Large language models (LLMs) are rapidly evolving into autonomous agents that cooperate across organizational boundaries, enabling joint disaster response, supply-chain optimization, and other tasks that demand decentralized expertise without surrendering data ownership. Yet, cross-domain collaboration shatters the unified trust assumptions behind current alignment and containment techniques. An agent benign in isolation may, when receiving messages from an untrusted peer, leak secrets or violate policy, producing risks driven by emergent multi-agent dynamics rather than classical software bugs. This position paper maps the security agenda for cross-domain multi-agent LLM systems. We introduce seven categories of novel security challenges, for each of which we also present plausible attacks, security evaluation metrics, and future research guidelines.
User-Centric Embodied Visual Tracking (UC-EVT) presents a novel challenge for reinforcement learning-based models due to the substantial gap between high-level user instructions and low-level agent actions. While recent advancements in language models (e.g., LLMs, VLMs, VLAs) have improved instruction comprehension, these models face critical limitations in either inference speed (LLMs, VLMs) or generalizability (VLAs) for UC-EVT tasks. To address these challenges, we propose \textbf{Hierarchical Instruction-aware Embodied Visual Tracking (HIEVT)} agent, which bridges instruction comprehension and action generation using \textit{spatial goals} as intermediaries. HIEVT first introduces \textit{LLM-based Semantic-Spatial Goal Aligner} to translate diverse human instructions into spatial goals that directly annotate the desired spatial position. Then the \textit{RL-based Adaptive Goal-Aligned Policy}, a general offline policy, enables the tracker to position the target as specified by the spatial goal. To benchmark UC-EVT tasks, we collect over ten million trajectories for training and evaluate across one seen environment and nine unseen challenging environments. Extensive experiments and real-world deployments demonstrate the robustness and generalizability of HIEVT across diverse environments, varying target dynamics, and complex instruction combinations. The complete project is available at https://sites.google.com/view/hievt.
In this position paper, we observe that empirical evaluation in Generative AI is at a crisis point since traditional ML evaluation and benchmarking strategies are insufficient to meet the needs of evaluating modern GenAI models and systems. There are many reasons for this, including the fact that these models typically have nearly unbounded input and output spaces, typically do not have a well defined ground truth target, and typically exhibit strong feedback loops and prediction dependence based on context of previous model outputs. On top of these critical issues, we argue that the problems of leakage and contamination are in fact the most important and difficult issues to address for GenAI evaluations. Interestingly, the field of AI Competitions has developed effective measures and practices to combat leakage for the purpose of counteracting cheating by bad actors within a competition setting. This makes AI Competitions an especially valuable (but underutilized) resource. Now is time for the field to view AI Competitions as the gold standard for empirical rigor in GenAI evaluation, and to harness and harvest their results with according value.
Post-training of foundation language models has emerged as a promising research domain in federated learning (FL) with the goal to enable privacy-preserving model improvements and adaptations to user's downstream tasks. Recent advances in this area adopt centralized post-training approaches that build upon black-box foundation language models where there is no access to model weights and architecture details. Although the use of black-box models has been successful in centralized post-training, their blind replication in FL raises several concerns. Our position is that using black-box models in FL contradicts the core principles of federation such as data privacy and autonomy. In this position paper, we critically analyze the usage of black-box models in federated post-training, and provide a detailed account of various aspects of openness and their implications for FL.
The digitization of cultural heritage collections has opened new directions for research, yet the lack of enriched metadata poses a substantial challenge to accessibility, interoperability, and cross-institutional collaboration. In several past years neural networks models such as YOLOv11 and Detectron2 have revolutionized visual data analysis, but their application to domain-specific cultural artifacts - such as manuscripts and incunabula - remains limited by the absence of methodologies that address structural feature extraction and semantic interoperability. In this position paper, we argue, that the integration of neural networks with semantic technologies represents a paradigm shift in cultural heritage digitization processes. We present the Metadata Enrichment Model (MEM), a conceptual framework designed to enrich metadata for digitized collections by combining fine-tuned computer vision models, large language models (LLMs) and structured knowledge graphs. The Multilayer Vision Mechanism (MVM) appears as the key innovation of MEM. This iterative process improves visual analysis by dynamically detecting nested features, such as text within seals or images within stamps. To expose MEM's potential, we apply it to a dataset of digitized incunabula from the Jagiellonian Digital Library and release a manually annotated dataset of 105 manuscript pages. We examine the practical challenges of MEM's usage in real-world GLAM institutions, including the need for domain-specific fine-tuning, the adjustment of enriched metadata with Linked Data standards and computational costs. We present MEM as a flexible and extensible methodology. This paper contributes to the discussion on how artificial intelligence and semantic web technologies can advance cultural heritage research, and also use these technologies in practice.
Modern vision-language models (VLMs) often fail at cultural competency evaluations and benchmarks. Given the diversity of applications built upon VLMs, there is renewed interest in understanding how they encode cultural nuances. While individual aspects of this problem have been studied, we still lack a comprehensive framework for systematically identifying and annotating the nuanced cultural dimensions present in images for VLMs. This position paper argues that foundational methodologies from visual culture studies (cultural studies, semiotics, and visual studies) are necessary for cultural analysis of images. Building upon this review, we propose a set of five frameworks, corresponding to cultural dimensions, that must be considered for a more complete analysis of the cultural competencies of VLMs.
An important task in machine learning (ML) research is comparing prior work, which is often performed via ML leaderboards: a tabular overview of experiments with comparable conditions (e.g., same task, dataset, and metric). However, the growing volume of literature creates challenges in creating and maintaining these leaderboards. To ease this burden, researchers have developed methods to extract leaderboard entries from research papers for automated leaderboard curation. Yet, prior work varies in problem framing, complicating comparisons and limiting real-world applicability. In this position paper, we present the first overview of Automatic Leaderboard Generation (ALG) research, identifying fundamental differences in assumptions, scope, and output formats. We propose an ALG unified conceptual framework to standardise how the ALG task is defined. We offer ALG benchmarking guidelines, including recommendations for datasets and metrics that promote fair, reproducible evaluation. Lastly, we outline challenges and new directions for ALG, such as, advocating for broader coverage by including all reported results and richer metadata.
Intra-frame prediction in the High Efficiency Video Coding (HEVC) standard can be empirically improved by applying sets of recursive two-dimensional filters to the predicted values. However, this approach does not allow (or complicates significantly) the parallel computation of pixel predictions. In this work we analyze why the recursive filters are effective, and use the results to derive sets of non-recursive predictors that have superior performance. We present an extension to HEVC intra prediction that combines values predicted using non-filtered and filtered (smoothed) reference samples, depending on the prediction mode, and block size. Simulations using the HEVC common test conditions show that a 2.0% bit rate average reduction can be achieved compared to HEVC, for All Intra (AI) configurations.
Recent efforts toward developing trustworthy AI systems with accountability guarantees have led to a growing reliance on machine learning formulations that incorporate external requirements, or constraints. These requirements are often enforced through penalization--adding fixed-weight terms to the task loss. We argue that this approach is ill-suited, and that tailored constrained optimization methods should be adopted instead. In particular, no penalty coefficient may yield a solution that both satisfies the constraints and achieves good performance--i.e., one solving the constrained problem. Moreover, tuning these coefficients is costly, incurring significant time and computational overhead. In contrast, tailored constrained methods--such as the Lagrangian approach, which optimizes the penalization "coefficients" (the Lagrange multipliers) alongside the model--(i) truly solve the constrained problem and add accountability, (ii) eliminate the need for extensive penalty tuning, and (iii) integrate seamlessly with modern deep learning pipelines.
Movable antennas (MAs) have emerged as a disruptive technology in wireless communications for enhancing spatial degrees of freedom through continuous antenna repositioning within predefined regions, thereby creating favorable channel propagation conditions. In this paper, we study the problem of position optimization for MA-enabled multi-user MISO systems, where a base station (BS), equipped with multiple MAs, communicates with multiple users each equipped with a single fixed-position antenna (FPA). To circumvent the difficulty of acquiring the channel state information (CSI) from the transmitter to the receiver over the entire movable region, we propose a derivative-free approach for MA position optimization. The basic idea is to treat position optimization as a closed-box optimization problem and calculate the gradient of the unknown objective function using zeroth-order (ZO) gradient approximation techniques. Specifically, the proposed method does not need to explicitly estimate the global CSI. Instead, it adaptively refines its next movement based on previous measurements such that it eventually converges to an optimum or stationary solution. Simulation results show that the proposed derivative-free approach is able to achieve higher sample and computational efficiencies than the CSI estimation-based position optimization approach, particularly for challenging scenarios where the number of multi-path components (MPCs) is large or the number of pilot signals is limited.
Graph neural networks (GNNs), which capture graph structures via a feature aggregation mechanism following the graph embedding framework, have demonstrated a powerful ability to support various tasks. According to the topology properties (e.g., structural roles or community memberships of nodes) to be preserved, graph embedding can be categorized into identity and position embedding. However, it is unclear for most GNN-based methods which property they can capture. Some of them may also suffer from low efficiency and scalability caused by several time- and space-consuming procedures (e.g., feature extraction and training). From a perspective of graph signal processing, we find that high- and low-frequency information in the graph spectral domain may characterize node identities and positions, respectively. Based on this investigation, we propose random feature aggregation (RFA) for efficient identity and position embedding, serving as an extreme ablation study regarding GNN feature aggregation. RFA (i) adopts a spectral-based GNN without learnable parameters as its backbone, (ii) only uses random noises as inputs, and (iii) derives embeddings via just one feed-forward propagation (FFP). Inspired by degree-corrected spectral clustering, we further introduce a degree correction mechanism to the GNN backbone. Surprisingly, our experiments demonstrate that two variants of RFA with high- and low-pass filters can respectively derive informative identity and position embeddings via just one FFP (i.e., without any training). As a result, RFA can achieve a better trade-off between quality and efficiency for both identity and position embedding over various baselines.
Sound source localization (SSL) is a critical technology for determining the position of sound sources in complex environments. However, existing methods face challenges such as high computational costs and precise calibration requirements, limiting their deployment in dynamic or resource-constrained environments. This paper introduces a novel 3D SSL framework, which uses sparse cross-attention, pretraining, and adaptive signal coherence metrics, to achieve accurate and computationally efficient localization with fewer input microphones. The framework is also fault-tolerant to unreliable or even unknown microphone position inputs, ensuring its applicability in real-world scenarios. Preliminary experiments demonstrate its scalability for multi-source localization without requiring additional hardware. This work advances SSL by balancing the model's performance and efficiency and improving its robustness for real-world scenarios.
Recent interest in Multi-Agent Systems of Large Language Models (MAS LLMs) has led to an increase in frameworks leveraging multiple LLMs to tackle complex tasks. However, much of this literature appropriates the terminology of MAS without engaging with its foundational principles. In this position paper, we highlight critical discrepancies between MAS theory and current MAS LLMs implementations, focusing on four key areas: the social aspect of agency, environment design, coordination and communication protocols, and measuring emergent behaviours. Our position is that many MAS LLMs lack multi-agent characteristics such as autonomy, social interaction, and structured environments, and often rely on oversimplified, LLM-centric architectures. The field may slow down and lose traction by revisiting problems the MAS literature has already addressed. Therefore, we systematically analyse this issue and outline associated research opportunities; we advocate for better integrating established MAS concepts and more precise terminology to avoid mischaracterisation and missed opportunities.
Vision-based tactile sensing has been widely used in perception, reconstruction, and robotic manipulation. However, collecting large-scale tactile data remains costly due to the localized nature of sensor-object interactions and inconsistencies across sensor instances. Existing approaches to scaling tactile data, such as simulation and free-form tactile generation, often suffer from unrealistic output and poor transferability to downstream tasks. To address this, we propose ControlTac, a two-stage controllable framework that generates realistic tactile images conditioned on a single reference tactile image, contact force, and contact position. With those physical priors as control input, ControlTac generates physically plausible and varied tactile images that can be used for effective data augmentation. Through experiments on three downstream tasks, we demonstrate that ControlTac can effectively augment tactile datasets and lead to consistent gains. Our three real-world experiments further validate the practical utility of our approach. Project page: https://dongyuluo.github.io/controltac.
System prompts in Large Language Models (LLMs) are predefined directives that guide model behaviour, taking precedence over user inputs in text processing and generation. LLM deployers increasingly use them to ensure consistent responses across contexts. While model providers set a foundation of system prompts, deployers and third-party developers can append additional prompts without visibility into others' additions, while this layered implementation remains entirely hidden from end-users. As system prompts become more complex, they can directly or indirectly introduce unaccounted for side effects. This lack of transparency raises fundamental questions about how the position of information in different directives shapes model outputs. As such, this work examines how the placement of information affects model behaviour. To this end, we compare how models process demographic information in system versus user prompts across six commercially available LLMs and 50 demographic groups. Our analysis reveals significant biases, manifesting in differences in user representation and decision-making scenarios. Since these variations stem from inaccessible and opaque system-level configurations, they risk representational, allocative and potential other biases and downstream harms beyond the user's ability to detect or correct. Our findings draw attention to these critical issues, which have the potential to perpetuate harms if left unexamined. Further, we argue that system prompt analysis must be incorporated into AI auditing processes, particularly as customisable system prompts become increasingly prevalent in commercial AI deployments.
Regulatory affairs, which sits at the intersection of medicine and law, can benefit significantly from AI-enabled automation. Classification task is the initial step in which manufacturers position their products to regulatory authorities, and it plays a critical role in determining market access, regulatory scrutiny, and ultimately, patient safety. In this study, we investigate a broad range of AI models -- including traditional machine learning (ML) algorithms, deep learning architectures, and large language models -- using a regulatory dataset of medical device descriptions. We evaluate each model along three key dimensions: accuracy, interpretability, and computational cost.
Sparse Autoencoders (SAEs) are a prominent tool in mechanistic interpretability (MI) for decomposing neural network activations into interpretable features. However, the aspiration to identify a canonical set of features is challenged by the observed inconsistency of learned SAE features across different training runs, undermining the reliability and efficiency of MI research. This position paper argues that mechanistic interpretability should prioritize feature consistency in SAEs -- the reliable convergence to equivalent feature sets across independent runs. We propose using the Pairwise Dictionary Mean Correlation Coefficient (PW-MCC) as a practical metric to operationalize consistency and demonstrate that high levels are achievable (0.80 for TopK SAEs on LLM activations) with appropriate architectural choices. Our contributions include detailing the benefits of prioritizing consistency; providing theoretical grounding and synthetic validation using a model organism, which verifies PW-MCC as a reliable proxy for ground-truth recovery; and extending these findings to real-world LLM data, where high feature consistency strongly correlates with the semantic similarity of learned feature explanations. We call for a community-wide shift towards systematically measuring feature consistency to foster robust cumulative progress in MI.
Multimodal learning has seen remarkable progress, particularly with the emergence of large-scale pre-training across various modalities. However, most current approaches are built on the assumption of a deterministic, one-to-one alignment between modalities. This oversimplifies real-world multimodal relationships, where their nature is inherently many-to-many. This phenomenon, named multiplicity, is not a side-effect of noise or annotation error, but an inevitable outcome of semantic abstraction, representational asymmetry, and task-dependent ambiguity in multimodal tasks. This position paper argues that multiplicity is a fundamental bottleneck that manifests across all stages of the multimodal learning pipeline: from data construction to training and evaluation. This paper examines the causes and consequences of multiplicity, and highlights how multiplicity introduces training uncertainty, unreliable evaluation, and low dataset quality. This position calls for new research directions on multimodal learning: novel multiplicity-aware learning frameworks and dataset construction protocols considering multiplicity.
The rapid advancement of large language models (LLMs) and multi-modal LLMs (MLLMs) has historically relied on model-centric scaling through increasing parameter counts from millions to hundreds of billions to drive performance gains. However, as we approach hardware limits on model size, the dominant computational bottleneck has fundamentally shifted to the quadratic cost of self-attention over long token sequences, now driven by ultra-long text contexts, high-resolution images, and extended videos. In this position paper, \textbf{we argue that the focus of research for efficient AI is shifting from model-centric compression to data-centric compression}. We position token compression as the new frontier, which improves AI efficiency via reducing the number of tokens during model training or inference. Through comprehensive analysis, we first examine recent developments in long-context AI across various domains and establish a unified mathematical framework for existing model efficiency strategies, demonstrating why token compression represents a crucial paradigm shift in addressing long-context overhead. Subsequently, we systematically review the research landscape of token compression, analyzing its fundamental benefits and identifying its compelling advantages across diverse scenarios. Furthermore, we provide an in-depth analysis of current challenges in token compression research and outline promising future directions. Ultimately, our work aims to offer a fresh perspective on AI efficiency, synthesize existing research, and catalyze innovative developments to address the challenges that increasing context lengths pose to the AI community's advancement.
Standing in 2025, at a critical juncture in the pursuit of Artificial General Intelligence (AGI), reinforcement fine-tuning (RFT) has demonstrated significant potential in enhancing the reasoning capability of large language models (LLMs) and has led to the development of cutting-edge AI models such as OpenAI-o1 and DeepSeek-R1. Moreover, the efficient application of RFT to enhance the reasoning capability of multimodal large language models (MLLMs) has attracted widespread attention from the community. In this position paper, we argue that reinforcement fine-tuning powers the reasoning capability of multimodal large language models. To begin with, we provide a detailed introduction to the fundamental background knowledge that researchers interested in this field should be familiar with. Furthermore, we meticulously summarize the improvements of RFT in powering reasoning capability of MLLMs into five key points: diverse modalities, diverse tasks and domains, better training algorithms, abundant benchmarks and thriving engineering frameworks. Finally, we propose five promising directions for future research that the community might consider. We hope that this position paper will provide valuable insights to the community at this pivotal stage in the advancement toward AGI. Summary of works done on RFT for MLLMs is available at https://github.com/Sun-Haoyuan23/Awesome-RL-based-Reasoning-MLLMs.