Gabriele Dominici

Gabriele Dominici, Pietro Barbiero, Lucie Charlotte Magister et al.

Multimodal learning is an essential paradigm for addressing complex real-world problems, where individual data modalities are typically insufficient to accurately solve a given modelling task. While various deep learning approaches have successfully addressed these challenges, their reasoning process is often opaque; limiting the capabilities for a principled explainable cross-modal analysis and any domain-expert intervention. In this paper, we introduce SHARCS (SHARed Concept Space) -- a novel concept-based approach for explainable multimodal learning. SHARCS learns and maps interpretable concepts from different heterogeneous modalities into a single unified concept-manifold, which leads to an intuitive projection of semantically similar cross-modal concepts. We demonstrate that such an approach can lead to inherently explainable task predictions while also improving downstream predictive performance. Moreover, we show that SHARCS can operate and significantly outperform other approaches in practically significant scenarios, such as retrieval of missing modalities and cross-modal explanations. Our approach is model-agnostic and easily applicable to different types (and number) of modalities, thus advancing the development of effective, interpretable, and trustworthy multimodal approaches.

Dario Fenoglio, Arianna Casanova, Francesco De Santis et al.

Concept-based models (CMs) enhance interpretability in deep learning by grounding predictions in human-understandable concepts. However, concept annotations are expensive to obtain and rarely available at scale within a single data source. Federated learning (FL) could alleviate this limitation by enabling cross-institutional training that leverages concept annotations distributed across multiple data owners. Yet, FL lacks interpretable modeling paradigms. Integrating CMs with FL is non-trivial: CMs assume a fixed concept space and a predefined model architecture, whereas real-world FL is heterogeneous and non-stationary, with institutions joining over time and bringing new supervision. In this work, we propose Federated Concept-based Models (F-CMs), a new methodology for deploying CMs in evolving FL settings. F-CMs aggregate concept-level information across institutions and efficiently adapt the model architecture in response to changes in the available concept supervision, while preserving institutional privacy. Empirically, F-CMs preserve the accuracy and intervention effectiveness of training settings with full concept supervision, while outperforming non-adaptive federated baselines. Notably, F-CMs enable interpretable inference on concepts not available to a given institution, a key novelty with respect to existing approaches.

Francesco Sovrano, Gabriele Dominici, Alberto Bacchelli

Prompt-induced cognitive biases are changes in a general-purpose AI (GPAI) system's decisions caused solely by biased wording in the input (e.g., framing, anchors), not task logic. In software engineering (SE) decision support (where problem statements and requirements are natural language) small phrasing shifts (e.g., popularity hints or outcome reveals) can push GPAI models toward suboptimal decisions. We study this with PROBE-SWE, a dynamic benchmark for SE that pairs biased and unbiased versions of the same SE dilemmas, controls for logic and difficulty, and targets eight SE-relevant biases (anchoring, availability, bandwagon, confirmation, framing, hindsight, hyperbolic discounting, overconfidence). We ask whether prompt engineering mitigates bias sensitivity in practice, focusing on actionable techniques that practitioners can apply off-the-shelf in real environments. Testing common strategies (e.g., chain-of-thought, self-debiasing) on cost-effective GPAI systems, we find no statistically significant reductions in bias sensitivity on a per-bias basis. We then adopt a Prolog-style view of the reasoning process: solving SE dilemmas requires making explicit any background axioms and inference assumptions (i.e., SE best practices) that are usually implicit in the prompt. So, we hypothesize that bias-inducing features short-circuit assumptions elicitation, pushing GPAI models toward biased shortcuts. Building on this, we introduce an end-to-end method that elicits best practices and injects axiomatic reasoning cues into the prompt before answering, reducing overall bias sensitivity by 51% on average (p < .001). Finally, we report a thematic analysis that surfaces linguistic patterns associated with heightened bias sensitivity, clarifying when GPAI use is less advisable for SE decision support and where to focus future countermeasures.

Nursena Koprucu, Meher Shashwat Nigam, Shicheng Xu et al.

Inspired by Geoffrey Hinton emphasis on generative modeling, To recognize shapes, first learn to generate them, we explore the use of 3D diffusion models for object classification. Leveraging the density estimates from these models, our approach, the Diffusion Classifier for 3D Objects (DC3DO), enables zero-shot classification of 3D shapes without additional training. On average, our method achieves a 12.5 percent improvement compared to its multiview counterparts, demonstrating superior multimodal reasoning over discriminative approaches. DC3DO employs a class-conditional diffusion model trained on ShapeNet, and we run inferences on point clouds of chairs and cars. This work highlights the potential of generative models in 3D object classification.

Francesco Sovrano, Gabriele Dominici, Marc Langheinrich

A key goal of explainable AI (XAI) is to express the decision logic of large language models (LLMs) in symbolic form and link it to internal mechanisms. Global rule-extraction methods typically learn symbolic surrogates without grounding rules in model circuitry, while mechanistic interpretability can connect behaviors to neuron sets but often depends on hand-crafted hypotheses and expensive neuron-level interventions. We introduce MechaRule, a pipeline that grounds rule extraction in LLM circuits by efficiently localizing sparse neurons called agonists, whose activation neutralization disrupts rule-related behaviors. MechaRule rests on two empirical observations. First, within a fixed baseline/flip regime, sparse agonist effects can be approximately monotone and saturating: a few dominant neuron activations can overtop weaker ones at coarse scales, while overlapping neurons flip many of the same examples. This motivates viewing localization as adaptive group testing driven by a regime-conditional strength predicate with confidence-guided conservative pruning, yielding Theta(k log(N/k) + k) interventions over N candidates when k << N neurons are agonists under the monotone-overtopping abstraction. Second, agonists emerge more reliably when ablations are verified through data splits aligned with close-to-faithful rule behavior; spectral splits remain a useful rule-free fallback, while unfaithful splits degrade localization. Empirically, overtopping appears mainly in learned, task-aligned regimes: on arithmetic and jailbreak tasks across Qwen2 and GPT-J, MechaRule recalls 96.8% of high-effect brute-force agonists in completed comparisons, and suppressing localized agonists reduces arithmetic accuracy and jailbreak success by up to 71.1% and 8.8%, respectively.

Francesco Bombassei De Bona, Gabriele Dominici, Tim Miller et al.

As AI becomes fundamental in sectors like healthcare, explainable AI (XAI) tools are essential for trust and transparency. However, traditional user studies used to evaluate these tools are often costly, time consuming, and difficult to scale. In this paper, we explore the use of Large Language Models (LLMs) to replicate human participants to help streamline XAI evaluation. We reproduce a user study comparing counterfactual and causal explanations, replicating human participants with seven LLMs under various settings. Our results show that (i) LLMs can replicate most conclusions from the original study, (ii) different LLMs yield varying levels of alignment in the results, and (iii) experimental factors such as LLM memory and output variability affect alignment with human responses. These initial findings suggest that LLMs could provide a scalable and cost-effective way to simplify qualitative XAI evaluation.

Gabriele Dominici, Pietro Barbiero, Francesco Giannini et al. · ibm-research

Current deep learning models are not designed to simultaneously address three fundamental questions: predict class labels to solve a given classification task (the "What?"), simulate changes in the situation to evaluate how this impacts class predictions (the "How?"), and imagine how the scenario should change to result in different class predictions (the "Why not?"). The inability to answer these questions represents a crucial gap in deploying reliable AI agents, calibrating human trust, and improving human-machine interaction. To bridge this gap, we introduce CounterFactual Concept Bottleneck Models (CF-CBMs), a class of models designed to efficiently address the above queries all at once without the need to run post-hoc searches. Our experimental results demonstrate that CF-CBMs: achieve classification accuracy comparable to black-box models and existing CBMs ("What?"), rely on fewer important concepts leading to simpler explanations ("How?"), and produce interpretable, concept-based counterfactuals ("Why not?"). Additionally, we show that training the counterfactual generator jointly with the CBM leads to two key improvements: (i) it alters the model's decision-making process, making the model rely on fewer important concepts (leading to simpler explanations), and (ii) it significantly increases the causal effect of concept interventions on class predictions, making the model more responsive to these changes.

Francesco Sovrano, Gabriele Dominici, Rita Sevastjanova et al.

Human cognitive biases in software engineering can lead to costly errors. While general-purpose AI (GPAI) systems may help mitigate these biases due to their non-human nature, their training on human-generated data raises a critical question: Do GPAI systems themselves exhibit cognitive biases? To investigate this, we present the first dynamic benchmarking framework to evaluate data-induced cognitive biases in GPAI within software engineering workflows. Starting with a seed set of 16 hand-crafted realistic tasks, each featuring one of 8 cognitive biases (e.g., anchoring, framing) and corresponding unbiased variants, we test whether bias-inducing linguistic cues unrelated to task logic can lead GPAI systems from correct to incorrect conclusions. To scale the benchmark and ensure realism, we develop an on-demand augmentation pipeline relying on GPAI systems to generate task variants that preserve bias-inducing cues while varying surface details. This pipeline ensures correctness (88--99% on average, according to human evaluation), promotes diversity, and controls reasoning complexity by leveraging Prolog-based reasoning and LLM-as-a-judge validation. It also verifies that the embedded biases are both harmful and undetectable by logic-based, unbiased reasoners. We evaluate leading GPAI systems (GPT, LLaMA, DeepSeek) and find a consistent tendency to rely on shallow linguistic heuristics over deep reasoning. All systems exhibit cognitive biases (ranging from 5.9% to 35% across types), with bias sensitivity increasing sharply with task complexity (up to 49%), highlighting critical risks in real-world software engineering deployments.

David Debot, Pietro Barbiero, Gabriele Dominici et al.

Concept-Based Models (CBMs) are a class of deep learning models that provide interpretability by explaining predictions through high-level concepts. These models first predict concepts and then use them to perform a downstream task. However, current CBMs offer interpretability only for the final task prediction, while the concept predictions themselves are typically made via black-box neural networks. To address this limitation, we propose Hierarchical Concept Memory Reasoner (H-CMR), a new CBM that provides interpretability for both concept and task predictions. H-CMR models relationships between concepts using a learned directed acyclic graph, where edges represent logic rules that define concepts in terms of other concepts. During inference, H-CMR employs a neural attention mechanism to select a subset of these rules, which are then applied hierarchically to predict all concepts and the final task. Experimental results demonstrate that H-CMR matches state-of-the-art performance while enabling strong human interaction through concept and model interventions. The former can significantly improve accuracy at inference time, while the latter can enhance data efficiency during training when background knowledge is available.

Mateo Espinosa Zarlenga, Gabriele Dominici, Pietro Barbiero et al.

In this paper, we investigate how concept-based models (CMs) respond to out-of-distribution (OOD) inputs. CMs are interpretable neural architectures that first predict a set of high-level concepts (e.g., stripes, black) and then predict a task label from those concepts. In particular, we study the impact of concept interventions (i.e., operations where a human expert corrects a CM's mispredicted concepts at test time) on CMs' task predictions when inputs are OOD. Our analysis reveals a weakness in current state-of-the-art CMs, which we term leakage poisoning, that prevents them from properly improving their accuracy when intervened on for OOD inputs. To address this, we introduce MixCEM, a new CM that learns to dynamically exploit leaked information missing from its concepts only when this information is in-distribution. Our results across tasks with and without complete sets of concept annotations demonstrate that MixCEMs outperform strong baselines by significantly improving their accuracy for both in-distribution and OOD samples in the presence and absence of concept interventions.

Andrea Pugnana, Riccardo Massidda, Francesco Giannini et al.

Concept Bottleneck Models (CBMs) are machine learning models that improve interpretability by grounding their predictions on human-understandable concepts, allowing for targeted interventions in their decision-making process. However, when intervened on, CBMs assume the availability of humans that can identify the need to intervene and always provide correct interventions. Both assumptions are unrealistic and impractical, considering labor costs and human error-proneness. In contrast, Learning to Defer (L2D) extends supervised learning by allowing machine learning models to identify cases where a human is more likely to be correct than the model, thus leading to deferring systems with improved performance. In this work, we gain inspiration from L2D and propose Deferring CBMs (DCBMs), a novel framework that allows CBMs to learn when an intervention is needed. To this end, we model DCBMs as a composition of deferring systems and derive a consistent L2D loss to train them. Moreover, by relying on a CBM architecture, DCBMs can explain why defer occurs on the final task. Our results show that DCBMs achieve high predictive performance and interpretability at the cost of deferring more to humans.

Teague McMillan, Gabriele Dominici, Martin Gjoreski et al.

Large Language Models (LLMs) often produce explanations that do not faithfully reflect the factors driving their predictions. In healthcare settings, such unfaithfulness is especially problematic: explanations that omit salient clinical cues or mask spurious shortcuts can undermine clinician trust and lead to unsafe decision support. We study how inference and training-time choices shape explanation faithfulness, focusing on factors practitioners can control at deployment. We evaluate three LLMs (GPT-4.1-mini, LLaMA 70B, LLaMA 8B) on two datasets-BBQ (social bias) and MedQA (medical licensing questions), and manipulate the number and type of few-shot examples, prompting strategies, and training procedure. Our results show: (i) both the quantity and quality of few-shot examples significantly impact model faithfulness; (ii) faithfulness is sensitive to prompting design; (iii) the instruction-tuning phase improves measured faithfulness on MedQA. These findings offer insights into strategies for enhancing the interpretability and trustworthiness of LLMs in sensitive domains.

Dario Fenoglio, Gabriele Dominici, Pietro Barbiero et al. · ibm-research

Federated Learning (FL), a privacy-aware approach in distributed deep learning environments, enables many clients to collaboratively train a model without sharing sensitive data, thereby reducing privacy risks. However, enabling human trust and control over FL systems requires understanding the evolving behaviour of clients, whether beneficial or detrimental for the training, which still represents a key challenge in the current literature. To address this challenge, we introduce Federated Behavioural Planes (FBPs), a novel method to analyse, visualise, and explain the dynamics of FL systems, showing how clients behave under two different lenses: predictive performance (error behavioural space) and decision-making processes (counterfactual behavioural space). Our experiments demonstrate that FBPs provide informative trajectories describing the evolving states of clients and their contributions to the global model, thereby enabling the identification of clusters of clients with similar behaviours. Leveraging the patterns identified by FBPs, we propose a robust aggregation technique named Federated Behavioural Shields to detect malicious or noisy client models, thereby enhancing security and surpassing the efficacy of existing state-of-the-art FL defense mechanisms. Our code is publicly available on GitHub.