and an information-theoretic objective to extract a single invariant subgraph capturing causal features. However, this single-subgraph focus can miss multiple causal patterns. Liu et al. (2025) addressed this with SuGAr, which learns and aggregates diverse invariant subgraphs via a sampler and diversity regularizer, improving robustness but still relying on simple uniform or greedy aggregation. To overcome this, the proposed PISA framework introduces a dynamic MLP-based aggregation that prioritizes and combines subgraph representations more effectively. Experiments on 15 datasets, including DrugOOD (Ji et al., 2023), show that PISA achieves up to 5% higher classification accuracy than prior methods.

Mostafa Mozafari, Farooq Ahmad Wani, Maria Sofia Bucarelli et al.

Corrupted training data are ubiquitous. Corrective Machine Unlearning (CMU) seeks to remove the influence of such corruption post-training. Prior CMU typically assumes access to identified corrupted training samples (a "forget set"). However, in many real-world scenarios the training data are no longer accessible. We formalize source-free CMU, where the original training data are unavailable and, consequently, no forget set of identified corrupted training samples can be specified. Instead, we assume a small proxy (surrogate) set of corrupted samples that reflect the suspected corruption type without needing to be the original training samples. In this stricter setting, methods relying on forget set are ineffective or narrow in scope. We introduce Corrective Unlearning in Task Space (CUTS), a lightweight weight space correction method guided by the proxy set using task arithmetic principles. CUTS treats the clean and the corruption signal as distinct tasks. Specifically, we briefly fine-tune the corrupted model on the proxy to amplify the corruption mechanism in the weight space, compute the difference between the corrupted and fine-tuned weights as a proxy task vector, and subtract a calibrated multiple of this vector to cancel the corruption. Without access to clean data or a forget set, CUTS recovers a large fraction of the lost utility under label noise and, for backdoor triggers, nearly eliminates the attack with minimal damage to utility, outperforming state-of-the-art specialized CMU methods in source-free setting.

Giovanni Trappolini, Florin Cuconasu, Simone Filice et al.

Traditional Information Retrieval (IR) metrics, such as nDCG, MAP, and MRR, assume that human users sequentially examine documents with diminishing attention to lower ranks. This assumption breaks down in Retrieval Augmented Generation (RAG) systems, where search results are consumed by Large Language Models (LLMs), which, unlike humans, process all retrieved documents as a whole rather than sequentially. Additionally, traditional IR metrics do not account for related but irrelevant documents that actively degrade generation quality, rather than merely being ignored. Due to these two major misalignments, namely human vs. machine position discount and human relevance vs. machine utility, classical IR metrics do not accurately predict RAG performance. We introduce a utility-based annotation schema that quantifies both the positive contribution of relevant passages and the negative impact of distracting ones. Building on this foundation, we propose UDCG (Utility and Distraction-aware Cumulative Gain), a metric using an LLM-oriented positional discount to directly optimize the correlation with the end-to-end answer accuracy. Experiments on five datasets and six LLMs demonstrate that UDCG improves correlation by up to 36% compared to traditional metrics. Our work provides a critical step toward aligning IR evaluation with LLM consumers and enables more reliable assessment of RAG components

Matteo Silvestri, Flavio Giorgi, Fabrizio Silvestri et al.

Large Language Models (LLMs) are increasingly evaluated on their ability to reason over structured data, yet such assessments often overlook a crucial confound: dataset contamination. In this work, we investigate whether LLMs exhibit prior knowledge of widely used tabular benchmarks such as Adult Income, Titanic, and others. Through a series of controlled probing experiments, we reveal that contamination effects emerge exclusively for datasets containing strong semantic cues-for instance, meaningful column names or interpretable value categories. In contrast, when such cues are removed or randomized, performance sharply declines to near-random levels. These findings suggest that LLMs' apparent competence on tabular reasoning tasks may, in part, reflect memorization of publicly available datasets rather than genuine generalization. We discuss implications for evaluation protocols and propose strategies to disentangle semantic leakage from authentic reasoning ability in future LLM assessments.

Maximo Eduardo Rulli, Simone Petruzzi, Edoardo Michielon et al.

Masked Diffusion Language Models (DLMs) have recently emerged as a promising alternative to traditional Autoregressive Models (ARMs). DLMs employ transformer encoders with bidirectional attention, enabling parallel token generation while maintaining competitive performance. Although their efficiency and effectiveness have been extensively studied, the internal mechanisms that govern DLMs remain largely unexplored. In this work, we conduct an empirical analysis of DLM attention patterns, focusing on the attention sinking phenomenon, an effect previously observed in various transformer-based architectures. Our findings reveal that DLMs also exhibit attention sinks, but with distinct characteristics. First, unlike in ARMs, the sink positions in DLMs tend to shift throughout the generation process, displaying a dynamic behaviour. Second, while ARMs are highly sensitive to the removal of attention sinks, DLMs remain robust: masking sinks leads to only a minor degradation in performance. These results provide new insights into the inner workings of diffusion-based language models and highlight fundamental differences in how they allocate and utilize attention compared to autoregressive models.

Gavriel Di Nepi, Federico Siciliano, Fabrizio Silvestri

By the end of 2024, Google researchers introduced Titans: Learning at Test Time, a neural memory model achieving strong empirical results across multiple tasks. However, the lack of publicly available code and ambiguities in the original description hinder reproducibility. In this work, we present a lightweight reimplementation of Titans and conduct a comprehensive evaluation on Masked Language Modeling, Time Series Forecasting, and Recommendation tasks. Our results reveal that Titans does not always outperform established baselines due to chunking. However, its Neural Memory component consistently improves performance compared to attention-only models. These findings confirm the model's innovative potential while highlighting its practical limitations and raising questions for future research.