Paolo Santi

Andrea La Grotteria, Paolo Santi, Titus Venverloo et al.

Police-reported crash statistics remain the standard input for urban road-safety assessment, but their incompleteness and reporting lag limit their usefulness for timely, fine-grained intervention design. Harsh acceleration and braking events are widely used as surrogate safety indicators, but have so far been studied only in comparatively small urban samples. This study analyses harsh events across the urban road network of Milan, combining high-resolution telematics from more than 4.2 million vehicles equipped with On-Board Units, segment-level traffic metrics from TomTom, street-network and infrastructure attributes from OpenStreetMap, and visual streetscape features extracted from Google Street View via semantic segmentation using a OneFormer model. We employ an analytical framework combining non-parametric Mann--Whitney U tests of segment-feature distributions between high- and low-harshness groups with supervised machine-learning regressors. We find that, once exposure is controlled for, wider carriageways, crossings and transit stops, and more open visual fields (higher sky- and road-pixel proportions) are associated with higher harsh-event intensity, while denser built frontage is associated with lower intensity. Finally, the cycling-infrastructure case study identifies a gradient in harsh-event intensity across facility types: markings-only cycle lanes are associated with a 19.5% higher harshness score, and mixed-traffic configurations with an 11.5% higher score, relative to physically separated cycle paths, conditional on the included controls. These results support context-specific rather than uniform urban-safety interventions and illustrate how large-scale telematics combined with open geospatial and visual data can inform Vision Zero decision-making at the metropolitan scale.

Mohammad M. Vazifeh, Hongmou Zhang, Paolo Santi et al.

With recent advances in battery technology and the resulting decrease in the charging times, public charging stations are becoming a viable option for Electric Vehicle (EV) drivers. Concurrently, wide-spread use of location-tracking devices in mobile phones and wearable devices makes it possible to track individual-level human movements to an unprecedented spatial and temporal grain. Motivated by these developments, we propose a novel methodology to perform data-driven optimization of EV charging stations location. We formulate the problem as a discrete optimization problem on a geographical grid, with the objective of covering the entire demand region while minimizing a measure of drivers' discomfort. Since optimally solving the problem is computationally infeasible, we present computationally efficient, near-optimal solutions based on greedy and genetic algorithms. We then apply the proposed methodology to optimize EV charging stations location in the city of Boston, starting from a massive cellular phone data sets covering 1 million users over 4 months. Results show that genetic algorithm based optimization provides the best solutions in terms of drivers' discomfort and the number of charging stations required, which are both reduced about 10 percent as compared to a randomized solution. We further investigate robustness of the proposed data-driven methodology, showing that, building upon well-known regularity of aggregate human mobility patterns, the near-optimal solution computed using single day movements preserves its properties also in later months. When collectively considered, the results presented in this paper clearly indicate the potential of data-driven approaches for optimally locating public charging facilities at the urban scale.

Yu Zheng, Yuan Yuan, Yue Zhuo et al.

Probing large language models (LLMs) has yielded valuable insights into their internal mechanisms by linking neural activations to interpretable semantics. However, the complex mechanisms that link neuron's functional co-activation with the emergent model capabilities remains largely unknown, hindering a deeper understanding and safer development of LLMs. In this work, we introduce graph probing, a method for uncovering the functional connectivity of LLM neurons and relating it to language generation performance. By probing models across diverse LLM families and scales, we discover a universal predictability of next-token prediction performance using only neural topology, which persists even when retaining just 1% of neuron connections. Strikingly, probing on topology outperforms probing on activation by up to 130.4%, suggesting that neural topology contains orders of richer information of LLM performance than neural activation, which can be easily extracted with simple linear or MLP probes. To explain the dependence between neural topology and language performance, we identify default networks and hub neurons in LLMs and provide causal evidence by interventional experiments on multiple benchmarks, showing that LLMs actually exploit these topological information. Further analyses suggest that neural topology can be effectively leveraged to improve the efficiency, reliability, and safety of LLMs through proof-of-concept applications in model pruning, hallucination detection, and LLM fingerprinting. Codes and data for the graph probing toolbox are available at https://github.com/DavyMorgan/llm-graph-probing.

Miho Koda, Yu Zheng, Ruixian Ma et al.

Recent advances in large language models (LLMs), particularly those enhanced through reinforced post-training, have demonstrated impressive reasoning capabilities, as exemplified by models such as OpenAI o1 and DeepSeek-R1. However, these capabilities are predominantly benchmarked on domains like mathematical problem solving and code generation, leaving open the question of whether such reasoning skills generalize to complex real-world scenarios. In this paper, we introduce LocationReasoner, a benchmark designed to evaluate LLMs' reasoning abilities in the context of real-world site selection, where models must identify feasible locations by reasoning over diverse and complicated spatial, environmental, and logistic constraints. The benchmark covers carefully crafted queries of varying difficulty levels and is supported by a sandbox environment with in-house tools for constraint-based location search. Automated verification further guarantees the scalability of the benchmark, enabling the addition of arbitrary number of queries. Extensive evaluations on real-world site selection data from Boston, New York, and Tampa reveal that state-of-the-art reasoning models offer limited improvement over their non-reasoning predecessors in real-world contexts, with even the latest OpenAI o4 model failing on 30% of site selection tasks. Moreover, agentic strategies such as ReAct and Reflexion often suffer from over-reasoning, leading to worse outcomes than direct prompting. With key limitations of LLMs in holistic and non-linear reasoning highlighted, we release LocationReasoner to foster the development of LLMs and agents capable of robust, grounded reasoning in real-world decision-making tasks. Codes and data for our benchmark are available at https://github.com/miho-koda/LocationReasoner.

Mohammad Ramezanali, Mo Vazifeh, Paolo Santi

We introduce seqBench, a parametrized benchmark for probing sequential reasoning limits in Large Language Models (LLMs) through precise, multi-dimensional control over several key complexity dimensions. seqBench allows systematic variation of (1) the logical depth, defined as the number of sequential actions required to solve the task; (2) the number of backtracking steps along the optimal path, quantifying how often the agent must revisit prior states to satisfy deferred preconditions (e.g., retrieving a key after encountering a locked door); and (3) the noise ratio, defined as the ratio between supporting and distracting facts about the environment. Our evaluations on state-of-the-art LLMs reveal a universal failure pattern: accuracy collapses exponentially beyond a model-specific logical depth. Unlike existing benchmarks, seqBench's fine-grained control facilitates targeted analyses of these reasoning failures, illuminating universal scaling laws and statistical limits, as detailed in this paper alongside its generation methodology and evaluation metrics. We find that even top-performing models systematically fail on seqBench's structured reasoning tasks despite minimal search complexity, underscoring key limitations in their commonsense reasoning capabilities. Designed for future evolution to keep pace with advancing models, the seqBench datasets are publicly released to spur deeper scientific inquiry into LLM reasoning, aiming to establish a clearer understanding of their true potential and current boundaries for robust real-world application.

Giacomo Orsi, Titus Venverloo, Andrea La Grotteria et al.

In recent years, cities have increasingly reduced speed limits from 50 km/h to 30 km/h to enhance road safety, reduce noise pollution, and promote sustainable modes of transportation. However, achieving compliance with these new limits remains a key challenge for urban planners. This study investigates drivers' compliance with the 30 km/h speed limit in Milan and examines how street characteristics influence driving behavior. Our findings suggest that the mere introduction of lower speed limits is not sufficient to reduce driving speeds effectively, highlighting the need to understand how street design can improve speed limit adherence. To comprehend this relationship, we apply computer vision-based semantic segmentation models to Google Street View images. A large-scale analysis reveals that narrower streets and densely built environments are associated with lower speeds, whereas roads with greater visibility and larger sky views encourage faster driving. To evaluate the influence of the local context on speeding behaviour, we apply the developed methodological framework to two additional cities: Amsterdam, which, similar to Milan, is a historic European city not originally developed for cars, and Dubai, which instead has developed in recent decades with a more car-centric design. The results of the analyses largely confirm the findings obtained in Milan, which demonstrates the broad applicability of the road design guidelines for driver speed compliance identified in this paper. Finally, we develop a machine learning model to predict driving speeds based on street characteristics. We showcase the model's predictive power by estimating the compliance with speed limits in Milan if the city were to adopt a 30 km/h speed limit city-wide. The tool provides actionable insights for urban planners, supporting the design of interventions to improve speed limit compliance.

Punit Rathore, James C. Bezdek, Paolo Santi et al.

The VAT method is a visual technique for determining the potential cluster structure and the possible number of clusters in numerical data. Its improved version, iVAT, uses a path-based distance transform to improve the effectiveness of VAT for "tough" cases. Both VAT and iVAT have also been used in conjunction with a single-linkage(SL) hierarchical clustering algorithm. However, they are sensitive to noise and bridge points between clusters in the dataset, and consequently, the corresponding VAT/iVAT images are often in-conclusive for such cases. In this paper, we propose a constraint-based version of iVAT, which we call ConiVAT, that makes use of background knowledge in the form of constraints, to improve VAT/iVAT for challenging and complex datasets. ConiVAT uses the input constraints to learn the underlying similarity metric and builds a minimum transitive dissimilarity matrix, before applying VAT to it. We demonstrate ConiVAT approach to visual assessment and single linkage clustering on nine datasets to show that, it improves the quality of iVAT images for complex datasets, and it also overcomes the limitation of SL clustering with VAT/iVAT due to "noisy" bridges between clusters. Extensive experiment results on nine datasets suggest that ConiVAT outperforms the other three semi-supervised clustering algorithms in terms of improved clustering accuracy.

Umberto Fugiglando, Emanuele Massaro, Paolo Santi et al.

Cars can nowadays record several thousands of signals through the CAN bus technology and potentially provide real-time information on the car, the driver and the surrounding environment. This paper proposes a new method for the analysis and classification of driver behavior using a selected subset of CAN bus signals, specifically gas pedal position, brake pedal pressure, steering wheel angle, steering wheel momentum, velocity, RPM, frontal and lateral acceleration. Data has been collected in a completely uncontrolled experiment, where 64 people drove 10 cars for or a total of over 2000 driving trips without any type of pre-determined driving instruction on a wide variety of road scenarios. We propose an unsupervised learning technique that clusters drivers in different groups, and offers a validation method to test the robustness of clustering in a wide range of experimental settings. The minimal amount of data needed to preserve robust driver clustering is also computed. The presented study provides a new methodology for near-real-time classification of driver behavior in uncontrolled environments.