Maurizio Gabbrielli

Stefano Pio Zingaro, Giuseppe Lisanti, Maurizio Gabbrielli

In this paper, we propose to exploit the side-tuning framework for multimodal document classification. Side-tuning is a methodology for network adaptation recently introduced to solve some of the problems related to previous approaches. Thanks to this technique it is actually possible to overcome model rigidity and catastrophic forgetting of transfer learning by fine-tuning. The proposed solution uses off-the-shelf deep learning architectures leveraging the side-tuning framework to combine a base model with a tandem of two side networks. We show that side-tuning can be successfully employed also when different data sources are considered, e.g. text and images in document classification. The experimental results show that this approach pushes further the limit for document classification accuracy with respect to the state of the art.

Andrea Gurioli, Davide D'Ascenzo, Federico Pennino et al.

Large language models (LLMs) for code completion and generation are increasingly used in software development, yet they may reproduce training examples verbatim and without authorship attribution, raising legal and ethical concerns around plagiarism and license compliance. Classical fingerprint-based plagiarism detectors based on fingerprinting, such as Winnowing, remain highly effective, yet the inspection requires comparing fragments of code to the entire training set, and their linear-time search makes them impractical for the billion-scale corpora used to train modern code LLMs. To bridge this gap, we introduce SOURCETRACKER, a 300M-parameter encoder tailored for code retrieval, together with a hybrid two-stage provenance-tracking pipeline HYBRIDSOURCETRACKER (HST). HST first narrows down a small set of candidate snippets via vector search, then re-ranks those candidates using Winnowing on exact fingerprints. We train and evaluate our system on a 10M-snippet subset of the THESTACKV2 dataset, with both verbatim and adapted snippets that emulate realistic identifier renaming. On an in vitro 100k-snippet search space with adapted queries, our hybrid approach reaches a mean reciprocal rank on par with Winnowing for 30-token fragments. Then, starting from windows >= 60 tokens, it consistently over-performs by up to 5.4% while preserving logarithmic-time query complexity. In a complementary evaluation using an LLM-based judge, we find that many retrieved snippets not labeled as ground truth are still highly similar to the expected sources, particularly with longer context windows, and thus remain useful for end users. Overall, our results demonstrate that integrating vector search with fingerprinting enables scalable, high-precision provenance tracking for code produced by LLMs.

Massimo Rondelli, Francesco Pivi, Maurizio Gabbrielli

Automatic generation of executable Blender code from natural language remains challenging, with state-of-the-art LLMs producing frequent syntactic errors and geometrically inconsistent objects. We present BlenderRAG, a retrieval-augmented generation system that operates on a curated multimodal dataset of 500 expert-validated examples (text, code, image) across 50 object categories. By retrieving semantically similar examples during generation, BlenderRAG improves compilation success rates from 40.8% to 70.0% and semantic normalized alignment from 0.41 to 0.77 (CLIP similarity) across four state-of-the-art LLMs, without requiring fine-tuning or specialized hardware, making it immediately accessible for deployment. The dataset and code will be available at https://github.com/MaxRondelli/BlenderRAG.

Davide Evangelista, Elena Morotti, Francesco Pivi et al.

Diffusion-based posterior sampling (PS) is a leading framework for imaging inverse problems, combining learned priors with measurement constraints. Yet, its standard formulations rely on instantaneous data-consistent estimates, which induce temporal variability in the reverse dynamics. We reinterpret PS from a dynamical perspective, showing that the standard PS update corresponds to a first-order discretization of the diffusion dynamics plus a residual correction capturing the mismatch between the denoised prediction and the data-consistent estimate. A second-order discretization, however, naturally introduces a temporal correction based on the variation of consecutive estimates. Building on this, we propose LAMP, combining the second-order update with the residual correction characterizing a PS technique. LAMP thus inherits a lagged temporal correction, and it can be implemented as a modular plug-in over the PS backbone. We show that LAMP preserves the structure of a posterior sampler, and we perform a one-step risk analysis to characterize when LAMP improves the reverse transition via a bias-variance trade-off. Experiments across multiple imaging tasks demonstrate consistent improvements over strong baselines such as DiffPIR and DDRM, without increasing the number of denoising evaluations.

Andrea Gurioli, Federico Pennino, Maurizio Gabbrielli

Embedding-based code retrieval often suffers when encoders overfit to surface syntax. Prior work mitigates this by using LLMs to rephrase queries and corpora into a normalized style, but leaves two questions open: how much representational shift helps, and when is the per-query LLM call justified? We study a hierarchy of three rewriting strategies: stylistic rephrasing, NL-enriched PseudoCode, and full Natural-Language transcription, under joint query-corpus (QC, online) and corpus-only (C, offline) augmentation, across six CoIR benchmarks, five encoders, and three rewriters spanning independent model families (Qwen, DeepSeek, Mistral). We are the first to evaluate NL-enriched PseudoCode and snippet-level Natural Language as direct retrieval representations, rather than as transient intermediates. Full NL rewriting with QC yields the largest gains (+0.51 absolute NDCG@10 on CT-Contest for MoSE-18), while corpus-only rewriting degrades retrieval in 56 of 90 configurations, about 62%. We introduce two diagnostics, Delta H, token entropy, and Delta s, embedding cosine, and show that Delta H predicts retrieval gain under QC across all three rewriter families: pooled Spearman rho = +0.436, p < 0.001 on DeepSeek+Codestral; rho = +0.593 on Codestral alone; rho = +0.356 on Qwen. This establishes Delta H as a cheap, rewriter-agnostic proxy for deciding when rewriting pays off before running retrieval. Our analysis reframes LLM rewriting as a cost-benefit decision: it is most effective as a remediation layer for lightweight encoders on code-dominant queries, with diminishing returns for strong encoders or NL-heavy queries.

Bianca Raimondi, Maurizio Gabbrielli

The black-box nature of Large Language Models necessitates novel evaluation frameworks that transcend surface-level performance metrics. This study investigates the internal neural representations of cognitive complexity using Bloom's Taxonomy as a hierarchical lens. By analyzing high-dimensional activation vectors from different LLMs, we probe whether different cognitive levels, ranging from basic recall (Remember) to abstract synthesis (Create), are linearly separable within the model's residual streams. Our results demonstrate that linear classifiers achieve approximately 95% mean accuracy across all Bloom levels, providing strong evidence that cognitive level is encoded in a linearly accessible subspace of the model's representations. These findings provide evidence that the model resolves the cognitive difficulty of a prompt early in the forward pass, with representations becoming increasingly separable across layers.

Gaia Ebli, Bianca Raimondi, Maurizio Gabbrielli

The increasing integration of AI tools in education has led prior research to explore their impact on learning processes. Nevertheless, most existing studies focus on higher education and conventional instructional contexts, leaving open questions about how key learning factors are related in AI-mediated learning environments and how these relationships may vary across different age groups. Addressing these gaps, our work investigates whether four critical learning factors, experience, clarity, comfort, and motivation, maintain coherent interrelationships in AI-augmented educational settings, and how the structure of these relationships differs between middle and high school students. The study was conducted in authentic classroom contexts where students interacted with AI tools as part of programming learning activities to collect data on the four learning factors and students' perceptions. Using a multimethod quantitative analysis, which combined correlation analysis and text mining, we revealed markedly different dimensional structures between the two age groups. Middle school students exhibit strong positive correlations across all dimensions, indicating holistic evaluation patterns whereby positive perceptions in one dimension generalise to others. In contrast, high school students show weak or near-zero correlations between key dimensions, suggesting a more differentiated evaluation process in which dimensions are assessed independently. These findings reveal that perception dimensions actively mediate AI-augmented learning and that the developmental stage moderates their interdependencies. This work establishes a foundation for the development of AI integration strategies that respond to learners' developmental levels and account for age-specific dimensional structures in student-AI interactions.

Federico Pennino, Maurizio Gabbrielli

The standard paradigm for training deep learning models on sensor data assumes that more data is always better. However, raw sensor streams are often imbalanced and contain significant redundancy, meaning that not all data points contribute equally to model generalization. In this paper, we show that, in some cases, "less is more" when considering datasets. We do this by reframing the data selection problem: rather than tuning model hyperparameters, we fix the model and optimize the composition of the training data itself. We introduce a framework for discovering the optimal "training diet" from a large, unlabeled time series corpus. Our framework first uses a large-scale encoder and k-means clustering to partition the dataset into distinct, behaviorally consistent clusters. These clusters represent the fundamental 'ingredients' available for training. We then employ the Optuna optimization framework to search the high-dimensional space of possible data mixtures. For each trial, Optuna proposes a specific sampling ratio for each cluster, and a new training set is constructed based on this recipe. A smaller target model is then trained and evaluated. Our experiments reveal that this data-centric search consistently discovers data mixtures that yield models with significantly higher performance compared to baselines trained on the entire dataset. Specifically - evaluated on PMSM dataset - our method improved performance from a baseline MSE of 1.70 to 1.37, a 19.41% improvement.

Federico Pennino, Bianca Raimondi, Massimo Rondelli et al.

Generating accurate and executable code using large language models (LLMs) is challenging for languages with limited public training data compared to popular languages such as Python. This paper introduces a generalizable approach that uses small-scale code versions of the Qwen 2.5 model combined with Group Relative Policy Optimization (GRPO) to enable effective code generation through explicit reasoning steps, which is particularly beneficial for languages with smaller source code databases. Using Prolog as a representative use case -- given its limited online presence -- the initial model faced challenges in generating executable code. After some training steps, the model successfully produces logically consistent and syntactically accurate code by directly integrating reasoning-driven feedback into the reinforcement learning loop. Experimental evaluations using mathematical logic problem benchmarks illustrate significant improvements in reasoning quality, code accuracy, and logical correctness, underscoring the potential of this approach to benefit a wide range of programming languages lacking extensive training resources.

Bianca Raimondi, Maurizio Gabbrielli

Large Language Models (LLMs) have become essential in many Natural Language Processing (NLP) tasks, leveraging extensive pre-training and fine-tuning to achieve high accuracy. However, like humans, LLMs exhibit biases, particularly positional biases such as primacy and recency effects, which can influence the accuracy of the answers. The primacy effect-where items presented first are more likely to be remembered or selected-plays a key role in Multiple Choice Question Answering (MCQA), where the order of answer options can affect prediction outcomes. This study focuses on primacy bias in fine-tuned LLMs: We first show that fine-tuning amplifies this bias, probably due to exposure to human-like patterns. Hence, we strategically leverage this effect by reordering response options based on semantic similarity to the query, without requiring knowledge of the correct answer. Our experimental results show that this approach significantly improves performance in MCQA. More generally, our findings underscore the dual nature of biases as both challenges and opportunities, offering insights for bias-aware model design and NLP applications.

Francesco Pivi, Simone Gazza, Davide Evangelista et al.

Generative models based on flow matching have demonstrated remarkable success in various domains, yet they suffer from a fundamental limitation: the lack of interpretability in their intermediate generation steps. In fact these models learn to transform noise into data through a series of vector field updates, however the meaning of each step remains opaque. We address this problem by proposing a general framework constraining each flow step to be sampled from a known physical distribution. Flow trajectories are mapped to (and constrained to traverse) the equilibrium states of the simulated physical process. We implement this approach through the 2D Ising model in such a way that flow steps become thermal equilibrium points along a parametric cooling schedule. Our proposed architecture includes an encoder that maps discrete Ising configurations into a continuous latent space, a flow-matching network that performs temperature-driven diffusion, and a projector that returns to discrete Ising states while preserving physical constraints. We validate this framework across multiple lattice sizes, showing that it preserves physical fidelity while outperforming Monte Carlo generation in speed as the lattice size increases. In contrast with standard flow matching, each vector field represents a meaningful stepwise transition in the 2D Ising model's latent space. This demonstrates that embedding physical semantics into generative flows transforms opaque neural trajectories into interpretable physical processes.

Bianca Raimondi, Daniela Dalbagno, Maurizio Gabbrielli

Large language models (LLMs) have been shown to internalize human-like biases during finetuning, yet the mechanisms by which these biases manifest remain unclear. In this work, we investigated whether the well-known Knobe effect, a moral bias in intentionality judgements, emerges in finetuned LLMs and whether it can be traced back to specific components of the model. We conducted a Layer-Patching analysis across 3 open-weights LLMs and demonstrated that the bias is not only learned during finetuning but also localized in a specific set of layers. Surprisingly, we found that patching activations from the corresponding pretrained model into just a few critical layers is sufficient to eliminate the effect. Our findings offer new evidence that social biases in LLMs can be interpreted, localized, and mitigated through targeted interventions, without the need for model retraining.

Francesco Balzan, Pedro P. Santos, Maurizio Gabbrielli et al.

Human education transcends mere knowledge transfer, it relies on co-adaptation dynamics -- the mutual adjustment of teaching and learning strategies between agents. Despite its centrality, computational models of co-adaptive teacher-student interactions (T-SI) remain underdeveloped. We argue that this gap impedes Educational Science in testing and scaling contextual insights across diverse settings, and limits the potential of Machine Learning systems, which struggle to emulate and adaptively support human learning processes. To address this, we present a computational T-SI model that integrates contextual insights on human education into a testable framework. We use the model to evaluate diverse T-SI strategies in a realistic synthetic classroom setting, simulating student groups with unequal access to sensory information. Results show that strategies incorporating co-adaptation principles (e.g., bidirectional agency) outperform unilateral approaches (i.e., where only the teacher or the student is active), improving the learning outcomes for all learning types. Beyond the testing and scaling of context-dependent educational insights, our model enables hypothesis generation in controlled yet adaptable environments. This work bridges non-computational theories of human education with scalable, inclusive AI in Education systems, providing a foundation for equitable technologies that dynamically adapt to learner needs.

Andrea Gurioli, Federico Pennino, João Monteiro et al.

Deploying language models often requires navigating accuracy vs. performance trade-offs to meet latency constraints while preserving utility. Traditional model distillation reduces size but incurs substantial costs through training separate models. We introduce ModularStarEncoder (MoSE), a 1-billion-parameter multi-exit encoder for code retrieval and classification that employs a novel Self-Distillation mechanism. This approach significantly enhances lower-layer representations, enabling flexible deployment of different model portions with favorable performance trade-offs. Our architecture improves text-to-code and code-to-code search by targeting specific encoder layers as exit heads, where higher layers guide earlier ones during training-improving intermediate representations at minimal additional cost. We further enhance MoSE with a repository-level contextual loss that maximizes training context window utilization. Additionally, we release a new dataset created through code translation that extends text-to-code benchmarks with cross-language code-to-code pairs. Evaluations demonstrate the effectiveness of Self-Distillation as a principled approach to trading inference cost for accuracy across various code understanding tasks.

Bianca Raimondi, Saverio Giallorenzo, Maurizio Gabbrielli

In education, the capability of generating human-like text of Large Language Models (LLMs) inspired work on how they can increase the efficiency of learning and teaching. We study the affordability of these models for educators and students by investigating how LLMs answer multiple-choice questions (MCQs) with respect to hardware constraints and refinement techniques. We explore this space by using generic pre-trained LLMs (the 7B, 13B, and 70B variants of LLaMA-2) to answer 162 undergraduate-level MCQs from a course on Programming Languages (PL) -- the MCQ dataset is a contribution of this work, which we make publicly available. Specifically, we dissect how different factors, such as using readily-available material -- (parts of) the course's textbook -- for fine-tuning and quantisation (to decrease resource usage) can change the accuracy of the responses. The main takeaway is that smaller textbook-based fine-tuned models outperform generic larger ones (whose pre-training requires conspicuous resources), making the usage of LLMs for answering MCQs resource- and material-wise affordable.

Roberto Amadini, Maurizio Gabbrielli, Tong Liu et al.

Meta-solver approaches exploits a number of individual solvers to potentially build a better solver. To assess the performance of meta-solvers, one can simply adopt the metrics typically used for individual solvers (e.g., runtime or solution quality), or employ more specific evaluation metrics (e.g., by measuring how close the meta-solver gets to its virtual best performance). In this paper, based on some recently published works, we provide an overview of different performance metrics for evaluating (meta-)solvers, by underlying their strengths and weaknesses.

Francesca Del Bonifro, Maurizio Gabbrielli, Stefano Zacchiroli

Programming language detection is a common need in the analysis of large source code bases. It is supported by a number of existing tools that rely on several features, and most notably file extensions, to determine file types. We consider the problem of accurately detecting the type of files commonly found in software code bases, based solely on textual file content. Doing so is helpful to classify source code that lack file extensions (e.g., code snippets posted on the Web or executable scripts), to avoid misclassifying source code that has been recorded with wrong or uncommon file extensions, and also shed some light on the intrinsic recognizability of source code files. We propose a simple model that (a) use a language-agnostic word tokenizer for textual files, (b) group tokens in 1-/2-grams, (c) build feature vectors based on N-gram frequencies, and (d) use a simple fully connected neural network as classifier. As training set we use textual files extracted from GitHub repositories with at least 1000 stars, using existing file extensions as ground truth. Despite its simplicity the proposed model reaches 85% in our experiments for a relatively high number of recognized classes (more than 130 file types).

Tong Liu, Roberto Amadini, Jacopo Mauro et al.

SUNNY is an Algorithm Selection (AS) technique originally tailored for Constraint Programming (CP). SUNNY enables to schedule, from a portfolio of solvers, a subset of solvers to be run on a given CP problem. This approach has proved to be effective for CP problems, and its parallel version won many gold medals in the Open category of the MiniZinc Challenge -- the yearly international competition for CP solvers. In 2015, the ASlib benchmarks were released for comparing AS systems coming from disparate fields (e.g., ASP, QBF, and SAT) and SUNNY was extended to deal with generic AS problems. This led to the development of sunny-as2, an algorithm selector based on SUNNY for ASlib scenarios. A preliminary version of sunny-as2 was submitted to the Open Algorithm Selection Challenge (OASC) in 2017, where it turned out to be the best approach for the runtime minimization of decision problems. In this work, we present the technical advancements of sunny-as2, including: (i) wrapper-based feature selection; (ii) a training approach combining feature selection and neighbourhood size configuration; (iii) the application of nested cross-validation. We show how sunny-as2 performance varies depending on the considered AS scenarios, and we discuss its strengths and weaknesses. Finally, we also show how sunny-as2 improves on its preliminary version submitted to OASC.

Roberto Amadini, Maurizio Gabbrielli, Jacopo Mauro

In Constraint Programming (CP) a portfolio solver combines a variety of different constraint solvers for solving a given problem. This fairly recent approach enables to significantly boost the performance of single solvers, especially when multicore architectures are exploited. In this work we give a brief overview of the portfolio solver sunny-cp, and we discuss its performance in the MiniZinc Challenge---the annual international competition for CP solvers---where it won two gold medals in 2015 and 2016. Under consideration in Theory and Practice of Logic Programming (TPLP)

Roberto Amadini, Maurizio Gabbrielli, Jacopo Mauro

*** To appear in IJCAI 2015 proceedings *** In Constraint Programming (CP), a portfolio solver uses a variety of different solvers for solving a given Constraint Satisfaction / Optimization Problem. In this paper we introduce sunny-cp2: the first parallel CP portfolio solver that enables a dynamic, cooperative, and simultaneous execution of its solvers in a multicore setting. It incorporates state-of-the-art solvers, providing also a usable and configurable framework. Empirical results are very promising. sunny-cp2 can even outperform the performance of the oracle solver which always selects the best solver of the portfolio for a given problem.

Stefano Bistarelli, Maurizio Gabbrielli, Maria Chiara Meo et al.

We propose a timed and soft extension of Concurrent Constraint Programming. The time extension is based on the hypothesis of bounded asynchrony: the computation takes a bounded period of time and is measured by a discrete global clock. Action prefixing is then considered as the syntactic marker which distinguishes a time instant from the next one. Supported by soft constraints instead of crisp ones, tell and ask agents are now equipped with a preference (or consistency) threshold which is used to determine their success or suspension. In the paper we provide a language to describe the agents behavior, together with its operational and denotational semantics, for which we also prove the compositionality and correctness properties. After presenting a semantics using maximal parallelism of actions, we also describe a version for their interleaving on a single processor (with maximal parallelism for time elapsing). Coordinating agents that need to take decisions both on preference values and time events may benefit from this language. To appear in Theory and Practice of Logic Programming (TPLP).

Roberto Amadini, Maurizio Gabbrielli, Jacopo Mauro

*** To appear in Theory and Practice of Logic Programming (TPLP) *** Within the context of constraint solving, a portfolio approach allows one to exploit the synergy between different solvers in order to create a globally better solver. In this paper we present SUNNY: a simple and flexible algorithm that takes advantage of a portfolio of constraint solvers in order to compute --- without learning an explicit model --- a schedule of them for solving a given Constraint Satisfaction Problem (CSP). Motivated by the performance reached by SUNNY vs. different simulations of other state of the art approaches, we developed sunny-csp, an effective portfolio solver that exploits the underlying SUNNY algorithm in order to solve a given CSP. Empirical tests conducted on exhaustive benchmarks of MiniZinc models show that the actual performance of SUNNY conforms to the predictions. This is encouraging both for improving the power of CSP portfolio solvers and for trying to export them to fields such as Answer Set Programming and Constraint Logic Programming.

Roberto Amadini, Maurizio Gabbrielli, Jacopo Mauro

Recent research has shown that a single arbitrarily efficient solver can be significantly outperformed by a portfolio of possibly slower on-average solvers. The solver selection is usually done by means of (un)supervised learning techniques which exploit features extracted from the problem specification. In this paper we present an useful and flexible framework that is able to extract an extensive set of features from a Constraint (Satisfaction/Optimization) Problem defined in possibly different modeling languages: MiniZinc, FlatZinc or XCSP. We also report some empirical results showing that the performances that can be obtained using these features are effective and competitive with state of the art CSP portfolio techniques.