Sparkle: Mastering Basic Spatial Capabilities in Vision Language Models Elicits Generalization to Composite Spatial Reasoning

Vision language models (VLMs) have demonstrated impressive performance across a wide range of downstream tasks. However, their proficiency in spatial reasoning remains limited, despite its crucial role in tasks involving navigation and interaction with physical environments. Specifically, much of the spatial reasoning in these tasks occurs in two-dimensional (2D) environments, and our evaluation reveals that state-of-the-art VLMs frequently generate implausible and incorrect responses to composite spatial reasoning problems, including simple pathfinding tasks that humans can solve effortlessly at a glance. To address this, we explore an effective approach to enhance 2D spatial reasoning within VLMs by training the model on basic spatial capabilities. We begin by disentangling the key components of 2D spatial reasoning: direction comprehension, distance estimation, and localization. Our central hypothesis is that mastering these basic spatial capabilities can significantly enhance a model's performance on composite spatial tasks requiring advanced spatial understanding and combinatorial problem-solving. To investigate this hypothesis, we introduce Sparkle, a framework that fine-tunes VLMs on these three basic spatial capabilities by synthetic data generation and targeted supervision to form an instruction dataset for each capability. Our experiments demonstrate that VLMs fine-tuned with Sparkle achieve significant performance gains, not only in the basic tasks themselves but also in generalizing to composite and out-of-distribution spatial reasoning tasks (e.g., improving from 13.5% to 40.0% on the shortest path problem). These findings underscore the effectiveness of mastering basic spatial capabilities in enhancing composite spatial problem-solving, offering insights for improving VLMs' spatial reasoning capabilities.

High and Low Resolution Tradeoffs in Roadside Multimodal Sensing

Designing roadside sensing for intelligent transportation applications requires balancing cost and performance,especially when choosing between high and low-resolution sensors. The tradeoff is challenging due to sensor heterogeneity,where different sensors produce unique data modalities due to varying physical principles. High-resolution LiDAR offers detailed point cloud, while 4D millimeter-wave radar, despite providing sparser data, delivers velocity information useful for distinguishing objects based on movement patterns. To assess whether reductions in spatial resolution can be compensated by the informational richness of sensors, particularly in recognizing both vehicles and vulnerable road users (VRUs), we propose Residual Fusion Net (ResFusionNet) to fuse multimodal data for 3D object detection. This enables a quantifiable tradeoff between spatial resolution and information richness across different modalities. Furthermore, we introduce a sensor placement algorithm utilizing probabilistic modeling to manage uncertainties in sensor visibility influenced by environmental or human-related factors. Through simulation-assisted ex-ante evaluation on a real-world testbed, our findings show marked marginal gains in detecting VRUs--an average of 16.7% for pedestrians and 11% for cyclists--when merging velocity-encoded radar with LiDAR, compared to LiDAR only configurations. Additionally, experimental results from 300 runs reveal a maximum loss of 11.5% and a average of 5.25% in sensor coverage due to uncertainty factors. These findings underscore the potential of using low spatial resolution but information-rich sensors to enhance detection capabilities for vulnerable road users while highlighting the necessity of thoroughly evaluating sensor modality heterogeneity, traffic participant diversity, and operational uncertainties when making sensor tradeoffs in practical applications.

RoleBreak: Character Hallucination as a Jailbreak Attack in Role-Playing Systems

Role-playing systems powered by large language models (LLMs) have become increasingly influential in emotional communication applications. However, these systems are susceptible to character hallucinations, where the model deviates from predefined character roles and generates responses that are inconsistent with the intended persona. This paper presents the first systematic analysis of character hallucination from an attack perspective, introducing the RoleBreak framework. Our framework identifies two core mechanisms-query sparsity and role-query conflict-as key factors driving character hallucination. Leveraging these insights, we construct a novel dataset, RoleBreakEval, to evaluate existing hallucination mitigation techniques. Our experiments reveal that even enhanced models trained to minimize hallucination remain vulnerable to attacks. To address these vulnerabilities, we propose a novel defence strategy, the Narrator Mode, which generates supplemental context through narration to mitigate role-query conflicts and improve query generalization. Experimental results demonstrate that Narrator Mode significantly outperforms traditional refusal-based strategies by reducing hallucinations, enhancing fidelity to character roles and queries, and improving overall narrative coherence.

MORPHEUS: Modeling Role from Personalized Dialogue History by Exploring and Utilizing Latent Space

Personalized Dialogue Generation (PDG) aims to create coherent responses according to roles or personas. Traditional PDG relies on external role data, which can be scarce and raise privacy concerns. Approaches address these issues by extracting role information from dialogue history, which often fail to generically model roles in continuous space. To overcome these limitations, we introduce a novel framework \textbf{MO}dels \textbf{R}oles from \textbf{P}ersonalized Dialogue \textbf{H}istory by \textbf{E}xploring and \textbf{U}tilizing Latent \textbf{S}pace (MORPHEUS) through a three-stage training process. Specifically, we create a persona codebook to represent roles in latent space compactly, and this codebook is used to construct a posterior distribution of role information. This method enables the model to generalize across roles, allowing the generation of personalized dialogues even for unseen roles. Experiments on both Chinese and English datasets demonstrate that MORPHEUS enhances the extraction of role information, and improves response generation without external role data. Additionally, MORPHEUS can be considered an efficient fine-tuning for large language models.

Enhancing Role-playing Systems through Aggressive Queries: Evaluation and Improvement

The advent of Large Language Models (LLMs) has propelled dialogue generation into new realms, particularly in the field of role-playing systems (RPSs). While enhanced with ordinary role-relevant training dialogues, existing LLM-based RPSs still struggle to align with roles when handling intricate and trapped queries in boundary scenarios. In this paper, we design the Modular ORchestrated Trap-setting Interaction SystEm (MORTISE) to benchmark and improve the role-playing LLMs' performance. MORTISE can produce highly role-relevant aggressive queries through the collaborative effort of multiple LLM-based modules, and formulate corresponding responses to create an adversarial training dataset via a consistent response generator. We select 190 Chinese and English roles to construct aggressive queries to benchmark existing role-playing LLMs. Through comprehensive evaluation, we find that existing models exhibit a general deficiency in role alignment capabilities. We further select 180 of the roles to collect an adversarial training dataset (named RoleAD) and retain the other 10 roles for testing. Experiments on models improved by RoleAD indicate that our adversarial dataset ameliorates this deficiency, with the improvements demonstrating a degree of generalizability in ordinary scenarios.

Synergizing Spatial Optimization with Large Language Models for Open-Domain Urban Itinerary Planning

In this paper, we for the first time propose the task of Open-domain Urban Itinerary Planning (OUIP) for citywalk, which directly generates itineraries based on users' requests described in natural language. OUIP is different from conventional itinerary planning, which limits users from expressing more detailed needs and hinders true personalization. Recently, large language models (LLMs) have shown potential in handling diverse tasks. However, due to non-real-time information, incomplete knowledge, and insufficient spatial awareness, they are unable to independently deliver a satisfactory user experience in OUIP. Given this, we present ItiNera, an OUIP system that synergizes spatial optimization with Large Language Models (LLMs) to provide services that customize urban itineraries based on users' needs. Specifically, we develop an LLM-based pipeline for extracting and updating POI features to create a user-owned personalized POI database. For each user request, we leverage LLM in cooperation with an embedding-based module for retrieving candidate POIs from the user's POI database. Then, a spatial optimization module is used to order these POIs, followed by LLM crafting a personalized, spatially coherent itinerary. To the best of our knowledge, this study marks the first integration of LLMs to innovate itinerary planning solutions. Extensive experiments on offline datasets and online subjective evaluation have demonstrated the capacities of our system to deliver more responsive and spatially coherent itineraries than current LLM-based solutions. Our system has been deployed in production at the TuTu online travel service and has attracted thousands of users for their urban travel planning.

DialogBench: Evaluating LLMs as Human-like Dialogue Systems

Large language models (LLMs) have achieved remarkable breakthroughs in new dialogue capabilities, refreshing human's impressions on dialogue systems. The long-standing goal of dialogue systems is to be human-like enough to establish long-term connections with users by satisfying the need for communication, affection and social belonging. Therefore, there has been an urgent need to evaluate LLMs as human-like dialogue systems. In this paper, we propose DialogBench, a dialogue evaluation benchmark that currently contains $12$ dialogue tasks to assess the capabilities of LLMs as human-like dialogue systems should have. Specifically, we prompt GPT-4 to generate evaluation instances for each task. We first design the basic prompt based on widely-used design principles and further mitigate the existing biases to generate higher-quality evaluation instances. Our extensive test over $28$ LLMs (including pre-trained and supervised instruction-tuning) shows that instruction fine-tuning benefits improve the human likeness of LLMs to a certain extent, but there is still much room to improve those capabilities for most LLMs as human-like dialogue systems. In addition, experimental results also indicate that LLMs perform differently in various abilities that human-like dialogue systems should have. We will publicly release DialogBench, along with the associated evaluation code for the broader research community.

Enhancing Personalized Dialogue Generation with Contrastive Latent Variables: Combining Sparse and Dense Persona

The personalized dialogue explores the consistent relationship between dialogue generation and personality. Existing personalized dialogue agents model persona profiles from three resources: sparse or dense persona descriptions and dialogue histories. However, sparse structured persona attributes are explicit but uninformative, dense persona texts contain rich persona descriptions with much noise, and dialogue history query is both noisy and uninformative for persona modeling. In this work, we combine the advantages of the three resources to obtain a richer and more accurate persona. We design a Contrastive Latent Variable-based model (CLV) that clusters the dense persona descriptions into sparse categories, which are combined with the history query to generate personalized responses. Experimental results on Chinese and English datasets demonstrate our model's superiority in personalization.

HGARN: Hierarchical Graph Attention Recurrent Network for Human Mobility Prediction

Human mobility prediction is a fundamental task essential for various applications, including urban planning, transportation services, and location recommendation. Existing approaches often ignore activity information crucial for reasoning human preferences and routines, or adopt a simplified representation of the dependencies between time, activities and locations. To address these issues, we present Hierarchical Graph Attention Recurrent Network (HGARN) for human mobility prediction. Specifically, we construct a hierarchical graph based on all users' history mobility records and employ a Hierarchical Graph Attention Module to capture complex time-activity-location dependencies. This way, HGARN can learn representations with rich contextual semantics to model user preferences at the global level. We also propose a model-agnostic history-enhanced confidence (MaHec) label to focus our model on each user's individual-level preferences. Finally, we introduce a Recurrent Encoder-Decoder Module, which employs recurrent structures to jointly predict users' next activities (as an auxiliary task) and locations. For model evaluation, we test the performances of our Hgarn against existing SOTAs in recurring and explorative settings. The recurring setting focuses more on assessing models' capabilities to capture users' individual-level preferences. In contrast, the results in the explorative setting tend to reflect the power of different models to learn users' global-level preferences. Overall, our model outperforms other baselines significantly in the main, recurring, and explorative settings based on two real-world human mobility data benchmarks. Source codes of HGARN are available at https://github.com/YihongT/HGARN.

Few-Shot Traffic Prediction with Graph Networks using Locale as Relational Inductive Biases

Accurate short-term traffic prediction plays a pivotal role in various smart mobility operation and management systems. Currently, most of the state-of-the-art prediction models are based on graph neural networks (GNNs), and the required training samples are proportional to the size of the traffic network. In many cities, the available amount of traffic data is substantially below the minimum requirement due to the data collection expense. It is still an open question to develop traffic prediction models with a small size of training data on large-scale networks. We notice that the traffic states of a node for the near future only depend on the traffic states of its localized neighborhoods, which can be represented using the graph relational inductive biases. In view of this, this paper develops a graph network (GN)-based deep learning model LocaleGn that depicts the traffic dynamics using localized data aggregating and updating functions, as well as the node-wise recurrent neural networks. LocaleGn is a light-weighted model designed for training on few samples without over-fitting, and hence it can solve the problem of few-shot traffic prediction. The proposed model is examined on predicting both traffic speed and flow with six datasets, and the experimental results demonstrate that LocaleGn outperforms existing state-of-the-art baseline models. It is also demonstrated that the learned knowledge from LocaleGn can be transferred across cities. The research outcomes can help to develop light-weighted traffic prediction systems, especially for cities lacking in historically archived traffic data.