# Research Papers — 2025-12 · bundle page 88 of 175

> Full-text markdown of papers #4351–4400 of 8703 published in 2025-12, newest first. AI-readable mirrors of the full 8703-paper 2025-12 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2025-12/page/88
[← Previous bundle](https://aipod.dev/papers/archive/2025-12/page/87.md)
[Next bundle →](https://aipod.dev/papers/archive/2025-12/page/89.md)

### SAGA: Open-World Mobile Manipulation via Structured Affordance Grounding

Authors: Kuan Fang, Yuxin Chen, Xinghao Zhu, Farzad Niroui, Lingfeng Sun, Jiuguang Wang
Published: 2025-12-14
Categories: cs.RO, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2512.12842v1

We present SAGA, a versatile and adaptive framework for visuomotor control that can generalize across various environments, task objectives, and user specifications. To efficiently learn such capability, our key idea is to disentangle high-level semantic intent from low-level visuomotor control by explicitly grounding task objectives in the observed environment. Using an affordance-based task representation, we express diverse and complex behaviors in a unified, structured form. By leveraging multimodal foundation models, SAGA grounds the proposed task representation to the robot's visual observation as 3D affordance heatmaps, highlighting task-relevant entities while abstracting away spurious appearance variations that would hinder generalization. These grounded affordances enable us to effectively train a conditional policy on multi-task demonstration data for whole-body control. In a unified framework, SAGA can solve tasks specified in different forms, including language instructions, selected points, and example demonstrations, enabling both zero-shot execution and few-shot adaptation. We instantiate SAGA on a quadrupedal manipulator and conduct extensive experiments across eleven real-world tasks. SAGA consistently outperforms end-to-end and modular baselines by substantial margins. Together, these results demonstrate that structured affordance grounding offers a scalable and effective pathway toward generalist mobile manipulation.

---

### PRIVEE: Privacy-Preserving Vertical Federated Learning Against Feature Inference Attacks

Authors: Sindhuja Madabushi, Haider Ali, Ahmad Faraz Khan, Rui Ning, Hongyi Wu, Chunsheng Xin, Ali. R. Butt, Jin-Hee Cho
Published: 2025-12-14
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2512.12840v2

Vertical Federated Learning (VFL) enables collaborative model training across organizations that share common user samples but hold disjoint feature spaces. Despite its potential, VFL is susceptible to feature inference attacks, in which adversarial parties exploit shared confidence scores (prediction probabilities) during inference to reconstruct private input features of other participants. To counter this threat, we propose PRIVEE (PRIvacy-preserving Vertical fEderated lEarning), a novel defense mechanism named after the French word privée, meaning "private." PRIVEE obfuscates confidence scores while preserving critical properties such as relative ranking and inter-score distances. Rather than exposing raw scores, PRIVEE only shares transformed representations, mitigating risk of reconstruction attacks without degrading model prediction accuracy. Extensive experiments show that PRIVEE achieves up to a 30 times increase in reconstruction error (MSE) against feature inference attacks, compared to the strongest competing defense, while preserving full predictive performance against advanced feature inference attacks.

---

### What Matters in Evaluating Book-Length Stories? A Systematic Study of Long Story Evaluation

Authors: Dingyi Yang, Qin Jin
Published: 2025-12-14
Categories: cs.CL
arXiv: https://arxiv.org/abs/2512.12839v1

In this work, we conduct systematic research in a challenging area: the automatic evaluation of book-length stories (>100K tokens). Our study focuses on two key questions: (1) understanding which evaluation aspects matter most to readers, and (2) exploring effective methods for evaluating lengthy stories. We introduce the first large-scale benchmark, LongStoryEval, comprising 600 newly published books with an average length of 121K tokens (maximum 397K). Each book includes its average rating and multiple reader reviews, presented as critiques organized by evaluation aspects. By analyzing all user-mentioned aspects, we propose an evaluation criteria structure and conduct experiments to identify the most significant aspects among the 8 top-level criteria. For evaluation methods, we compare the effectiveness of three types: aggregation-based, incremental-updated, and summary-based evaluations. Our findings reveal that aggregation- and summary-based evaluations perform better, with the former excelling in detail assessment and the latter offering greater efficiency. Building on these insights, we further propose NovelCritique, an 8B model that leverages the efficient summary-based framework to review and score stories across specified aspects. NovelCritique outperforms commercial models like GPT-4o in aligning with human evaluations. Our datasets and codes are available at https://github.com/DingyiYang/LongStoryEval.

---

### Algorithmic Criminal Liability in Greenwashing: Comparing India, United States, and European Union

Authors: Sahibpreet Singh, Manjit Singh
Published: 2025-12-14
Categories: cs.CY, cs.AI, cs.CR
arXiv: https://arxiv.org/abs/2512.12837v1

AI-powered greenwashing has emerged as an insidious challenge within corporate sustainability governance, exacerbating the opacity of environmental disclosures and subverting regulatory oversight. This study conducts a comparative legal analysis of criminal liability for AI-mediated greenwashing across India, the US, and the EU, exposing doctrinal lacunae in attributing culpability when deceptive claims originate from algorithmic systems. Existing statutes exhibit anthropocentric biases by predicating liability on demonstrable human intent, rendering them ill-equipped to address algorithmic deception. The research identifies a critical gap in jurisprudential adaptation, as prevailing fraud statutes remain antiquated vis-à-vis AI-generated misrepresentation. Utilising a doctrinal legal methodology, this study systematically dissects judicial precedents and statutory instruments, yielding results regarding the potential expansion of corporate criminal liability. Findings underscore the viability of strict liability models, recalibrated governance frameworks for AI accountability, and algorithmic due diligence mandates under ESG regimes. Comparative insights reveal jurisdictional disparities, with the EU Corporate Sustainability Due Diligence Directive (CSDDD) offering a potential transnational model. This study contributes to AI ethics and environmental jurisprudence by advocating for a hybrid liability framework integrating algorithmic risk assessment with legal personhood constructs, ensuring algorithmic opacity does not preclude liability enforcement.

---

### Network Level Evaluation of Hangup Susceptibility of HRGCs using Deep Learning and Sensing Techniques: A Goal Towards Safer Future

Authors: Kaustav Chatterjee, Joshua Li, Kundan Parajulee, Jared Schwennesen
Published: 2025-12-14
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2512.12832v2

Steep-profiled Highway Railway Grade Crossings (HRGCs) pose safety hazards to vehicles with low ground clearance, which may become stranded on the tracks, creating risks of train vehicle collisions. This research develops a framework for network level evaluation of hang-up susceptibility of HRGCs. Profile data from different crossings in Oklahoma were collected using both a walking profiler and the Pave3D8K Laser Imaging System. A hybrid deep learning model, combining Long Short Term Memory (LSTM) and Transformer architectures, was developed to reconstruct accurate HRGC profiles from Pave3D8K Laser Imaging System data. Vehicle dimension data from around 350 specialty vehicles were collected at various locations across Oklahoma to enable up-to-date statistical design dimensions. Hang-up susceptibility was analyzed using three vehicle dimension scenarios: (a) median dimension (median wheelbase and ground clearance), (b) 75-25 percentile dimension (75 percentile wheelbase, 25 percentile ground clearance), and (c) worst case dimension (maximum wheelbase and minimum ground clearance). Results indicate 70, 80, and 95 crossings at the highest hang-up risk levels under these scenarios, respectively. An ArcGIS database and a software interface were developed to support transportation agencies in mitigating crossing hazards. This framework advances safety evaluation by integrating next-generation sensing, deep learning, and infrastructure datasets into practical decision support tools.

---

### GradID: Adversarial Detection via Intrinsic Dimensionality of Gradients

Authors: Mohammad Mahdi Razmjoo, Mohammad Mahdi Sharifian, Saeed Bagheri Shouraki
Published: 2025-12-14
Categories: cs.LG, cs.CR, cs.CV
arXiv: https://arxiv.org/abs/2512.12827v1

Despite their remarkable performance, deep neural networks exhibit a critical vulnerability: small, often imperceptible, adversarial perturbations can lead to drastically altered model predictions. Given the stringent reliability demands of applications such as medical diagnosis and autonomous driving, robust detection of such adversarial attacks is paramount. In this paper, we investigate the geometric properties of a model's input loss landscape. We analyze the Intrinsic Dimensionality (ID) of the model's gradient parameters, which quantifies the minimal number of coordinates required to describe the data points on their underlying manifold. We reveal a distinct and consistent difference in the ID for natural and adversarial data, which forms the basis of our proposed detection method. We validate our approach across two distinct operational scenarios. First, in a batch-wise context for identifying malicious data groups, our method demonstrates high efficacy on datasets like MNIST and SVHN. Second, in the critical individual-sample setting, we establish new state-of-the-art results on challenging benchmarks such as CIFAR-10 and MS COCO. Our detector significantly surpasses existing methods against a wide array of attacks, including CW and AutoAttack, achieving detection rates consistently above 92\% on CIFAR-10. The results underscore the robustness of our geometric approach, highlighting that intrinsic dimensionality is a powerful fingerprint for adversarial detection across diverse datasets and attack strategies.

---

### Adapting Multimodal Foundation Models for Few-Shot Learning: A Comprehensive Study on Contrastive Captioners

Authors: N. K. B. M. P. K. B. Narasinghe, Uthayasanker Thayasivam
Published: 2025-12-14
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2512.12824v1

Large-scale multimodal foundation models, particularly Contrastive Captioners (CoCa), have achieved state-of-the-art results by unifying contrastive alignment with generative captioning. While zero-shot transfer capabilities are well-documented, the adaptation of these generative-contrastive hybrids to downstream tasks with extreme data scarcity (few-shot learning) remains under-explored. Existing literature predominantly focuses on dual-encoder architectures like CLIP, leaving a gap in understanding how CoCa's distinct latent space responds to parameter-efficient fine-tuning (PEFT). This paper presents a comprehensive empirical study on adapting the CoCa visual backbone for few-shot image classification. We systematically evaluate a hierarchy of strategies, ranging from training-free hybrid prototyping to deep parameter adaptation via Low-Rank Adaptation (LoRA). First, we identify an "augmentation divergence": while strong data augmentation degrades the performance of linear probing in low-shot settings, it is essential for stabilizing LoRA fine-tuning. We also demonstrate that hybrid objectives incorporating Supervised Contrastive (SupCon) loss yield consistent performance improvements over standard Cross-Entropy across varying shot counts. Crucially, we characterize the sensitivity of training configurations to data scarcity, providing empirical reference settings for scaling regularization, rank, and sampling strategies to facilitate the efficient adaptation of generative-contrastive foundation models.

---

### Lemon: A Unified and Scalable 3D Multimodal Model for Universal Spatial Understanding

Authors: Yongyuan Liang, Xiyao Wang, Yuanchen Ju, Jianwei Yang, Furong Huang
Published: 2025-12-14
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2512.12822v1

Scaling large multimodal models (LMMs) to 3D understanding poses unique challenges: point cloud data is sparse and irregular, existing models rely on fragmented architectures with modality-specific encoders, and training pipelines often suffer from instability and poor scalability. We introduce Lemon, a unified transformer architecture that addresses these challenges by jointly processing 3D point cloud patches and language tokens as a single sequence. Unlike prior work that relies on modality-specific encoders and cross-modal alignment modules, this design enables early spatial-linguistic fusion, eliminates redundant encoders, improves parameter efficiency, and supports more effective model scaling. To handle the complexity of 3D data, we develop a structured patchification and tokenization scheme that preserves spatial context, and a three-stage training curriculum that progressively builds capabilities from object-level recognition to scene-level spatial reasoning. Lemon establishes new state-of-the-art performance across comprehensive 3D understanding and reasoning tasks, from object recognition and captioning to spatial reasoning in 3D scenes, while demonstrating robust scaling properties as model size and training data increase. Our work provides a unified foundation for advancing 3D spatial intelligence in real-world applications.

---

### On the continuity of flows

Authors: Congzhou M Sha
Published: 2025-12-14
Categories: cs.LG, cs.AI, physics.data-an
arXiv: https://arxiv.org/abs/2512.12821v1

Flow matching has emerged as a powerful framework for generative modeling through continuous normalizing flows. We investigate a potential topological constraint: when the prior distribution and target distribution have mismatched topology (e.g., unimodal to multimodal), the optimal velocity field under standard flow matching objectives may exhibit spatial discontinuities. We suggest that this discontinuity arises from the requirement that continuous flows must bifurcate to map a single mode to multiple modes, forcing particles to make discrete routing decisions at intermediate times. Through theoretical analysis on bimodal Gaussian mixtures, we demonstrate that the optimal velocity field exhibits jump discontinuities along decision boundaries, with magnitude approaching infinity as time approaches the target distribution. Our analysis suggests that this phenomenon is not specific to $L^2$ loss, but rather may be a consequence of topological mismatch between distributions. We validate our theory empirically and discuss potential implications for flow matching on manifolds, connecting our findings to recent work on Riemannian flow matching and the challenge of learning discontinuous representations in neural networks.

---

### Hindsight is 20/20: Building Agent Memory that Retains, Recalls, and Reflects

Authors: Chris Latimer, Nicoló Boschi, Andrew Neeser, Chris Bartholomew, Gaurav Srivastava, Xuan Wang, Naren Ramakrishnan
Published: 2025-12-14
Categories: cs.CL, cs.AI, cs.IR, cs.LG
arXiv: https://arxiv.org/abs/2512.12818v1

Agent memory has been touted as a dimension of growth for LLM-based applications, enabling agents that can accumulate experience, adapt across sessions, and move beyond single-shot question answering. The current generation of agent memory systems treats memory as an external layer that extracts salient snippets from conversations, stores them in vector or graph-based stores, and retrieves top-k items into the prompt of an otherwise stateless model. While these systems improve personalization and context carry-over, they still blur the line between evidence and inference, struggle to organize information over long horizons, and offer limited support for agents that must explain their reasoning. We present Hindsight, a memory architecture that treats agent memory as a structured, first-class substrate for reasoning by organizing it into four logical networks that distinguish world facts, agent experiences, synthesized entity summaries, and evolving beliefs. This framework supports three core operations -- retain, recall, and reflect -- that govern how information is added, accessed, and updated. Under this abstraction, a temporal, entity aware memory layer incrementally turns conversational streams into a structured, queryable memory bank, while a reflection layer reasons over this bank to produce answers and to update information in a traceable way. On key long-horizon conversational memory benchmarks like LongMemEval and LoCoMo, Hindsight with an open-source 20B model lifts overall accuracy from 39% to 83.6% over a full-context baseline with the same backbone and outperforms full context GPT-4o. Scaling the backbone further pushes Hindsight to 91.4% on LongMemEval and up to 89.61% on LoCoMo (vs. 75.78% for the strongest prior open system), consistently outperforming existing memory architectures on multi-session and open-domain questions.

---

### Decoding Human and AI Persuasion in National College Debate: Analyzing Prepared Arguments Through Aristotle's Rhetorical Principles

Authors: Mengqian Wu, Jiayi Zhang, Raymond Z. Zhang
Published: 2025-12-14
Categories: cs.HC, cs.AI
arXiv: https://arxiv.org/abs/2512.12817v1

Debate has been widely adopted as a strategy to enhance critical thinking skills in English Language Arts (ELA). One important skill in debate is forming effective argumentation, which requires debaters to select supportive evidence from literature and construct compelling claims. However, the training of this skill largely depends on human coaching, which is labor-intensive and difficult to scale. To better support students in preparing for debates, this study explores the potential of leveraging artificial intelligence to generate effective arguments. Specifically, we prompted GPT-4 to create an evidence card and compared it to those produced by human debaters. The evidence cards outline the arguments students will present and how those arguments will be delivered, including components such as literature-based evidence quotations, summaries of core ideas, verbatim reading scripts, and tags (i.e., titles of the arguments). We compared the quality of the arguments in the evidence cards created by GPT and student debaters using Aristotle's rhetorical principles: ethos (credibility), pathos (emotional appeal), and logos (logical reasoning). Through a systematic qualitative and quantitative analysis, grounded in the rhetorical principles, we identify the strengths and limitations of human and GPT in debate reasoning, outlining areas where AI's focus and justifications align with or diverge from human reasoning. Our findings contribute to the evolving role of AI-assisted learning interventions, offering insights into how student debaters can develop strategies that enhance their argumentation and reasoning skills.

---

### Optimal Resource Allocation for ML Model Training and Deployment under Concept Drift

Authors: Hasan Burhan Beytur, Haris Vikalo, Kevin S Chan, Gustavo de Veciana
Published: 2025-12-14
Categories: cs.LG, cs.NI
arXiv: https://arxiv.org/abs/2512.12816v2

We study how to allocate resources for training and deployment of machine learning (ML) models under concept drift and limited budgets. We consider a setting in which a model provider distributes trained models to multiple clients whose devices support local inference but lack the ability to retrain those models, placing the burden of performance maintenance on the provider. We introduce a model-agnostic framework that captures the interaction between resource allocation, concept drift dynamics, and deployment timing. We show that optimal training policies depend critically on the aging properties of concept durations. Under sudden concept changes, we derive optimal training policies subject to budget constraints when concept durations follow distributions with Decreasing Mean Residual Life (DMRL), and show that intuitive heuristics are provably suboptimal under Increasing Mean Residual Life (IMRL). We further study model deployment under communication constraints, prove that the associated optimization problem is quasi-convex under mild conditions, and propose a randomized scheduling strategy that achieves near-optimal client-side performance. These results offer theoretical and algorithmic foundations for cost-efficient ML model management under concept drift, with implications for continual learning, distributed inference, and adaptive ML systems.

---

### Does Tone Change the Answer? Evaluating Prompt Politeness Effects on Modern LLMs: GPT, Gemini, and LLaMA

Authors: Hanyu Cai, Binqi Shen, Lier Jin, Lan Hu, Xiaojing Fan
Published: 2025-12-14
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2512.12812v2

Prompt engineering has emerged as a critical factor influencing large language model (LLM) performance, yet the impact of pragmatic elements such as linguistic tone and politeness remains underexplored, particularly across different model families. In this work, we propose a systematic evaluation framework to examine how interaction tone affects model accuracy and apply it to three recently released and widely available LLMs: GPT-4o mini (OpenAI), Gemini 2.0 Flash (Google DeepMind), and Llama 4 Scout (Meta). Using the MMMLU benchmark, we evaluate model performance under Very Polite, Neutral, and Very Rude prompt variants across six tasks spanning STEM and Humanities domains, and analyze pairwise accuracy differences with statistical significance testing. Our results show that tone sensitivity is both model-dependent and domain-specific. Neutral or Very Polite prompts generally yield higher accuracy than Very Rude prompts, but statistically significant effects appear only in a subset of Humanities tasks, where rude tone reduces accuracy for GPT and Llama, while Gemini remains comparatively tone-insensitive. When performance is aggregated across tasks within each domain, tone effects diminish and largely lose statistical significance. Compared with earlier research, these findings suggest that dataset scale and coverage materially influence the detection of tone effects. Overall, our study indicates that while interaction tone can matter in specific interpretive settings, modern LLMs are broadly robust to tonal variation in typical mixed-domain use, providing practical guidance for prompt design and model selection in real-world deployments.

---

### OPAL: Operator-Programmed Algorithms for Landscape-Aware Black-Box Optimization

Authors: Junbo Jacob Lian, Mingyang Yu, Kaichen Ouyang, Shengwei Fu, Rui Zhong, Yujun Zhang, Jun Zhang, Huiling Chen
Published: 2025-12-14
Categories: cs.NE, cs.AI
arXiv: https://arxiv.org/abs/2512.12809v1

Black-box optimization often relies on evolutionary and swarm algorithms whose performance is highly problem dependent. We view an optimizer as a short program over a small vocabulary of search operators and learn this operator program separately for each problem instance. We instantiate this idea in Operator-Programmed Algorithms (OPAL), a landscape-aware framework for continuous black-box optimization that uses a small design budget with a standard differential evolution baseline to probe the landscape, builds a $k$-nearest neighbor graph over sampled points, and encodes this trajectory with a graph neural network. A meta-learner then maps the resulting representation to a phase-wise schedule of exploration, restart, and local search operators. On the CEC~2017 test suite, a single meta-trained OPAL policy is statistically competitive with state-of-the-art adaptive differential evolution variants and achieves significant improvements over simpler baselines under nonparametric tests. Ablation studies on CEC~2017 justify the choices for the design phase, the trajectory graph, and the operator-program representation, while the meta-components add only modest wall-clock overhead. Overall, the results indicate that operator-programmed, landscape-aware per-instance design is a practical way forward beyond ad hoc metaphor-based algorithms in black-box optimization.

---

### Fault-Tolerant Sandboxing for AI Coding Agents: A Transactional Approach to Safe Autonomous Execution

Authors: Boyang Yan
Published: 2025-12-14
Categories: cs.AI
arXiv: https://arxiv.org/abs/2512.12806v1

The transition of Large Language Models (LLMs) from passive code generators to autonomous agents introduces significant safety risks, specifically regarding destructive commands and inconsistent system states. Existing commercial solutions often prioritize interactive user safety, enforcing authentication barriers that break the headless loops required for true autonomy. This paper presents a Fault-Tolerant Sandboxing framework designed to mitigate these risks through a policy-based interception layer and a transactional filesystem snapshot mechanism. We hypothesize that wrapping agent actions in atomic transactions can guarantee safety with acceptable latency, outperforming the heavy initialization overhead of containers or the interactive friction of commercial CLIs. We validated this approach by deploying the Minimind-MoE LLM served via nano-vllm on a custom Proxmox-based testbed utilizing EVPN/VXLAN isolation. Experimental results demonstrate a 100\% interception rate for high-risk commands and a 100\% success rate in rolling back failed states. Crucially, our prototype incurs only a 14.5\% performance overhead (approx. 1.8s) per transaction. In contrast, benchmarking against the Gemini CLI sandbox revealed that it requires interactive authentication ("Sign in"), rendering it unusable for headless, autonomous agent workflows.

---

### From Small to Large: Generalization Bounds for Transformers on Variable-Size Inputs

Authors: Anastasiia Alokhina, Pan Li
Published: 2025-12-14
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2512.12805v2

Transformers exhibit a notable property of \emph{size generalization}, demonstrating an ability to extrapolate from smaller token sets to significantly longer ones. This behavior has been documented across diverse applications, including point clouds, graphs, and natural language. Despite its empirical success, this capability still lacks some rigorous theoretical characterizations. In this paper, we develop a theoretical framework to analyze this phenomenon for geometric data, which we represent as discrete samples from a continuous source (e.g., point clouds from manifolds, graphs from graphons). Our core contribution is a bound on the error between the Transformer's output for a discrete sample and its continuous-domain equivalent. We prove that for Transformers with stable positional encodings, this bound is determined by the sampling density and the intrinsic dimensionality of the data manifold. Experiments on graphs and point clouds of various sizes confirm the tightness of our theoretical bound.

---

### Causal Counterfactuals Reconsidered

Authors: Sander Beckers
Published: 2025-12-14
Categories: cs.AI
arXiv: https://arxiv.org/abs/2512.12804v1

I develop a novel semantics for probabilities of counterfactuals that generalizes the standard Pearlian semantics: it applies to probabilistic causal models that cannot be extended into realistic structural causal models and are therefore beyond the scope of Pearl's semantics. This generalization is needed because, as I show, such probabilistic causal models arise even in simple settings. My semantics offer a natural compromize in the long-standing debate between Pearl and Dawid over counterfactuals: I agree with Dawid that universal causal determinism and unrealistic variables should be rejected, but I agree with Pearl that a general semantics of counterfactuals is nonetheless possible. I restrict attention to causal models that satisfy the Markov condition, only contain realistic variables, and are causally complete. Although I formulate my proposal using structural causal models, as does Pearl, I refrain from using so-called response variables. Moreover, I prove that my semantics is equivalent to two other recent proposals that do not involve structural causal models, and that it is in line with various comments on stochastic counterfactuals that have appeared in the literature more broadly. Throughout I also reflect on the universality of the Markov condition and explore a novel generalization of causal abstractions

---

### A Disproof of Large Language Model Consciousness: The Necessity of Continual Learning for Consciousness

Authors: Erik Hoel
Published: 2025-12-14
Categories: q-bio.NC, cs.AI
arXiv: https://arxiv.org/abs/2512.12802v3

Scientific theories of consciousness should be falsifiable and non-trivial. Recent research has given us formal tools to analyze these requirements of falsifiability and non-triviality for theories of consciousness. Surprisingly, many contemporary theories of consciousness fail to pass this bar, including theories based on causal structure but also (as I demonstrate) theories based on function. Herein, I show these requirements of falsifiability and non-triviality especially constrain the potential consciousness of contemporary Large Language Models (LLMs) because of their proximity to systems that are equivalent to LLMs in terms of input/output function; yet, for these functionally equivalent systems, there cannot be any falsifiable and non-trivial theory of consciousness that judges them conscious. This forms the basis of a disproof of contemporary LLM consciousness. I then show a positive result, which is that theories of consciousness based on (or requiring) continual learning do satisfy the stringent formal constraints for a theory of consciousness in humans. Intriguingly, this work supports a hypothesis: If continual learning is linked to consciousness in humans, the current limitations of LLMs (which do not continually learn) are intimately tied to their lack of consciousness.

---

### Learning Common and Salient Generative Factors Between Two Image Datasets

Authors: Yunlong He, Gwilherm Lesné, Ziqian Liu, Michaël Soumm, Pietro Gori
Published: 2025-12-14
Categories: cs.CV
arXiv: https://arxiv.org/abs/2512.12800v1

Recent advancements in image synthesis have enabled high-quality image generation and manipulation. Most works focus on: 1) conditional manipulation, where an image is modified conditioned on a given attribute, or 2) disentangled representation learning, where each latent direction should represent a distinct semantic attribute. In this paper, we focus on a different and less studied research problem, called Contrastive Analysis (CA). Given two image datasets, we want to separate the common generative factors, shared across the two datasets, from the salient ones, specific to only one dataset. Compared to existing methods, which use attributes as supervised signals for editing (e.g., glasses, gender), the proposed method is weaker, since it only uses the dataset signal. We propose a novel framework for CA, that can be adapted to both GAN and Diffusion models, to learn both common and salient factors. By defining new and well-adapted learning strategies and losses, we ensure a relevant separation between common and salient factors, preserving a high-quality generation. We evaluate our approach on diverse datasets, covering human faces, animal images and medical scans. Our framework demonstrates superior separation ability and image quality synthesis compared to prior methods.

---

### DrivePI: Spatial-aware 4D MLLM for Unified Autonomous Driving Understanding, Perception, Prediction and Planning

Authors: Zhe Liu, Runhui Huang, Rui Yang, Siming Yan, Zining Wang, Lu Hou, Di Lin, Xiang Bai, Hengshuang Zhao
Published: 2025-12-14
Categories: cs.CV
arXiv: https://arxiv.org/abs/2512.12799v1

Although multi-modal large language models (MLLMs) have shown strong capabilities across diverse domains, their application in generating fine-grained 3D perception and prediction outputs in autonomous driving remains underexplored. In this paper, we propose DrivePI, a novel spatial-aware 4D MLLM that serves as a unified Vision-Language-Action (VLA) framework that is also compatible with vision-action (VA) models. Our method jointly performs spatial understanding, 3D perception (i.e., 3D occupancy), prediction (i.e., occupancy flow), and planning (i.e., action outputs) in parallel through end-to-end optimization. To obtain both precise geometric information and rich visual appearance, our approach integrates point clouds, multi-view images, and language instructions within a unified MLLM architecture. We further develop a data engine to generate text-occupancy and text-flow QA pairs for 4D spatial understanding. Remarkably, with only a 0.5B Qwen2.5 model as MLLM backbone, DrivePI as a single unified model matches or exceeds both existing VLA models and specialized VA models. Specifically, compared to VLA models, DrivePI outperforms OpenDriveVLA-7B by 2.5% mean accuracy on nuScenes-QA and reduces collision rate by 70% over ORION (from 0.37% to 0.11%) on nuScenes. Against specialized VA models, DrivePI surpasses FB-OCC by 10.3 RayIoU for 3D occupancy on OpenOcc, reduces the mAVE from 0.591 to 0.509 for occupancy flow on OpenOcc, and achieves 32% lower L2 error than VAD (from 0.72m to 0.49m) for planning on nuScenes. Code will be available at https://github.com/happinesslz/DrivePI

---

### Diffusion Model-Based Posterior Sampling in Full Waveform Inversion

Authors: Mohammad H. Taufik, Tariq Alkhalifah
Published: 2025-12-14
Categories: physics.geo-ph, stat.ML
arXiv: https://arxiv.org/abs/2512.12797v1

Bayesian full waveform inversion (FWI) offers uncertainty-aware subsurface models; however, posterior sampling directly on observed seismic shot records is rarely practical at the field scale because each sample requires numerous wave-equation solves. We aim to make such sampling feasible for large surveys while preserving calibration, that is, high uncertainty in less illuminated areas. Our approach couples diffusion-based posterior sampling with simultaneous-source FWI data. At each diffusion noise level, a network predicts a clean velocity model. We then apply a stochastic refinement step in model space using Langevin dynamics under the wave-equation likelihood and reintroduce noise to decouple successive levels before proceeding. Simultaneous-source batches reduce forward and adjoint solves approximately in proportion to the supergather size, while an unconditional diffusion prior trained on velocity patches and volumes helps suppress source-related numerical artefacts. We evaluate the method on three 2D synthetic datasets (SEG/EAGE Overthrust, SEG/EAGE Salt, SEAM Arid), a 2D field line, and a 3D upscaling study. Relative to a particle-based variational baseline, namely Stein variational gradient descent without a learned prior and with single-source (non-simultaneous-source) FWI, our sampler achieves lower model error and better data fit at a substantially reduced computational cost. By aligning encoded-shot likelihoods with diffusion-based sampling and exploiting straightforward parallelization over samples and source batches, the method provides a practical path to calibrated posterior inference on observed shot records that scales to large 2D and 3D problems.

---

### TRACER: Transfer Learning based Real-time Adaptation for Clinical Evolving Risk

Authors: Mengying Yan, Ziye Tian, Siqi Li, Nan Liu, Benjamin A. Goldstein, Molei Liu, Chuan Hong
Published: 2025-12-14
Categories: cs.LG, stat.ME
arXiv: https://arxiv.org/abs/2512.12795v1

Clinical decision support tools built on electronic health records often experience performance drift due to temporal population shifts, particularly when changes in the clinical environment initially affect only a subset of patients, resulting in a transition to mixed populations. Such case-mix changes commonly arise following system-level operational updates or the emergence of new diseases, such as COVID-19. We propose TRACER (Transfer Learning-based Real-time Adaptation for Clinical Evolving Risk), a framework that identifies encounter-level transition membership and adapts predictive models using transfer learning without full retraining. In simulation studies, TRACER outperformed static models trained on historical or contemporary data. In a real-world application predicting hospital admission following emergency department visits across the COVID-19 transition, TRACER improved both discrimination and calibration. TRACER provides a scalable approach for maintaining robust predictive performance under evolving and heterogeneous clinical conditions.

---

### Liquid Reasoning Transformers: A Sudoku-Based Prototype for Chess-Scale Algorithmic Tasks

Authors: Shivansh Sahni, Wenzhi Zhang
Published: 2025-12-14
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2512.12792v1

The Liquid Reasoning Transformer (LRT) is a transformer architecture designed for inference with adaptive depths using iterative changes, discard-based correction, and a learned stopping mechanism. Instead of relying on a single feedforward pass, the model updates a recurrent reasoning token across multiple internal steps, allowing it to correct early errors and allocate computation based on input difficulty. We evaluate the LRT on Sudoku as a controlled testbed for structured reasoning and show that it achieves strong performance, reaching 98.68% digit accuracy and 36.30% full-puzzle accuracy without using symbolic rules or search. Analyzing internal patterns shows that the discard and stop gates play different, important roles in stabilizing inferences and adjusting computational depth. We discuss how these mechanisms extend naturally to chess-scale reasoning tasks and outline extensions for multi-token reasoning and larger domains.

---

### Beyond Task Completion: An Assessment Framework for Evaluating Agentic AI Systems

Authors: Sreemaee Akshathala, Bassam Adnan, Mahisha Ramesh, Karthik Vaidhyanathan, Basil Muhammed, Kannan Parthasarathy
Published: 2025-12-14
Categories: cs.MA, cs.AI, cs.SE
arXiv: https://arxiv.org/abs/2512.12791v2

Recent advances in agentic AI have shifted the focus from standalone Large Language Models (LLMs) to integrated systems that combine LLMs with tools, memory, and other agents to perform complex tasks. These multi-agent architectures enable coordinated reasoning, planning, and execution across diverse domains, allowing agents to collaboratively automate complex workflows. Despite these advances, evaluation and assessment of LLM agents and the multi-agent systems they constitute remain a fundamental challenge. Although various approaches have been proposed in the software engineering literature for evaluating conventional software components, existing methods for AI-based systems often overlook the non-deterministic nature of models. This non-determinism introduces behavioral uncertainty during execution, yet existing evaluations rely on binary task completion metrics that fail to capture it. Evaluating agentic systems therefore requires examining additional dimensions, including the agent ability to invoke tools, ingest and retrieve memory, collaborate with other agents, and interact effectively with its environment. These challenges emerged during our ongoing industry collaboration with MontyCloud Inc., when we deployed an agentic system in production. These limitations surfaced during deployment, highlighting practical gaps in the current evaluation methods and the need for a systematic assessment of agent behavior beyond task outcomes. Informed by these observations and established definitions of agentic systems, we propose an end-to-end Agent Assessment Framework with four evaluation pillars encompassing LLMs, Memory, Tools, and Environment. We validate the framework on a representative Autonomous CloudOps use case, where experiments reveal behavioral deviations overlooked by conventional metrics, demonstrating its effectiveness in capturing runtime uncertainties.

---

### L-STEC: Learned Video Compression with Long-term Spatio-Temporal Enhanced Context

Authors: Tiange Zhang, Zhimeng Huang, Xiandong Meng, Kai Zhang, Zhipin Deng, Siwei Ma
Published: 2025-12-14
Categories: cs.CV
arXiv: https://arxiv.org/abs/2512.12790v1

Neural Video Compression has emerged in recent years, with condition-based frameworks outperforming traditional codecs. However, most existing methods rely solely on the previous frame's features to predict temporal context, leading to two critical issues. First, the short reference window misses long-term dependencies and fine texture details. Second, propagating only feature-level information accumulates errors over frames, causing prediction inaccuracies and loss of subtle textures. To address these, we propose the Long-term Spatio-Temporal Enhanced Context (L-STEC) method. We first extend the reference chain with LSTM to capture long-term dependencies. We then incorporate warped spatial context from the pixel domain, fusing spatio-temporal information through a multi-receptive field network to better preserve reference details. Experimental results show that L-STEC significantly improves compression by enriching contextual information, achieving 37.01% bitrate savings in PSNR and 31.65% in MS-SSIM compared to DCVC-TCM, outperforming both VTM-17.0 and DCVC-FM and establishing new state-of-the-art performance.

---

### Unveiling Statistical Significance of Online Regression over Multiple Datasets

Authors: Mohammad Abu-Shaira, Weishi Shi
Published: 2025-12-14
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2512.12787v1

Despite extensive focus on techniques for evaluating the performance of two learning algorithms on a single dataset, the critical challenge of developing statistical tests to compare multiple algorithms across various datasets has been largely overlooked in most machine learning research. Additionally, in the realm of Online Learning, ensuring statistical significance is essential to validate continuous learning processes, particularly for achieving rapid convergence and effectively managing concept drifts in a timely manner. Robust statistical methods are needed to assess the significance of performance differences as data evolves over time. This article examines the state-of-the-art online regression models and empirically evaluates several suitable tests. To compare multiple online regression models across various datasets, we employed the Friedman test along with corresponding post-hoc tests. For thorough evaluations, utilizing both real and synthetic datasets with 5-fold cross-validation and seed averaging ensures comprehensive assessment across various data subsets. Our tests generally confirmed the performance of competitive baselines as consistent with their individual reports. However, some statistical test results also indicate that there is still room for improvement in certain aspects of state-of-the-art methods.

---

### OLC-WA: Drift Aware Tuning-Free Online Classification with Weighted Average

Authors: Mohammad Abu Shaira, Yunhe Feng, Heng Fan, Weishi Shi
Published: 2025-12-14
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2512.12785v1

Real-world data sets often exhibit temporal dynamics characterized by evolving data distributions. Disregarding this phenomenon, commonly referred to as concept drift, can significantly diminish a model's predictive accuracy. Furthermore, the presence of hyperparameters in online models exacerbates this issue. These parameters are typically fixed and cannot be dynamically adjusted by the user in response to the evolving data distribution. This paper introduces Online Classification with Weighted Average (OLC-WA), an adaptive, hyperparameter-free online classification model equipped with an automated optimization mechanism. OLC-WA operates by blending incoming data streams with an existing base model. This blending is facilitated by an exponentially weighted moving average. Furthermore, an integrated optimization mechanism dynamically detects concept drift, quantifies its magnitude, and adjusts the model based on the observed data stream characteristics. This approach empowers the model to effectively adapt to evolving data distributions within streaming environments. Rigorous empirical evaluation across diverse benchmark datasets shows that OLC-WA achieves performance comparable to batch models in stationary environments, maintaining accuracy within 1-3%, and surpasses leading online baselines by 10-25% under drift, demonstrating its effectiveness in adapting to dynamic data streams.

---

### Credit Risk Estimation with Non-Financial Features: Evidence from a Synthetic Istanbul Dataset

Authors: Atalay Denknalbant, Emre Sezdi, Zeki Furkan Kutlu
Published: 2025-12-14
Categories: cs.LG, q-fin.ST, stat.AP
arXiv: https://arxiv.org/abs/2512.12783v3

Financial exclusion constrains entrepreneurship, increases income volatility, and widens wealth gaps. Underbanked consumers in Istanbul often have no bureau file because their earnings and payments flow through informal channels. To study how such borrowers can be evaluated we create a synthetic dataset of one hundred thousand Istanbul residents that reproduces first quarter 2025 TÜİK (TURKSTAT) census marginals and telecom usage patterns. Retrieval augmented generation feeds these public statistics into the OpenAI o3 model, which synthesises realistic yet private records. Each profile contains seven socio demographic variables and nine alternative attributes that describe phone specifications, online shopping rhythm, subscription spend, car ownership, monthly rent, and a credit card flag. To test the impact of the alternative financial data CatBoost, LightGBM, and XGBoost are each trained in two versions. Demo models use only the socio demographic variables; Full models include both socio demographic and alternative attributes. Across five fold stratified validation the alternative block raises area under the curve by about one point three percentage and lifts balanced F 1 from roughly 0.84 to 0.95, a fourteen percent gain. We contribute an open Istanbul 2025 Q1 synthetic dataset, a fully reproducible modeling pipeline, and empirical evidence that a concise set of behavioural attributes can approach bureau level discrimination power while serving borrowers who lack formal credit records. These findings give lenders and regulators a transparent blueprint for extending fair and safe credit access to the underbanked.

---

### OLR-WAA: Adaptive and Drift-Resilient Online Regression with Dynamic Weighted Averaging

Authors: Mohammad Abu-Shaira, Weishi Shi
Published: 2025-12-14
Categories: cs.LG
arXiv: https://arxiv.org/abs/2512.12779v1

Real-world datasets frequently exhibit evolving data distributions, reflecting temporal variations and underlying shifts. Overlooking this phenomenon, known as concept drift, can substantially degrade the predictive performance of the model. Furthermore, the presence of hyperparameters in online models exacerbates this issue, as these parameters are typically fixed and lack the flexibility to dynamically adjust to evolving data. This paper introduces "OLR-WAA: An Adaptive and Drift-Resilient Online Regression with Dynamic Weighted Average", a hyperparameter-free model designed to tackle the challenges of non-stationary data streams and enable effective, continuous adaptation. The objective is to strike a balance between model stability and adaptability. OLR-WAA incrementally updates its base model by integrating incoming data streams, utilizing an exponentially weighted moving average. It further introduces a unique optimization mechanism that dynamically detects concept drift, quantifies its magnitude, and adjusts the model based on real-time data characteristics. Rigorous evaluations show that it matches batch regression performance in static settings and consistently outperforms or rivals state-of-the-art online models, confirming its effectiveness. Concept drift datasets reveal a performance gap that OLR-WAA effectively bridges, setting it apart from other online models. In addition, the model effectively handles confidence-based scenarios through a conservative update strategy that prioritizes stable, high-confidence data points. Notably, OLR-WAA converges rapidly, consistently yielding higher R2 values compared to other online models.

---

### State over Tokens: Characterizing the Role of Reasoning Tokens

Authors: Mosh Levy, Zohar Elyoseph, Shauli Ravfogel, Yoav Goldberg
Published: 2025-12-14
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2512.12777v1

Large Language Models (LLMs) can generate reasoning tokens before their final answer to boost performance on complex tasks. While these sequences seem like human thought processes, empirical evidence reveals that they are not a faithful explanation of the model's actual reasoning process. To address this gap between appearance and function, we introduce the State over Tokens (SoT) conceptual framework. SoT reframes reasoning tokens not as a linguistic narrative, but as an externalized computational state -- the sole persistent information carrier across the model's stateless generation cycles. This explains how the tokens can drive correct reasoning without being a faithful explanation when read as text and surfaces previously overlooked research questions on these tokens. We argue that to truly understand the process that LLMs do, research must move beyond reading the reasoning tokens as text and focus on decoding them as state.

---

### Persistent Personas? Role-Playing, Instruction Following, and Safety in Extended Interactions

Authors: Pedro Henrique Luz de Araujo, Michael A. Hedderich, Ali Modarressi, Hinrich Schuetze, Benjamin Roth
Published: 2025-12-14
Categories: cs.CL
arXiv: https://arxiv.org/abs/2512.12775v2

Persona-assigned large language models (LLMs) are used in domains such as education, healthcare, and sociodemographic simulation. Yet, they are typically evaluated only in short, single-round settings that do not reflect real-world usage. We introduce an evaluation protocol that combines long persona dialogues (over 100 rounds) and evaluation datasets to create dialogue-conditioned benchmarks that can robustly measure long-context effects. We then investigate the effects of dialogue length on persona fidelity, instruction-following, and safety of seven state-of-the-art open- and closed-weight LLMs. We find that persona fidelity degrades over the course of dialogues, especially in goal-oriented conversations, where models must sustain both persona fidelity and instruction following. We identify a trade-off between fidelity and instruction following, with non-persona baselines initially outperforming persona-assigned models; as dialogues progress and fidelity fades, persona responses become increasingly similar to baseline responses. Our findings highlight the fragility of persona applications in extended interactions and our work provides a protocol to systematically measure such failures.

---

### Persistent Personas? Role-Playing, Instruction Following, and Safety in Extended Interactions

Authors: Pedro Henrique Luz de Araujo, Michael A. Hedderich, Ali Modarressi, Hinrich Schuetze, Benjamin Roth
Published: 2025-12-14
Categories: cs.CL
arXiv: https://arxiv.org/abs/2512.12775v1

Persona-assigned large language models (LLMs) are used in domains such as education, healthcare, and sociodemographic simulation. Yet, they are typically evaluated only in short, single-round settings that do not reflect real-world usage. We introduce an evaluation protocol that combines long persona dialogues (over 100 rounds) and evaluation datasets to create dialogue-conditioned benchmarks that can robustly measure long-context effects. We then investigate the effects of dialogue length on persona fidelity, instruction-following, and safety of seven state-of-the-art open- and closed-weight LLMs. We find that persona fidelity degrades over the course of dialogues, especially in goal-oriented conversations, where models must sustain both persona fidelity and instruction following. We identify a trade-off between fidelity and instruction following, with non-persona baselines initially outperforming persona-assigned models; as dialogues progress and fidelity fades, persona responses become increasingly similar to baseline responses. Our findings highlight the fragility of persona applications in extended interactions and our work provides a protocol to systematically measure such failures.

---

### Fast 2DGS: Efficient Image Representation with Deep Gaussian Prior

Authors: Hao Wang, Ashish Bastola, Chaoyi Zhou, Wenhui Zhu, Xiwen Chen, Xuanzhao Dong, Siyu Huang, Abolfazl Razi
Published: 2025-12-14
Categories: cs.CV
arXiv: https://arxiv.org/abs/2512.12774v1

As generative models become increasingly capable of producing high-fidelity visual content, the demand for efficient, interpretable, and editable image representations has grown substantially. Recent advances in 2D Gaussian Splatting (2DGS) have emerged as a promising solution, offering explicit control, high interpretability, and real-time rendering capabilities (>1000 FPS). However, high-quality 2DGS typically requires post-optimization. Existing methods adopt random or heuristics (e.g., gradient maps), which are often insensitive to image complexity and lead to slow convergence (>10s). More recent approaches introduce learnable networks to predict initial Gaussian configurations, but at the cost of increased computational and architectural complexity. To bridge this gap, we present Fast-2DGS, a lightweight framework for efficient Gaussian image representation. Specifically, we introduce Deep Gaussian Prior, implemented as a conditional network to capture the spatial distribution of Gaussian primitives under different complexities. In addition, we propose an attribute regression network to predict dense Gaussian properties. Experiments demonstrate that this disentangled architecture achieves high-quality reconstruction in a single forward pass, followed by minimal fine-tuning. More importantly, our approach significantly reduces computational cost without compromising visual quality, bringing 2DGS closer to industry-ready deployment.

---

### Designing The Drive: Enhancing User Experience through Adaptive Interfaces in Autonomous Vehicles

Authors: Reeteesha Roy
Published: 2025-12-14
Categories: cs.HC, cs.AI
arXiv: https://arxiv.org/abs/2512.12773v1

With the recent development and integration of autonomous vehicles (AVs) in transportation systems of the modern world, the emphasis on customizing user interfaces to optimize the overall user experience has been growing expediently. Therefore, understanding user needs and preferences is essential to the acceptance and trust of these technologies as they continue to grow in prevalence. This paper addresses the implementation of HCI principles in the personalization of interfaces to improve safety, security, and usability for the users. This paper explores the way that personalized interfaces can be devised to increase user engagement and satisfaction through various HCI strategies such as adaptive design, multi-modal interaction, and user feedback mechanisms. Moreover, this paper puts emphasis on factors of transparency and user control in the design of an interface; hence, allowing users to design or modify their experience could foster an increase in trust in autonomous systems. In so doing, this research touches on the quite influential role HCI will play in this future scenario of autonomous vehicles while designing to ensure relevance to the diverse needs of users while maintaining high standards of safety and security. Discussing various HCI strategies such as adaptive design, multi-modal interaction, and feedback mechanisms to the user, this paper demonstrates how personalized interfaces can enhance significantly both user engagement and satisfaction. Transparency and user control also in designing an interface are further discussed, pointing out the need for a prerequisite condition of enabling the user to take control of their experience as a state of trust in autonomous systems. In summary, this paper points out the role of HCI in the development of autonomous vehicles and addresses numerous needs with respect to those enforced safety and security standards.

---

### JointAVBench: A Benchmark for Joint Audio-Visual Reasoning Evaluation

Authors: Jianghan Chao, Jianzhang Gao, Wenhui Tan, Yuchong Sun, Ruihua Song, Liyun Ru
Published: 2025-12-14
Categories: cs.MM, cs.CV
arXiv: https://arxiv.org/abs/2512.12772v2

Understanding videos inherently requires reasoning over both visual and auditory information. To properly evaluate Omni-Large Language Models (Omni-LLMs), which are capable of processing multi-modal information including vision and audio, an effective benchmark must comprehensively cover three key aspects: (1) multi-modal dependency (i.e., questions that cannot be answered using vision or audio alone), (2) diverse audio information types (e.g., speech, sound events), and (3) varying scene spans. However, existing datasets fall short in one or more of these dimensions, limiting strict and comprehensive evaluation. To address this gap, we introduce JointAVBench, a novel benchmark with strict audio-video correlation, spanning five cognitive dimensions, four audio information types (speech, sound events, music, vocal traits), and three scene spans (single-, cross-, and full-scene). Given the high cost of manual annotation, we propose an automated pipeline that leverages state-of-the-art vision-LLMs, audio-LLMs, and general-purpose LLMs to synthesize questions and answers that strictly require joint audio-visual understanding. We evaluate leading vision-only, audio-only, and Omni-LLMs on our dataset. Results show that even the best-performing Omni-LLM achieves an average accuracy of only 65.3\%, outperforming uni-modal baselines but revealing substantial room for improvement, especially in cross-scene reasoning.

---

### JointAVBench: A Benchmark for Joint Audio-Visual Reasoning Evaluation

Authors: Jianghan Chao, Jianzhang Gao, Wenhui Tan, Yuchong Sun, Ruihua Song, Liyun Ru
Published: 2025-12-14
Categories: cs.MM, cs.CV
arXiv: https://arxiv.org/abs/2512.12772v1

Understanding videos inherently requires reasoning over both visual and auditory information. To properly evaluate Omni-Large Language Models (Omni-LLMs), which are capable of processing multi-modal information including vision and audio, an effective benchmark must comprehensively cover three key aspects: (1) multi-modal dependency (i.e., questions that cannot be answered using vision or audio alone), (2) diverse audio information types (e.g., speech, sound events), and (3) varying scene spans. However, existing datasets fall short in one or more of these dimensions, limiting strict and comprehensive evaluation. To address this gap, we introduce JointAVBench, a novel benchmark with strict audio-video correlation, spanning five cognitive dimensions, four audio information types (speech, sound events, music, vocal traits), and three scene spans (single-, cross-, and full-scene). Given the high cost of manual annotation, we propose an automated pipeline that leverages state-of-the-art vision-LLMs, audio-LLMs, and general-purpose LLMs to synthesize questions and answers that strictly require joint audio-visual understanding. We evaluate leading vision-only, audio-only, and Omni-LLMs on our dataset. Results show that even the best-performing Omni-LLM achieves an average accuracy of only 62.6\%, outperforming uni-modal baselines but revealing substantial room for improvement, especially in cross-scene reasoning.

---

### Curió-Edu 7B: Examining Data Selection Impacts in LLM Continued Pretraining

Authors: Thales Sales Almeida, Rodrigo Nogueira, Hélio Pedrini
Published: 2025-12-14
Categories: cs.CL
arXiv: https://arxiv.org/abs/2512.12770v1

Continued pretraining extends a language model's capabilities by further exposing it to additional data, often tailored to a specific linguistic or domain context. This strategy has emerged as an efficient alternative to full retraining when adapting general-purpose models to new settings. In this work, we investigate this paradigm through Curió 7B, a 7-billion-parameter model derived from LLaMA-2 and trained on 100 billion Portuguese tokens from the ClassiCC-PT corpus - the most extensive Portuguese-specific continued-pretraining effort above the three-billion-parameter scale to date. Beyond scale, we investigate whether quantity alone suffices or whether data quality plays a decisive role in linguistic adaptation. To this end, we introduce Curió-Edu 7B, a variant trained exclusively on the educational and STEM-filtered subset of the same corpus, totaling just 10 billion tokens. Despite using only 10% of the data and 20% of the computation, Curió-Edu 7B surpasses the full-corpus model in our evaluations, demonstrating that data selection can be fundamental even when adapting models with limited prior exposure to the target language. The developed models are available at https://huggingface.co/collections/ClassiCC-Corpus/curio-edu

---

### Adaptive Edge-Cloud Inference for Speech-to-Action Systems Using ASR and Large Language Models

Authors: Mohammad Jalili Torkamani, Israt Zarin
Published: 2025-12-14
Categories: cs.SD, cs.AI
arXiv: https://arxiv.org/abs/2512.12769v2

Voice-based interaction has emerged as a natural and intuitive modality for controlling IoT devices. However, speech-driven edge devices face a fundamental trade-off between cloud-based solutions, which offer stronger language understanding capabilities at the cost of latency, connectivity dependence, and privacy concerns, and edge-based solutions, which provide low latency and improved privacy but are limited by computational constraints. This paper presents ASTA, an adaptive speech-to-action solution that dynamically routes voice commands between edge and cloud inference to balance performance and system resource utilization. ASTA integrates on-device automatic speech recognition and lightweight offline language-model inference with cloud-based LLM processing, guided by real-time system metrics such as CPU workload, device temperature, and network latency. A metric-aware routing mechanism selects the inference path at runtime, while a rule-based command validation and repair component ensures successful end-to-end command execution. We implemented our solution on an NVIDIA Jetson-based edge platform and evaluated it using a diverse dataset of 80 spoken commands. Experimental results show that ASTA successfully routes all input commands for execution, achieving a balanced distribution between online and offline inference. The system attains an ASR accuracy of 62.5% and generates executable commands without repair for only 47.5% of inputs, highlighting the importance of the repair mechanism in improving robustness. These results suggest that adaptive edge-cloud orchestration is a viable approach for resilient and resource-aware voice-controlled IoT systems.

---

### Adaptive Edge-Cloud Inference for Speech-to-Action Systems Using ASR and Large Language Models (ASTA)

Authors: Mohammad Jalili Torkamani, Israt Zarin
Published: 2025-12-14
Categories: cs.SD, cs.AI
arXiv: https://arxiv.org/abs/2512.12769v1

Voice-based interaction has emerged as a natural and intuitive modality for controlling IoT devices. However, speech-driven edge devices face a fundamental trade-off between cloud-based solutions, which offer stronger language understanding capabilities at the cost of latency, connectivity dependence, and privacy concerns, and edge-based solutions, which provide low latency and improved privacy but are limited by computational constraints. This paper presents ASTA, an adaptive speech-to-action solution that dynamically routes voice commands between edge and cloud inference to balance performance and system resource utilization. ASTA integrates on-device automatic speech recognition and lightweight offline language-model inference with cloud-based LLM processing, guided by real-time system metrics such as CPU workload, device temperature, and network latency. A metric-aware routing mechanism selects the inference path at runtime, while a rule-based command validation and repair component ensures successful end-to-end command execution. We implemented our solution on an NVIDIA Jetson-based edge platform and evaluated it using a diverse dataset of 80 spoken commands. Experimental results show that ASTA successfully routes all input commands for execution, achieving a balanced distribution between online and offline inference. The system attains an ASR accuracy of 62.5% and generates executable commands without repair for only 47.5% of inputs, highlighting the importance of the repair mechanism in improving robustness. These results suggest that adaptive edge-cloud orchestration is a viable approach for resilient and resource-aware voice-controlled IoT systems.

---

### CoRe3D: Collaborative Reasoning as a Foundation for 3D Intelligence

Authors: Tianjiao Yu, Xinzhuo Li, Yifan Shen, Yuanzhe Liu, Ismini Lourentzou
Published: 2025-12-14
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2512.12768v2

Recent advances in large multimodal models suggest that explicit reasoning mechanisms play a critical role in improving model reliability, interpretability, and cross-modal alignment. While such reasoning-centric approaches have been proven effective in language and vision tasks, their extension to 3D remains underdeveloped. CoRe3D introduces a unified 3D understanding and generation reasoning framework that jointly operates over semantic and spatial abstractions, enabling high-level intent inferred from language to directly guide low-level 3D content formation. Central to this design is a spatially grounded reasoning representation that decomposes 3D latent space into localized regions, allowing the model to reason over geometry in a compositional and procedural manner. By tightly coupling semantic chain-of-thought inference with structured spatial reasoning, CoRe3D produces 3D outputs that exhibit strong local consistency and faithful alignment with linguistic descriptions.

---

### CoRe3D: Collaborative Reasoning as a Foundation for 3D Intelligence

Authors: Tianjiao Yu, Xinzhuo Li, Yifan Shen, Yuanzhe Liu, Ismini Lourentzou
Published: 2025-12-14
Categories: cs.CV, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2512.12768v1

Recent advances in large multimodal models suggest that explicit reasoning mechanisms play a critical role in improving model reliability, interpretability, and cross-modal alignment. While such reasoning-centric approaches have been proven effective in language and vision tasks, their extension to 3D remains underdeveloped. CoRe3D introduces a unified 3D understanding and generation reasoning framework that jointly operates over semantic and spatial abstractions, enabling high-level intent inferred from language to directly guide low-level 3D content formation. Central to this design is a spatially grounded reasoning representation that decomposes 3D latent space into localized regions, allowing the model to reason over geometry in a compositional and procedural manner. By tightly coupling semantic chain-of-thought inference with structured spatial reasoning, CoRe3D produces 3D outputs that exhibit strong local consistency and faithful alignment with linguistic descriptions.

---

### Random matrix theory of sparse neuronal networks with heterogeneous timescales

Authors: Thiparat Chotibut, Oleg Evnin, Weerawit Horinouchi
Published: 2025-12-14
Categories: q-bio.NC, cond-mat.dis-nn, cs.LG, hep-th, math.PR
arXiv: https://arxiv.org/abs/2512.12767v2

Training recurrent neuronal networks consisting of excitatory (E) and inhibitory (I) units with additive noise for working memory computation slows and diversifies inhibitory timescales, leading to improved task performance that is attributed to emergent marginally stable equilibria (PNAS 122 (2025) e2316745122). Yet the link between trained network characteristics and their roles in shaping desirable dynamical landscapes remains unexplored. Here, we investigate the Jacobian matrices describing the dynamics near these equilibria and show that they are sparse, non-Hermitian rectangular-block matrices modified by heterogeneous synaptic decay timescales and activation-function gains. We specify a random matrix ensemble that faithfully captures the spectra of trained Jacobian matrices, arising from the inhibitory core - excitatory periphery network motif (pruned E weights, broadly distributed I weights) observed post-training. An analytic theory of this ensemble is developed using statistical field theory methods: a Hermitized resolvent representation of the spectral density is processed with a supersymmetry-based treatment in the style of Fyodorov and Mirlin. In this manner, an analytic description of the spectral edge is obtained, relating statistical parameters of the Jacobians (sparsity, weight variances, E/I ratio, and the distributions of timescales and gains) to near-critical features of the equilibria essential for robust working memory computation.

---

### Federated Learning with Feedback Alignment

Authors: Incheol Baek, Hyungbin Kim, Minseo Kim, Yon Dohn Chung
Published: 2025-12-14
Categories: cs.LG, cs.AI, cs.CV
arXiv: https://arxiv.org/abs/2512.12762v1

Federated Learning (FL) enables collaborative training across multiple clients while preserving data privacy, yet it struggles with data heterogeneity, where clients' data are not distributed independently and identically (non-IID). This causes local drift, hindering global model convergence. To address this, we introduce Federated Learning with Feedback Alignment (FLFA), a novel framework that integrates feedback alignment into FL. FLFA uses the global model's weights as a shared feedback matrix during local training's backward pass, aligning local updates with the global model efficiently. This approach mitigates local drift with minimal additional computational cost and no extra communication overhead. Our theoretical analysis supports FLFA's design by showing how it alleviates local drift and demonstrates robust convergence for both local and global models. Empirical evaluations, including accuracy comparisons and measurements of local drift, further illustrate that FLFA can enhance other FL methods demonstrating its effectiveness.

---

### Intelligent Scientific Literature Explorer using Machine Learning (ISLE)

Authors: Sina Jani, Arman Heidari, Amirmohammad Anvari, Zahra Rahimi
Published: 2025-12-14
Categories: cs.IR, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2512.12760v1

The rapid acceleration of scientific publishing has created substantial challenges for researchers attempting to discover, contextualize, and interpret relevant literature. Traditional keyword-based search systems provide limited semantic understanding, while existing AI-driven tools typically focus on isolated tasks such as retrieval, clustering, or bibliometric visualization. This paper presents an integrated system for scientific literature exploration that combines large-scale data acquisition, hybrid retrieval, semantic topic modeling, and heterogeneous knowledge graph construction. The system builds a comprehensive corpus by merging full-text data from arXiv with structured metadata from OpenAlex. A hybrid retrieval architecture fuses BM25 lexical search with embedding-based semantic search using Reciprocal Rank Fusion. Topic modeling is performed on retrieved results using BERTopic or non-negative matrix factorization depending on computational resources. A knowledge graph unifies papers, authors, institutions, countries, and extracted topics into an interpretable structure. The system provides a multi-layered exploration environment that reveals not only relevant publications but also the conceptual and relational landscape surrounding a query. Evaluation across multiple queries demonstrates improvements in retrieval relevance, topic coherence, and interpretability. The proposed framework contributes an extensible foundation for AI-assisted scientific discovery.

---

### FysicsWorld: A Unified Full-Modality Benchmark for Any-to-Any Understanding, Generation, and Reasoning

Authors: Yue Jiang, Dingkang Yang, Minghao Han, Jinghang Han, Zizhi Chen, Yizhou Liu, Mingcheng Li, Peng Zhai, Lihua Zhang
Published: 2025-12-14
Categories: cs.CV
arXiv: https://arxiv.org/abs/2512.12756v1

Despite rapid progress in multimodal large language models (MLLMs) and emerging omni-modal architectures, current benchmarks remain limited in scope and integration, suffering from incomplete modality coverage, restricted interaction to text-centric outputs, and weak interdependence and complementarity among modalities. To bridge these gaps, we introduce FysicsWorld, the first unified full-modality benchmark that supports bidirectional input-output across image, video, audio, and text, enabling comprehensive any-to-any evaluation across understanding, generation, and reasoning. FysicsWorld encompasses 16 primary tasks and 3,268 curated samples, aggregated from over 40 high-quality sources and covering a rich set of open-domain categories with diverse question types. We also propose the Cross-Modal Complementarity Screening (CMCS) strategy integrated in a systematic data construction framework that produces omni-modal data for spoken interaction and fusion-dependent cross-modal reasoning. Through a comprehensive evaluation of over 30 state-of-the-art baselines, spanning MLLMs, modality-specific models, unified understanding-generation models, and omni-modal language models, FysicsWorld exposes the performance disparities and limitations across models in understanding, generation, and reasoning. Our benchmark establishes a unified foundation and strong baselines for evaluating and advancing next-generation full-modality architectures.

---

### An End-to-End Approach for Microgrid Probabilistic Forecasting and Robust Operation via Decision-focused Learning

Authors: Tingwei Cao, Yan Xu
Published: 2025-12-14
Categories: eess.SY, cs.LG
arXiv: https://arxiv.org/abs/2512.12755v2

High penetration of renewable energy sources (RES) introduces significant uncertainty and intermittency into microgrid operations, posing challenges to economic and reliable scheduling. To address this, this paper proposes an end-to-end decision-focused framework that jointly optimizes probabilistic forecasting and robust operation for microgrids. A multilayer encoder-decoder (MED) probabilistic forecasting model is integrated with a two-stage robust optimization (TSRO) model involving direct load control (DLC) through a differentiable decision pathway, enabling gradient-based feedback from operational outcomes to improve forecasting performance. Unlike conventional sequential approaches, the proposed method aligns forecasting accuracy with operational objectives by directly minimizing decision regret via a surrogate smart predict-then-optimize (SPO) loss function. This integration ensures that probabilistic forecasts are optimized for downstream decisions, enhancing both economic efficiency and robustness. Case studies on modified IEEE 33-bus and 69-bus systems demonstrate that the proposed framework achieves superior forecasting accuracy and operational performance, reducing total and net operation costs by up to 18% compared with conventional forecasting and optimization combinations. The results verify the effectiveness and scalability of the end-to-end decision-focused approach for resilient and cost-efficient microgrid management under uncertainty.

---

### An End-to-End Approach for Microgrid Probabilistic Forecasting and Robust Operation via Decision-focused Learning

Authors: Tingwei Cao, Yan Xu
Published: 2025-12-14
Categories: eess.SY, cs.LG
arXiv: https://arxiv.org/abs/2512.12755v1

High penetration of renewable energy sources (RES) introduces significant uncertainty and intermittency into microgrid operations, posing challenges to economic and reliable scheduling. To address this, this paper proposes an end-to-end decision-focused framework that jointly optimizes probabilistic forecasting and robust operation for microgrids. A multilayer encoder-decoder (MED) probabilistic forecasting model is integrated with a two-stage robust optimization (TSRO) model involving direct load control (DLC) through a differentiable decision pathway, enabling gradient-based feedback from operational outcomes to improve forecasting performance. Unlike conventional sequential approaches, the proposed method aligns forecasting accuracy with operational objectives by directly minimizing decision regret via a surrogate smart predict-then-optimize (SPO) loss function. This integration ensures that probabilistic forecasts are optimized for downstream decisions, enhancing both economic efficiency and robustness. Case studies on modified IEEE 33-bus and 69-bus systems demonstrate that the proposed framework achieves superior forecasting accuracy and operational performance, reducing total and net operation costs by up to 18% compared with conventional forecasting and optimization combinations. The results verify the effectiveness and scalability of the end-to-end decision-focused approach for resilient and cost-efficient microgrid management under uncertainty.

---

### GenieDrive: Towards Physics-Aware Driving World Model with 4D Occupancy Guided Video Generation

Authors: Zhenya Yang, Zhe Liu, Yuxiang Lu, Liping Hou, Chenxuan Miao, Siyi Peng, Bailan Feng, Xiang Bai, Hengshuang Zhao
Published: 2025-12-14
Categories: cs.CV
arXiv: https://arxiv.org/abs/2512.12751v1

Physics-aware driving world model is essential for drive planning, out-of-distribution data synthesis, and closed-loop evaluation. However, existing methods often rely on a single diffusion model to directly map driving actions to videos, which makes learning difficult and leads to physically inconsistent outputs. To overcome these challenges, we propose GenieDrive, a novel framework designed for physics-aware driving video generation. Our approach starts by generating 4D occupancy, which serves as a physics-informed foundation for subsequent video generation. 4D occupancy contains rich physical information, including high-resolution 3D structures and dynamics. To facilitate effective compression of such high-resolution occupancy, we propose a VAE that encodes occupancy into a latent tri-plane representation, reducing the latent size to only 58% of that used in previous methods. We further introduce Mutual Control Attention (MCA) to accurately model the influence of control on occupancy evolution, and we jointly train the VAE and the subsequent prediction module in an end-to-end manner to maximize forecasting accuracy. Together, these designs yield a 7.2% improvement in forecasting mIoU at an inference speed of 41 FPS, while using only 3.47 M parameters. Additionally, a Normalized Multi-View Attention is introduced in the video generation model to generate multi-view driving videos with guidance from our 4D occupancy, significantly improving video quality with a 20.7% reduction in FVD. Experiments demonstrate that GenieDrive enables highly controllable, multi-view consistent, and physics-aware driving video generation.

---

### Residual-augmented flow matching operators for probabilistic partial differential equations

Authors: Sahil Bhola, Karthik Duraisamy
Published: 2025-12-14
Categories: stat.CO, cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2512.12749v3

Learning surrogate models for physical systems with latent uncertainty remains challenging in data-scarce regimes: deterministic neural operators fail to characterize uncertainty, while generative approaches require large ensembles of high-fidelity solution operator simulations and often sacrifice resolution generalizability. In this work, we propose a residual-augmented probabilistic operator learning framework that casts flow-matching-based generative modeling in infinite-dimensional function spaces while leveraging inexpensive low-fidelity solution operators as an inductive bias. Rather than learning the full high-fidelity stochastic solution operator directly, the proposed framework learns probabilistic residual operators that characterize the discrepancy between low- and high-fidelity solutions. By parameterizing the vector field in flow matching using neural operators conditioned on both the known system input and low-fidelity solution, the framework amortizes probabilistic inference across input conditions while enabling uncertainty-aware and resolution-generalizable predictions across spatial discretizations. Numerical experiments on stochastic advection, Burgers', and Darcy flow systems demonstrate that the residual-augmented formulation improves predictive accuracy under the same high-fidelity data budget, while the probabilistic operator learning formulation enables accurate characterization of uncertainty in low-data regimes compared to learning high-fidelity stochastic operators directly from data.

---

### Flow-matching Operators for Residual-Augmented Probabilistic Learning of Partial Differential Equations

Authors: Sahil Bhola, Karthik Duraisamy
Published: 2025-12-14
Categories: stat.CO, cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2512.12749v1

Learning probabilistic surrogates for PDEs remains challenging in data-scarce regimes: neural operators require large amounts of high-fidelity data, while generative approaches typically sacrifice resolution invariance. We formulate flow matching in an infinite-dimensional function space to learn a probabilistic transport that maps low-fidelity approximations to the manifold of high-fidelity PDE solutions via learned residual corrections. We develop a conditional neural operator architecture based on feature-wise linear modulation for flow-matching vector fields directly in function space, enabling inference at arbitrary spatial resolutions without retraining. To improve stability and representational control of the induced neural ODE, we parameterize the flow vector field as a sum of a linear operator and a nonlinear operator, combining lightweight linear components with a conditioned Fourier neural operator for expressive, input-dependent dynamics. We then formulate a residual-augmented learning strategy where the flow model learns probabilistic corrections from inexpensive low-fidelity surrogates to high-fidelity solutions, rather than learning the full solution mapping from scratch. Finally, we derive tractable training objectives that extend conditional flow matching to the operator setting with input-function-dependent couplings. To demonstrate the effectiveness of our approach, we present numerical experiments on a range of PDEs, including the 1D advection and Burgers' equation, and a 2D Darcy flow problem for flow through a porous medium. We show that the proposed method can accurately learn solution operators across different resolutions and fidelities and produces uncertainty estimates that appropriately reflect model confidence, even when trained on limited high-fidelity data.

---