Andrea Giovannini

Niharika S. D'Souza, Hongzhi Wang, Andrea Giovannini et al.

In a complex disease such as tuberculosis, the evidence for the disease and its evolution may be present in multiple modalities such as clinical, genomic, or imaging data. Effective patient-tailored outcome prediction and therapeutic guidance will require fusing evidence from these modalities. Such multimodal fusion is difficult since the evidence for the disease may not be uniform across all modalities, not all modality features may be relevant, or not all modalities may be present for all patients. All these nuances make simple methods of early, late, or intermediate fusion of features inadequate for outcome prediction. In this paper, we present a novel fusion framework using multiplexed graphs and derive a new graph neural network for learning from such graphs. Specifically, the framework allows modalities to be represented through their targeted encodings, and models their relationship explicitly via multiplexed graphs derived from salient features in a combined latent space. We present results that show that our proposed method outperforms state-of-the-art methods of fusing modalities for multi-outcome prediction on a large Tuberculosis (TB) dataset.

Niharika S. D'Souza, Hongzhi Wang, Andrea Giovannini et al.

With the emergence of multimodal electronic health records, the evidence for an outcome may be captured across multiple modalities ranging from clinical to imaging and genomic data. Predicting outcomes effectively requires fusion frameworks capable of modeling fine-grained and multi-faceted complex interactions between modality features within and across patients. We develop an innovative fusion approach called MaxCorr MGNN that models non-linear modality correlations within and across patients through Hirschfeld-Gebelein-Renyi maximal correlation (MaxCorr) embeddings, resulting in a multi-layered graph that preserves the identities of the modalities and patients. We then design, for the first time, a generalized multi-layered graph neural network (MGNN) for task-informed reasoning in multi-layered graphs, that learns the parameters defining patient-modality graph connectivity and message passing in an end-to-end fashion. We evaluate our model an outcome prediction task on a Tuberculosis (TB) dataset consistently outperforming several state-of-the-art neural, graph-based and traditional fusion techniques.

Corentin Royer, Debarun Bhattacharjya, Gaetano Rossiello et al.

Multi-step reasoning improves the capabilities of large language models (LLMs) but increases the risk of errors propagating through intermediate steps. Process reward models (PRMs) mitigate this by scoring each step individually, enabling fine-grained supervision and improved reliability. Existing methods for training PRMs rely on costly human annotations or computationally intensive automatic labeling. We propose a novel approach to automatically generate step-level labels using Information Theory. Our method estimates how each reasoning step affects the likelihood of the correct answer, providing a signal of step quality. Importantly, it reduces computational complexity to $\mathcal{O}(N)$, improving over the previous $\mathcal{O}(N \log N)$ methods. We demonstrate that these labels enable effective chain-of-thought selection in best-of-$K$ evaluation settings across diverse reasoning benchmarks, including mathematics, Python programming, SQL, and scientific question answering. This work enables scalable and efficient supervision of LLM reasoning, particularly for tasks where error propagation is critical.

Christopher Zanoli, Andrea Giovannini, Tengjun Jin et al.

Constructing Extract-Load-Transform (ELT) pipelines is a labor-intensive data engineering task and a high-impact target for AI automation. On ELT-Bench, the first benchmark for end-to-end ELT pipeline construction, AI agents initially showed low success rates, suggesting they lacked practical utility. We revisit these results and identify two factors causing a substantial underestimation of agent capabilities. First, re-evaluating ELT-Bench with upgraded large language models reveals that the extraction and loading stage is largely solved, while transformation performance improves significantly. Second, we develop an Auditor-Corrector methodology that combines scalable LLM-driven root-cause analysis with rigorous human validation (inter-annotator agreement Fleiss' kappa = 0.85) to audit benchmark quality. Applying this to ELT-Bench uncovers that most failed transformation tasks contain benchmark-attributable errors -- including rigid evaluation scripts, ambiguous specifications, and incorrect ground truth -- that penalize correct agent outputs. Based on these findings, we construct ELT-Bench-Verified, a revised benchmark with refined evaluation logic and corrected ground truth. Re-evaluating on this version yields significant improvement attributable entirely to benchmark correction. Our results show that both rapid model improvement and benchmark quality issues contributed to underestimating agent capabilities. More broadly, our findings echo observations of pervasive annotation errors in text-to-SQL benchmarks, suggesting quality issues are systemic in data engineering evaluation. Systematic quality auditing should be standard practice for complex agentic tasks. We release ELT-Bench-Verified to provide a more reliable foundation for progress in AI-driven data engineering automation.

Ben Schaper, Christopher Lohse, Marcell Streile et al.

Despite extensive recent advances in summary generation models, evaluation of auto-generated summaries still widely relies on single-score systems insufficient for transparent assessment and in-depth qualitative analysis. Towards bridging this gap, we propose the multifaceted interpretable summary evaluation method (MISEM), which is based on allocation of a summary's contextual token embeddings to semantic topics identified in the reference text. We further contribute an interpretability toolbox for automated summary evaluation and interactive visual analysis of summary scoring, topic identification, and token-topic allocation. MISEM achieves a promising .404 Pearson correlation with human judgment on the TAC'08 dataset.

Yotam Perlitz, Elad Venezian, Corentin Royer et al.

SQL dialects vary in syntax, types, and functions across database engines. Text-to-SQL benchmarks, however, predominantly support only SQLite. This creates a critical evaluation gap: cross-dialect evaluation reveals weak per-query agreement (Cohen's ), showing that SQLite performance is an unreliable proxy for other dialects. Yet such evaluation remains prohibitively difficult: existing approaches either require expensive manual query transpilation or rely on tools that often fail on complex SQL. To close this gap, we introduce PolySQL, a novel dual-execution method that eliminates the need for query transpilation by comparing normalized execution results. Notably, our approach achieves higher evaluation fidelity than query transpilation with 100% query coverage. PolySQL comprises three datasets, enabling the first large-scale cross-dialect study. Our study reveals a 10.1% average accuracy drop from SQLite to other dialects and identifies a significant dialect difficulty hierarchy. We find this degradation stems from logical rather than syntactic errors (61% vs. 8%). We release our framework code and leaderboard to enable rigorous dialect-robust evaluation.

Michael Nidd, Christoph Miksovic, Thomas Gschwind et al.

Having access to realistic workloads for a given database instance is extremely important to enable stress and vulnerability testing, as well as to optimize for cost and performance. Recent advances in learned cost models have shown that when enough diverse SQL queries are available, one can effectively and efficiently predict the cost of running a given query against a specific database engine. In this paper, we describe our experience in exploiting modern synthetic data generation techniques, inspired by the generative AI and LLM community, to create high-quality datasets enabling the effective training of such learned cost models. Initial results show that we can improve a learned cost model's predictive accuracy by training it with 45% fewer queries than when using competitive generation approaches.

Ivan Girardi, Panagiotis Vagenas, Dario Arcos-Díaz et al.

We develop various AI models to predict hospitalization on a large (over 110$k$) cohort of COVID-19 positive-tested US patients, sourced from March 2020 to February 2021. Models range from Random Forest to Neural Network (NN) and Time Convolutional NN, where combination of the data modalities (tabular and time dependent) are performed at different stages (early vs. model fusion). Despite high data unbalance, the models reach average precision 0.96-0.98 (0.75-0.85), recall 0.96-0.98 (0.74-0.85), and $F_1$-score 0.97-0.98 (0.79-0.83) on the non-hospitalized (or hospitalized) class. Performances do not significantly drop even when selected lists of features are removed to study model adaptability to different scenarios. However, a systematic study of the SHAP feature importance values for the developed models in the different scenarios shows a large variability across models and use cases. This calls for even more complete studies on several explainability methods before their adoption in high-stakes scenarios.

Joy T. Wu, Nkechinyere N. Agu, Ismini Lourentzou et al.

Despite the progress in automatic detection of radiologic findings from chest X-ray (CXR) images in recent years, a quantitative evaluation of the explainability of these models is hampered by the lack of locally labeled datasets for different findings. With the exception of a few expert-labeled small-scale datasets for specific findings, such as pneumonia and pneumothorax, most of the CXR deep learning models to date are trained on global "weak" labels extracted from text reports, or trained via a joint image and unstructured text learning strategy. Inspired by the Visual Genome effort in the computer vision community, we constructed the first Chest ImaGenome dataset with a scene graph data structure to describe $242,072$ images. Local annotations are automatically produced using a joint rule-based natural language processing (NLP) and atlas-based bounding box detection pipeline. Through a radiologist constructed CXR ontology, the annotations for each CXR are connected as an anatomy-centered scene graph, useful for image-level reasoning and multimodal fusion applications. Overall, we provide: i) $1,256$ combinations of relation annotations between $29$ CXR anatomical locations (objects with bounding box coordinates) and their attributes, structured as a scene graph per image, ii) over $670,000$ localized comparison relations (for improved, worsened, or no change) between the anatomical locations across sequential exams, as well as ii) a manually annotated gold standard scene graph dataset from $500$ unique patients.

Kevin Thandiackal, Tiziano Portenier, Andrea Giovannini et al.

Neural networks are prone to catastrophic forgetting when trained incrementally on different tasks. Popular incremental learning methods mitigate such forgetting by retaining a subset of previously seen samples and replaying them during the training on subsequent tasks. However, this is not always possible, e.g., due to data protection regulations. In such restricted scenarios, one can employ generative models to replay either artificial images or hidden features to a classifier. In this work, we propose Genifer (GENeratIve FEature-driven image Replay), where a generative model is trained to replay images that must induce the same hidden features as real samples when they are passed through the classifier. Our technique therefore incorporates the benefits of both image and feature replay, i.e.: (1) unlike conventional image replay, our generative model explicitly learns the distribution of features that are relevant for classification; (2) in contrast to feature replay, our entire classifier remains trainable; and (3) we can leverage image-space augmentations, which increase distillation performance while also mitigating overfitting during the training of the generative model. We show that Genifer substantially outperforms the previous state of the art for various settings on the CIFAR-100 and CUB-200 datasets.

Chiara Marchiori, Douglas Dykeman, Ivan Girardi et al.

Applying state-of-the-art machine learning and natural language processing on approximately one million of teleconsultation records, we developed a triage system, now certified and in use at the largest European telemedicine provider. The system evaluates care alternatives through interactions with patients via a mobile application. Reasoning on an initial set of provided symptoms, the triage application generates AI-powered, personalized questions to better characterize the problem and recommends the most appropriate point of care and time frame for a consultation. The underlying technology was developed to meet the needs for performance, transparency, user acceptance and ease of use, central aspects to the adoption of AI-based decision support systems. Providing such remote guidance at the beginning of the chain of care has significant potential for improving cost efficiency, patient experience and outcomes. Being remote, always available and highly scalable, this service is fundamental in high demand situations, such as the current COVID-19 outbreak.