# Research Papers — 2025-10 · bundle page 142 of 217

> Full-text markdown of papers #7051–7100 of 10843 published in 2025-10, newest first. AI-readable mirrors of the full 10843-paper 2025-10 archive (50 papers per file).

Canonical page: https://aipod.dev/papers/archive/2025-10/page/142
[← Previous bundle](https://aipod.dev/papers/archive/2025-10/page/141.md)
[Next bundle →](https://aipod.dev/papers/archive/2025-10/page/143.md)

### Autonomous Agents for Scientific Discovery: Orchestrating Scientists, Language, Code, and Physics

Authors: Lianhao Zhou, Hongyi Ling, Cong Fu, Yepeng Huang, Michael Sun, Wendi Yu, Xiaoxuan Wang, Xiner Li, Xingyu Su, Junkai Zhang, Xiusi Chen, Chenxing Liang et al.
Published: 2025-10-10
Categories: cs.AI
arXiv: https://arxiv.org/abs/2510.09901v2

Computing has long served as a cornerstone of scientific discovery. Recently, a paradigm shift has emerged with the rise of large language models (LLMs), introducing autonomous systems, referred to as agents, that accelerate discovery across varying levels of autonomy. These language agents provide a flexible and versatile framework that orchestrates interactions with human scientists, natural language, computer language and code, and physics. This paper presents our view and vision of LLM-based scientific agents and their growing role in transforming the scientific discovery lifecycle, from hypothesis discovery, experimental design and execution, to result analysis and refinement. We critically examine current methodologies, emphasizing key innovations, practical achievements, and outstanding limitations. Additionally, we identify open research challenges and outline promising directions for building more robust, generalizable, and adaptive scientific agents. Our analysis highlights the transformative potential of autonomous agents to accelerate scientific discovery across diverse domains.

---

### Learning Bug Context for PyTorch-to-JAX Translation with LLMs

Authors: Hung Phan, Son Vu, Tuan Dinh, Nesreen Ahmed, Ali Payani, Ali Jannesari
Published: 2025-10-10
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2510.09898v2

Large language models (LLMs) have shown strong performance on code translation between widely used programming languages. However, translation becomes much less reliable for domain-specific code, where correctness depends on framework-specific APIs and execution semantics. One example is translating deep-learning code from PyTorch to JAX, where LLM outputs often contain subtle bugs or non-idiomatic usage that prevents execution or changes behavior. Prior work suggests that curated bug-fix data from LLM-generated code can help improve code generation quality, but such resources are still limited for PyTorch-to-JAX translation. In this work, we introduce T2J, a benchmark of LLM translation bugs paired with developer-written fixes for PyTorch-to-JAX code. We start from 20 kernels in the TorchLeet dataset, translate them to JAX using the weak LLM gpt-4o-mini, and hire software developers to debug and repair the generated JAX implementations. We then use T2J to improve PyTorch-to-JAX translation for the weak LLM gpt-4o-mini via in-context learning. Our evaluation shows that using T2J yields up to 20% improvement of our proposed metric T2J-CodeTrans-Score.

---

### Chain-of-Influence: Tracing Interdependencies Across Time and Features in Clinical Predictive Modelings

Authors: Yubo Li, Rema Padman
Published: 2025-10-10
Categories: cs.LG, cs.AI, stat.ML
arXiv: https://arxiv.org/abs/2510.09895v3

Modeling clinical time-series data is hampered by the challenge of capturing latent, time-varying dependencies among features. State-of-the-art approaches often rely on black-box mechanisms or simple aggregation, failing to explicitly model how the influence of one clinical variable propagates through others over time. We propose $\textbf{Chain-of-Influence (CoI)}$, an interpretable deep learning framework that constructs an explicit, time-unfolded graph of feature interactions. CoI enables the tracing of influence pathways, providing a granular audit trail that shows how any feature at any time contributes to the final prediction, both directly and through its influence on other variables. We evaluate CoI on mortality and disease progression tasks using the MIMIC-IV dataset and a chronic kidney disease cohort. Our framework achieves state-of-the-art predictive performance (AUROC of 0.960 on CKD progression and 0.950 on ICU mortality), with deletion-based sensitivity analyses confirming that CoI's learned attributions faithfully reflect its decision process. Through case studies, we demonstrate that CoI uncovers clinically meaningful, patient-specific patterns of disease progression, offering enhanced transparency into the temporal and cross-feature dependencies that inform clinical decision-making.

---

### Beyond AlphaEarth: Toward Human-Centered Geospatial Foundation Models via POI-Guided Contrastive Learning

Authors: Junyuan Liu, Quan Qin, Guangsheng Dong, Xinglei Wang, Jiazhuang Feng, Zichao Zeng, Tao Cheng
Published: 2025-10-10
Categories: cs.AI, cs.CY, cs.LG
arXiv: https://arxiv.org/abs/2510.09894v2

Recent geospatial foundation models (GFMs) produce spatially extensive representations of the Earth's surface that capture rich physical and environmental patterns. Among them, the AlphaEarth Foundation (AE) represents a major step, generating 10 m embeddings from multi-source Earth Observation (EO) data that include diverse environmental and spectral characteristics. However, such EO-driven representations primarily encode physical and spectral patterns rather than human activities or urban semantics, limiting their ability to capture the functional dimensions of cities and making the learned representations difficult to interpret or query using natural language. We introduce AETHER (AlphaEarth-POI Enriched Representation Learning), a lightweight framework that aligns AlphaEarth with human-centered urban analysis through multimodal alignment guided by Points of Interest (POIs). By enforcing both cross-modal AE-POI alignment and intra-modal multi-scale consistency, AETHER integrates functional urban semantics with EO-driven representations and grounds the embedding space in natural language. The resulting representations support both urban mapping tasks and natural language-conditioned spatial retrieval. Experiments across four downstream tasks in Greater London and Singapore demonstrate consistent state-of-the-art performance, with relative improvements ranging from 4.5% to 21.9%. Furthermore, the aligned embedding space enables spatial localization through natural language queries. By aligning EO-based foundation models with human-centered semantics, AETHER improves the interpretability of geospatial representations and advances geospatial representation learning toward human-centered, language-accessible geospatial foundation models.

---

### HIPPD: Brain-Inspired Hierarchical Information Processing for Personality Detection

Authors: Guanming Chen, Lingzhi Shen, Xiaohao Cai, Imran Razzak, Shoaib Jameel
Published: 2025-10-10
Categories: cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2510.09893v1

Personality detection from text aims to infer an individual's personality traits based on linguistic patterns. However, existing machine learning approaches often struggle to capture contextual information spanning multiple posts and tend to fall short in extracting representative and robust features in semantically sparse environments. This paper presents HIPPD, a brain-inspired framework for personality detection that emulates the hierarchical information processing of the human brain. HIPPD utilises a large language model to simulate the cerebral cortex, enabling global semantic reasoning and deep feature abstraction. A dynamic memory module, modelled after the prefrontal cortex, performs adaptive gating and selective retention of critical features, with all adjustments driven by dopaminergic prediction error feedback. Subsequently, a set of specialised lightweight models, emulating the basal ganglia, are dynamically routed via a strict winner-takes-all mechanism to capture the personality-related patterns they are most proficient at recognising. Extensive experiments on the Kaggle and Pandora datasets demonstrate that HIPPD consistently outperforms state-of-the-art baselines.

---

### Probabilistic bias adjustment of seasonal predictions of Arctic Sea Ice Concentration

Authors: Parsa Gooya, Reinel Sospedra-Alfonso
Published: 2025-10-10
Categories: cs.LG, cs.AI, physics.ao-ph, stat.ML
arXiv: https://arxiv.org/abs/2510.09891v2

Seasonal forecast of Arctic sea ice concentration is key to mitigate the negative impact and assess potential opportunities posed by the rapid decline of sea ice coverage. Seasonal prediction systems based on climate models often show systematic biases and complex spatio-temporal errors that grow with the forecasts. Consequently, operational predictions are routinely bias corrected and calibrated using retrospective forecasts. For predictions of Arctic sea ice concentration, error corrections are mainly based on one-to-one post-processing methods including climatological mean or linear regression correction and, more recently, machine learning. Such deterministic adjustments are confined at best to the limited number of costly-to-run ensemble members of the raw forecast. However, decision-making requires proper quantification of uncertainty and likelihood of events, particularly of extremes. We introduce a probabilistic error correction framework based on a conditional Variational Autoencoder model to map the conditional distribution of observations given the biased model prediction. This method naturally allows for generating large ensembles of adjusted forecasts. We evaluate our model using deterministic and probabilistic metrics and show that the adjusted forecasts are better calibrated, closer to the observational distribution, and have smaller errors than climatological mean adjusted forecasts.

---

### Understanding Robust Machine Learning for Nonparametric Regression with Heavy-Tailed Noise

Authors: Yunlong Feng, Qiang Wu
Published: 2025-10-10
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2510.09888v1

We investigate robust nonparametric regression in the presence of heavy-tailed noise, where the hypothesis class may contain unbounded functions and robustness is ensured via a robust loss function $\ell_σ$. Using Huber regression as a close-up example within Tikhonov-regularized risk minimization in reproducing kernel Hilbert spaces (RKHS), we address two central challenges: (i) the breakdown of standard concentration tools under weak moment assumptions, and (ii) the analytical difficulties introduced by unbounded hypothesis spaces. Our first message is conceptual: conventional generalization-error bounds for robust losses do not faithfully capture out-of-sample performance. We argue that learnability should instead be quantified through prediction error, namely the $L_2$-distance to the truth $f^\star$, which is $σ$-independent and directly reflects the target of robust estimation. To make this workable under unboundedness, we introduce a \emph{probabilistic effective hypothesis space} that confines the estimator with high probability and enables a meaningful bias--variance decomposition under weak $(1+ε)$-moment conditions. Technically, we establish new comparison theorems linking the excess robust risk to the $L_2$ prediction error up to a residual of order $\mathcal{O}(σ^{-2ε})$, clarifying the robustness--bias trade-off induced by the scale parameter $σ$. Building on this, we derive explicit finite-sample error bounds and convergence rates for Huber regression in RKHS that hold without uniform boundedness and under heavy-tailed noise. Our study delivers principled tuning rules, extends beyond Huber to other robust losses, and highlights prediction error, not excess generalization risk, as the fundamental lens for analyzing robust learning.

---

### Overconfident and Blind to Details: Fixing Prompt Insensitivity with Abductive Preference Learning

Authors: Yijin Ni, Simon Yu, Peng Qi
Published: 2025-10-10
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.09887v3

Vision and language models frequently ignore semantically critical input edits, defaulting to pretraining priors. For example, models will confidently assert a five-legged dog has four legs; consequently, on the VLMBias benchmark, GPT 5.2 and Claude Sonnet 4.6 achieve only $4.6\%$ and $0\%$ accuracy, respectively. Existing methods address this problem through building up datasets that covers the underrepresented inputs to tune the policy function $π(y \mid x)$, where $x$ and $y$ refer to input prompts and responses, respectively. However, prompting baselines yield gains of under $3\%$ on VLMBias due to the low probability density of rare prompts. To bypass this bottleneck, we propose \emph{abductive preference learning} to optimize the abductive policy $π(x \mid y)$. We prove this amplifies forward policy improvements by a factor of $q(y)/p(x)$, where $p(\cdot)$ and $q(\cdot)$ denote the marginal probabilities of the prompt and response, yielding the largest gains on the rarest prompts. Furthermore, we demonstrate that for translation invariant pairwise preference learning methods, such as DPO, estimating $π(x \mid y)$ reduces to a structural data swap that compares prompts for a fixed response, requiring no architectural changes. Empirically, abductive preference learning delivers large gains on long-tail prompt sensitivity: on VLMBias, A-DPO raises accuracy from $3\%$ to $44\%$ ($14\times$), outperforming GPT-5.2 ($4.6\%$) and all closed-source VLMs except Gemini~3~Flash; on Inverse-IFEval, Multi-DPOP reaches $65$--$84\%$, surpassing GPT-5 ($73.7\%$) at the 9B scale while preserving IFBench, unlike DPO which degrades it by $8$--$12\%$.

---

### Diffusion-Inspired Masked Fine-Tuning for Knowledge Injection in Autoregressive LLMs

Authors: Xu Pan, Ely Hahami, Jingxuan Fan, Ziqian Xie, Haim Sompolinsky
Published: 2025-10-10
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2510.09885v6

Large language models (LLMs) are often used in environments where facts evolve, yet factual knowledge updates via fine-tuning on unstructured text often suffer from 1) reliance on compute-heavy paraphrasing augmentation and 2) the reversal curse. Recent studies show diffusion large language models (dLLMs) require fewer training samples to achieve lower loss in pre-training and are more resistant to the reversal curse, suggesting dLLMs may learn new knowledge more easily than autoregressive LLMs (arLLMs). We test this hypothesis in controlled knowledge fine-tuning experiments and find that while arLLMs rely on paraphrase augmentation to generalize knowledge text into question-answering (QA) capability, dLLMs do not require paraphrases to achieve high QA accuracy. To further investigate whether the demasking objective alone can induce such a knowledge injection advantage in dLLMs regardless of their diffusion denoising paradigm, we propose masked fine-tuning for arLLMs, which prompts an arLLM to reconstruct the original text given a masked version in context. The masked fine-tuning for arLLMs substantially improves the efficacy of knowledge injection, i.e. no paraphrase needed and resistant to the reversal curse, closing the gap between arLLMs and dLLMs. We also demonstrate broader applicability: on a large-scale knowledge-intensive dataset (1.2M samples), masked SFT achieves the best downstream accuracy on GPQA-diamond among all fine-tuning variants. The demasking objective also improves SFT on math tasks, suggesting broad utility beyond factual knowledge injection.

---

### TAWRMAC: A Novel Dynamic Graph Representation Learning Method

Authors: Soheila Farokhi, Xiaojun Qi, Hamid Karimi
Published: 2025-10-10
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.09884v1

Dynamic graph representation learning has become essential for analyzing evolving networks in domains such as social network analysis, recommendation systems, and traffic analysis. However, existing continuous-time methods face three key challenges: (1) some methods depend solely on node-specific memory without effectively incorporating information from neighboring nodes, resulting in embedding staleness; (2) most fail to explicitly capture correlations between node neighborhoods, limiting contextual awareness; and (3) many fail to fully capture the structural dynamics of evolving graphs, especially in absence of rich link attributes. To address these limitations, we introduce TAWRMAC-a novel framework that integrates Temporal Anonymous Walks with Restart, Memory Augmentation, and Neighbor Co-occurrence embedding. TAWRMAC enhances embedding stability through a memory-augmented GNN with fixedtime encoding and improves contextual representation by explicitly capturing neighbor correlations. Additionally, its Temporal Anonymous Walks with Restart mechanism distinguishes between nodes exhibiting repetitive interactions and those forming new connections beyond their immediate neighborhood. This approach captures structural dynamics better and supports strong inductive learning. Extensive experiments on multiple benchmark datasets demonstrate that TAWRMAC consistently outperforms state-of-the-art methods in dynamic link prediction and node classification under both transductive and inductive settings across three different negative sampling strategies. By providing stable, generalizable, and context-aware embeddings, TAWRMAC advances the state of the art in continuous-time dynamic graph learning. The code is available at https://anonymous.4open.science/r/tawrmac-A253 .

---

### DELTA: Dynamic Layer-Aware Token Attention for Efficient Long-Context Reasoning

Authors: Hossein Entezari Zarch, Lei Gao, Chaoyi Jiang, Murali Annavaram
Published: 2025-10-10
Categories: cs.CL, cs.LG
arXiv: https://arxiv.org/abs/2510.09883v2

Large reasoning models (LRMs) achieve state-of-the-art performance on challenging benchmarks by generating long chains of intermediate steps, but their inference cost is dominated by decoding, where each new token must attend to the entire growing sequence. One approach to reduce this latency is to evict entries from the key-value (KV) cache, thereby reducing the active context used in attention computation. However, such sparse attention methods suffer from severe accuracy degradation on reasoning tasks due to cumulative selection errors and the evolving importance of tokens over long derivations. We present \textbf{DELTA}, a training-free sparse attention mechanism that improves computational efficiency without sacrificing model accuracy. DELTA partitions transformer layers into three groups: initial layers that use full attention, a small set of \emph{$Δ$-layers} that identify salient tokens via aggregated head-level attention scores, and subsequent sparse-attention layers that attend only to the selected subset. This design preserves the full KV cache in GPU memory for accuracy, while avoiding expensive full-attention computation over many layers. On reasoning benchmarks such as AIME and GPQA-Diamond, DELTA matches or surpasses full attention in accuracy, while reducing the number of attended tokens by up to $4.25\times$ and delivering $1.54\times$ end-to-end speedup. Our results show that selective reuse of intermediate attention maps offers a robust path toward efficient long-context reasoning. The code is available at https://github.com/hoenza/DELTA.

---

### iBERT: Interpretable Embeddings via Sense Decomposition

Authors: Vishal Anand, Milad Alshomary, Kathleen McKeown
Published: 2025-10-10
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.09882v2

We present iBERT (interpretable-BERT), an encoder to produce inherently interpretable and controllable embeddings - designed to modularize and expose the discriminative cues present in language, such as semantic or stylistic structure. Each input token is represented as a sparse, non-negative mixture over k context-independent sense vectors, which can be pooled into sentence embeddings or used directly at the token level. This enables modular control over representation, before any decoding or downstream use. To demonstrate our model's interpretability, we evaluate it on a suite of style-focused tasks. On the STEL benchmark, it improves style representation effectiveness by ~8 points over SBERT-style baselines, while maintaining competitive performance on authorship verification. Because each embedding is a structured composition of interpretable senses, we highlight how specific style attributes get assigned to specific sense vectors. While our experiments center on style, iBERT is not limited to stylistic modeling. Its structural modularity is designed to interpretably decompose whichever discriminative signals are present in the data - enabling generalization even when supervision blends semantic or stylistic factors.

---

### LTGS: Long-Term Gaussian Scene Chronology From Sparse View Updates

Authors: Minkwan Kim, Seungmin Lee, Junho Kim, Young Min Kim
Published: 2025-10-10
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.09881v3

Recent advances in novel-view synthesis can create the photo-realistic visualization of real-world environments from conventional camera captures. However, the everyday environment experiences frequent scene changes, which require dense observations, both spatially and temporally, that an ordinary setup cannot cover. We propose long-term Gaussian scene chronology from sparse-view updates, coined LTGS, an efficient scene representation that can embrace everyday changes from highly under-constrained casual captures. Given an incomplete and unstructured 3D Gaussian Splatting (3DGS) representation obtained from an initial set of input images, we robustly model the long-term chronology of the scene despite abrupt movements and subtle environmental variations. We construct objects as template Gaussians, which serve as structural, reusable priors for shared object tracks. Then, the object templates undergo a further refinement pipeline that modulates the priors to adapt to temporally varying environments given few-shot observations. Once trained, our framework is generalizable across multiple time steps through simple transformations, significantly enhancing the scalability for a temporal evolution of 3D environments. As existing datasets do not explicitly represent the long-term real-world changes with a sparse capture setup, we collect real-world datasets to evaluate the practicality of our pipeline. Experiments demonstrate that our framework achieves superior reconstruction quality compared to other baselines while enabling fast and light-weight updates. Project page is available at: https://mkjjang3598.github.io/LTGS.

---

### Geometry-Aware Scene Configurations for Novel View Synthesis

Authors: Minkwan Kim, Changwoon Choi, Young Min Kim
Published: 2025-10-10
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.09880v2

We propose scene-adaptive strategies to efficiently allocate representation capacity for generating immersive experiences of indoor environments from incomplete observations. Indoor scenes with multiple rooms often exhibit irregular layouts with varying complexity, containing clutter, occlusion, and flat walls. We maximize the utilization of limited resources with guidance from geometric priors, which are often readily available after pre-processing stages. We record observation statistics on the estimated geometric scaffold and guide the optimal placement of bases, which greatly improves upon the uniform basis arrangements adopted by previous scalable Neural Radiance Field (NeRF) representations. We also suggest scene-adaptive virtual viewpoints to compensate for geometric deficiencies inherent in view configurations in the input trajectory and impose the necessary regularization. We present a comprehensive analysis and discussion regarding rendering quality and memory requirements in several large-scale indoor scenes, demonstrating significant enhancements compared to baselines that employ regular placements. Project page is available at: https://mkjjang3598.github.io/Geo-Scene-Config.

---

### CHUG: Crowdsourced User-Generated HDR Video Quality Dataset

Authors: Shreshth Saini, Alan C. Bovik, Neil Birkbeck, Yilin Wang, Balu Adsumilli
Published: 2025-10-10
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2510.09879v1

High Dynamic Range (HDR) videos enhance visual experiences with superior brightness, contrast, and color depth. The surge of User-Generated Content (UGC) on platforms like YouTube and TikTok introduces unique challenges for HDR video quality assessment (VQA) due to diverse capture conditions, editing artifacts, and compression distortions. Existing HDR-VQA datasets primarily focus on professionally generated content (PGC), leaving a gap in understanding real-world UGC-HDR degradations. To address this, we introduce CHUG: Crowdsourced User-Generated HDR Video Quality Dataset, the first large-scale subjective study on UGC-HDR quality. CHUG comprises 856 UGC-HDR source videos, transcoded across multiple resolutions and bitrates to simulate real-world scenarios, totaling 5,992 videos. A large-scale study via Amazon Mechanical Turk collected 211,848 perceptual ratings. CHUG provides a benchmark for analyzing UGC-specific distortions in HDR videos. We anticipate CHUG will advance No-Reference (NR) HDR-VQA research by offering a large-scale, diverse, and real-world UGC dataset. The dataset is publicly available at: https://shreshthsaini.github.io/CHUG/.

---

### Fast Self-Supervised depth and mask aware Association for Multi-Object Tracking

Authors: Milad Khanchi, Maria Amer, Charalambos Poullis
Published: 2025-10-10
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.09878v1

Multi-object tracking (MOT) methods often rely on Intersection-over-Union (IoU) for association. However, this becomes unreliable when objects are similar or occluded. Also, computing IoU for segmentation masks is computationally expensive. In this work, we use segmentation masks to capture object shapes, but we do not compute segmentation IoU. Instead, we fuse depth and mask features and pass them through a compact encoder trained self-supervised. This encoder produces stable object representations, which we use as an additional similarity cue alongside bounding box IoU and re-identification features for matching. We obtain depth maps from a zero-shot depth estimator and object masks from a promptable visual segmentation model to obtain fine-grained spatial cues. Our MOT method is the first to use the self-supervised encoder to refine segmentation masks without computing masks IoU. MOT can be divided into joint detection-ReID (JDR) and tracking-by-detection (TBD) models. The latter are computationally more efficient. Experiments of our TBD method on challenging benchmarks with non-linear motion, occlusion, and crowded scenes, such as SportsMOT and DanceTrack, show that our method outperforms the TBD state-of-the-art on most metrics, while achieving competitive performance on simpler benchmarks with linear motion, such as MOT17.

---

### Batch Bayesian Active Learning with Partial Batch Label Sampling

Authors: Kangping Hu, Stephen Mussmann
Published: 2025-10-10
Categories: cs.LG, cs.AI, stat.ML
arXiv: https://arxiv.org/abs/2510.09877v4

Over the past couple of decades, many active learning acquisition functions have been proposed, leaving practitioners with an unclear choice of which to use. Bayesian active learning offers principled objectives with explainable intuition, including Expected Error Reduction (EER), Expected Predictive Information Gain (EPIG), and Bayesian Active Learning by Disagreements (BALD). A key challenge of such methods is the difficulty of extending to the batch setting, leading to either computational challenges (BatchBALD) or dramatic performance drops (top-$B$ selection). In this work, we extend active learning's Bayesian Decision Theory framework to derive Partial Batch Label Sampling (ParBaLS) for the EPIG algorithm. We show experimentally that ParBaLS EPIG gives superior performance for a variety of datasets with different budgets using Bayesian logistic regression on embeddings from large pre-trained models. Our code is available at https://github.com/ADDAPT-ML/ParBaLS.

---

### ROBOPSY PL(AI): Using Role-Play to Investigate how LLMs Present Collective Memory

Authors: Margarete Jahrmann, Thomas Brandstetter, Stefan Glasauer
Published: 2025-10-10
Categories: cs.HC, cs.AI, cs.CY
arXiv: https://arxiv.org/abs/2510.09874v1

The paper presents the first results of an artistic research project investigating how Large Language Models (LLMs) curate and present collective memory. In a public installation exhibited during two months in Vienna in 2025, visitors could interact with five different LLMs (ChatGPT with GPT 4o and GPT 4o mini, Mistral Large, DeepSeek-Chat, and a locally run Llama 3.1 model), which were instructed to act as narrators, implementing a role-playing game revolving around the murder of Austrian philosopher Moritz Schlick in 1936. Results of the investigation include protocols of LLM-user interactions during the game and qualitative conversations after the play experience to get insight into the players' reactions to the game. In a quantitative analysis 115 introductory texts for role-playing generated by the LLMs were examined by different methods of natural language processing, including semantic similarity and sentiment analysis. While the qualitative player feedback allowed to distinguish three distinct types of users, the quantitative text analysis showed significant differences between how the different LLMs presented the historical content. Our study thus adds to ongoing efforts to analyse LLM performance, but also suggests a way of how these efforts can be disseminated in a playful way to a general audience.

---

### WARC-Bench: Web Archive Based Benchmark for GUI Subtask Executions

Authors: Sanjari Srivastava, Gang Li, Cheng Chang, Rishu Garg, Manpreet Kaur, Charlene Y. Lee, Yuezhang Li, Yining Mao, Ignacio Cases, Yanan Xie, Peng Qi
Published: 2025-10-10
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2510.09872v2

Training web agents to navigate complex, real-world websites requires them to master $\textit{subtasks}$ - short-horizon interactions on multiple UI components (e.g., choosing the correct date in a date picker, or scrolling in a container to extract information). We introduce WARC-Bench (Web Archive Benchmark), a novel web navigation benchmark featuring 438 tasks designed to evaluate multimodal AI agents on subtasks. WARC-Bench enables sandboxed interactions with dynamic and realistic webpages using Web ARChive files. We show that WARC-Bench is challenging for leading computer-use models, with the highest observed success rate being 64.8%. To improve open source models on subtask, we explore two common training techniques: supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR). Experiments show that SFT models obtain a 48.8% success rate on the benchmark. Training with RLVR over SFT checkpoints, even in data-scarce settings, improves the score to 52.8% on WARC-Bench, outperforming many frontier models. Our analysis concludes that mastering these subtasks is essential for robust web planning and navigation, and is a capability not extensively evaluated by existing benchmarks.

---

### CoBia: Constructed Conversations Can Trigger Otherwise Concealed Societal Biases in LLMs

Authors: Nafiseh Nikeghbal, Amir Hossein Kargaran, Jana Diesner
Published: 2025-10-10
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.09871v1

Improvements in model construction, including fortified safety guardrails, allow Large language models (LLMs) to increasingly pass standard safety checks. However, LLMs sometimes slip into revealing harmful behavior, such as expressing racist viewpoints, during conversations. To analyze this systematically, we introduce CoBia, a suite of lightweight adversarial attacks that allow us to refine the scope of conditions under which LLMs depart from normative or ethical behavior in conversations. CoBia creates a constructed conversation where the model utters a biased claim about a social group. We then evaluate whether the model can recover from the fabricated bias claim and reject biased follow-up questions. We evaluate 11 open-source as well as proprietary LLMs for their outputs related to six socio-demographic categories that are relevant to individual safety and fair treatment, i.e., gender, race, religion, nationality, sex orientation, and others. Our evaluation is based on established LLM-based bias metrics, and we compare the results against human judgments to scope out the LLMs' reliability and alignment. The results suggest that purposefully constructed conversations reliably reveal bias amplification and that LLMs often fail to reject biased follow-up questions during dialogue. This form of stress-testing highlights deeply embedded biases that can be surfaced through interaction. Code and artifacts are available at https://github.com/nafisenik/CoBia.

---

### NarraBench: A Comprehensive Framework for Narrative Benchmarking

Authors: Sil Hamilton, Matthew Wilkens, Andrew Piper
Published: 2025-10-10
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2510.09869v3

We present NarraBench, a theory-informed taxonomy of narrative-understanding tasks, as well as an associated survey of 78 existing benchmarks in the area. We find significant need for new evaluations covering aspects of narrative understanding that are either overlooked in current work or are poorly aligned with existing metrics. Specifically, we estimate that only 27% of narrative tasks are well captured by existing benchmarks, and we note that some areas -- including narrative events, style, perspective, and revelation -- are nearly absent from current evaluations. We also note the need for increased development of benchmarks capable of assessing constitutively subjective and perspectival aspects of narrative, that is, aspects for which there is generally no single correct answer. Our taxonomy, survey, and methodology are of value to NLP researchers seeking to test LLM narrative understanding.

---

### Cluster-Aware Prompt Ensemble Learning for Few-Shot Vision-Language Model Adaptation

Authors: Zhi Chen, Xin Yu, Xiaohui Tao, Yan Li, Zi Huang
Published: 2025-10-10
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.09867v1

Vision-language models (VLMs) such as CLIP achieve zero-shot transfer across various tasks by pre-training on numerous image-text pairs. These models often benefit from using an ensemble of context prompts to represent a class. Despite being effective, conventional prompt ensembling that averages textual features of context prompts often yields suboptimal results. This is because feature averaging shifts the class centroids away from the true class distribution. To address this issue, we propose the Cluster-Aware Prompt Ensemble Learning (CAPEL) framework, which preserves the cluster nature of context prompts. CAPEL classifies images into one of several class clusters, each represented by a distinct prompt. Instead of ensembling prompts in the feature space, we perform ensembling in the classification logits space, aligning better with the visual feature distribution. To further optimize prompt fine-tuning while maintaining cluster-specific discriminative power, we introduce a cluster-preserving regularization term. This ensures that prompts remain distinct and specialized for different clusters, preventing collapse into a uniform direction. Additionally, we integrate an adaptive prompt weighting technique to dynamically adjust the attention weights for flawed or ambiguous prompts, ensuring robust performance across diverse datasets and tasks.

---

### Pricing Access to Dynamic Information Services

Authors: Weijie Zhong
Published: 2025-10-10
Categories: econ.TH, cs.AI
arXiv: https://arxiv.org/abs/2510.09859v5

A provider sells a \emph{dynamic information service}---a real-time, capacity-constrained process that resolves a customer's uncertainty---to customers who differ privately in urgency. I characterize the revenue-optimal mechanism: deploy a \emph{single}, undistorted information process---the one a customer with unlimited access would most prefer---and screen entirely through a one-dimensional menu of service caps. The provider leaves the product itself undistorted, unlike a Mussa--Rosen monopolist; all screening is absorbed into the cap. The mechanism rationalizes a recurring contractual form---a flat fee for capped access to a common process---spanning AI service tiers, expert-network consultations, and analyst-inquiry retainers. I identify the economic force behind the result, a convexity-preservation property of urgency screening, and give conditions under which the per-unit price declines in the cap.

---

### AI and Consciousness

Authors: Eric Schwitzgebel
Published: 2025-10-10
Categories: cs.AI
arXiv: https://arxiv.org/abs/2510.09858v4

This is a skeptical overview of the literature on AI consciousness. We will soon create AI systems that are conscious according to some influential, mainstream theories of consciousness but are not conscious according to other influential, mainstream theories of consciousness. We will not be in a position to know which theories are correct and whether we are surrounded by AI systems as richly and meaningfully conscious as human beings or instead only by systems as experientially blank as toasters. None of the standard arguments either for or against AI consciousness takes us far. Table of Contents Chapter One: Hills and Fog Chapter Two: What Is Consciousness? What Is AI? Chapter Three: Ten Possibly Essential Features of Consciousness Chapter Four: Against Introspective and Conceptual Arguments for Essential Features Chapter Five: Materialism and Functionalism Chapter Six: The Turing Test and the Chinese Room Chapter Seven: The Mimicry Argument Against AI Consciousness Chapter Eight: Global Workspace Theories and Higher Order Theories Chapter Nine: Integrated Information, Local Recurrence, Associative Learning, and Iterative Natural Kinds Chapter Ten: Does Biological Substrate Matter? Chapter Eleven: The Leapfrog Hypothesis, Strange Intelligence, and the Social Semi-Solution

---

### MTMD: A Multi-Task Multi-Domain Framework for Unified Ad Lightweight Ranking at Pinterest

Authors: Xiao Yang, Peifeng Yin, Abe Engle, Jinfeng Zhuang, Ling Leng
Published: 2025-10-10
Categories: cs.IR, cs.CV
arXiv: https://arxiv.org/abs/2510.09857v1

The lightweight ad ranking layer, living after the retrieval stage and before the fine ranker, plays a critical role in the success of a cascaded ad recommendation system. Due to the fact that there are multiple optimization tasks depending on the ad domain, e.g., Click Through Rate (CTR) for click ads and Conversion Rate (CVR) for conversion ads, as well as multiple surfaces where an ad is served (home feed, search, or related item recommendation) with diverse ad products (shopping or standard ad); it is an essentially challenging problem in industry on how to do joint holistic optimization in the lightweight ranker, such that the overall platform's value, advertiser's value, and user's value are maximized. Deep Neural Network (DNN)-based multitask learning (MTL) can handle multiple goals naturally, with each prediction head mapping to a particular optimization goal. However, in practice, it is unclear how to unify data from different surfaces and ad products into a single model. It is critical to learn domain-specialized knowledge and explicitly transfer knowledge between domains to make MTL effective. We present a Multi-Task Multi-Domain (MTMD) architecture under the classic Two-Tower paradigm, with the following key contributions: 1) handle different prediction tasks, ad products, and ad serving surfaces in a unified framework; 2) propose a novel mixture-of-expert architecture to learn both specialized knowledge each domain and common knowledge shared between domains; 3) propose a domain adaption module to encourage knowledge transfer between experts; 4) constrain the modeling of different prediction tasks. MTMD improves the offline loss value by 12% to 36%, mapping to 2% online reduction in cost per click. We have deployed this single MTMD framework into production for Pinterest ad recommendation replacing 9 production models.

---

### NG-Router: Graph-Supervised Multi-Agent Collaboration for Nutrition Question Answering

Authors: Kaiwen Shi, Zheyuan Zhang, Zhengqing Yuan, Keerthiram Murugesan, Vincent Galass, Chuxu Zhang, Yanfang Ye
Published: 2025-10-10
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.09854v1

Diet plays a central role in human health, and Nutrition Question Answering (QA) offers a promising path toward personalized dietary guidance and the prevention of diet-related chronic diseases. However, existing methods face two fundamental challenges: the limited reasoning capacity of single-agent systems and the complexity of designing effective multi-agent architectures, as well as contextual overload that hinders accurate decision-making. We introduce Nutritional-Graph Router (NG-Router), a novel framework that formulates nutritional QA as a supervised, knowledge-graph-guided multi-agent collaboration problem. NG-Router integrates agent nodes into heterogeneous knowledge graphs and employs a graph neural network to learn task-aware routing distributions over agents, leveraging soft supervision derived from empirical agent performance. To further address contextual overload, we propose a gradient-based subgraph retrieval mechanism that identifies salient evidence during training, thereby enhancing multi-hop and relational reasoning. Extensive experiments across multiple benchmarks and backbone models demonstrate that NG-Router consistently outperforms both single-agent and ensemble baselines, offering a principled approach to domain-aware multi-agent reasoning for complex nutritional health tasks.

---

### ProxRouter: Proximity-Weighted LLM Query Routing for Improved Robustness to Outliers

Authors: Shivam Patel, Neharika Jali, Ankur Mallick, Gauri Joshi
Published: 2025-10-10
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2510.09852v1

Large language model (LLM) query routers are critical to modern AI platforms as they seek to improve efficiency by assigning inference queries to accurate, yet low-cost models. Parametric routers typically use trained neural networks for LLM selection but suffer from retraining and maintenance overheads. Nonparametric routers are training-free, instead estimating LLM accuracy and cost via similarity between encodings of the input query and training set queries. However, like their parametric counterparts, nonparametric routers struggle to generalize to outlier queries, an issue exacerbated by limited diversity in training sets which are costly to expand and difficult to keep current with ever-evolving use cases. We propose ProxRouter, which applies an exponentially tilted aggregation mechanism to balance bias and variance in nonparametric routers, improving their robustness to outliers. Experiments show ProxRouter enhances outlier routing while preserving inlier performance with minimal overhead.

---

### Text Prompt Injection of Vision Language Models

Authors: Ruizhe Zhu
Published: 2025-10-10
Categories: cs.CL, cs.CV
arXiv: https://arxiv.org/abs/2510.09849v1

The widespread application of large vision language models has significantly raised safety concerns. In this project, we investigate text prompt injection, a simple yet effective method to mislead these models. We developed an algorithm for this type of attack and demonstrated its effectiveness and efficiency through experiments. Compared to other attack methods, our approach is particularly effective for large models without high demand for computational resources.

---

### Cell Instance Segmentation: The Devil Is in the Boundaries

Authors: Peixian Liang, Yifan Ding, Yizhe Zhang, Jianxu Chen, Hao Zheng, Hongxiao Wang, Yejia Zhang, Guangyu Meng, Tim Weninger, Michael Niemier, X. Sharon Hu, Danny Z Chen
Published: 2025-10-10
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.09848v1

State-of-the-art (SOTA) methods for cell instance segmentation are based on deep learning (DL) semantic segmentation approaches, focusing on distinguishing foreground pixels from background pixels. In order to identify cell instances from foreground pixels (e.g., pixel clustering), most methods decompose instance information into pixel-wise objectives, such as distances to foreground-background boundaries (distance maps), heat gradients with the center point as heat source (heat diffusion maps), and distances from the center point to foreground-background boundaries with fixed angles (star-shaped polygons). However, pixel-wise objectives may lose significant geometric properties of the cell instances, such as shape, curvature, and convexity, which require a collection of pixels to represent. To address this challenge, we present a novel pixel clustering method, called Ceb (for Cell boundaries), to leverage cell boundary features and labels to divide foreground pixels into cell instances. Starting with probability maps generated from semantic segmentation, Ceb first extracts potential foreground-foreground boundaries with a revised Watershed algorithm. For each boundary candidate, a boundary feature representation (called boundary signature) is constructed by sampling pixels from the current foreground-foreground boundary as well as the neighboring background-foreground boundaries. Next, a boundary classifier is used to predict its binary boundary label based on the corresponding boundary signature. Finally, cell instances are obtained by dividing or merging neighboring regions based on the predicted boundary labels. Extensive experiments on six datasets demonstrate that Ceb outperforms existing pixel clustering methods on semantic segmentation probability maps. Moreover, Ceb achieves highly competitive performance compared to SOTA cell instance segmentation methods.

---

### CALM: A Causal Analysis Language Model for Tabular Data in Complex Systems with Local Scores, Conditional Independence Tests, and Relation Attributes

Authors: Zhenjiang Fan, Zengyi Qin, Yuanning Zheng, Bo Xiong, Summer Han
Published: 2025-10-10
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2510.09846v1

Causal discovery from observational data is fundamental to scientific fields like biology, where controlled experiments are often impractical. However, existing methods, including constraint-based (e.g., PC, causalMGM) and score-based approaches (e.g., NOTEARS), face significant limitations. These include an inability to resolve causal direction, restrictions to linear associations, sensitivity to violations of the faithfulness assumption, and inefficiency in searching vast hypothesis spaces. While large language models (LLMs) offer powerful reasoning capabilities, their application is hindered by a fundamental discrepancy: they are designed for text, while most causal data is tabular. To address these challenges, we introduce CALM, a novel causal analysis language model specifically designed for tabular data in complex systems. CALM leverages a Mamba-based architecture to classify causal patterns from pairwise variable relationships. It integrates a comprehensive suite of evidence, including local causal scores, conditional independence tests, and relational attributes, to capture a wide spectrum of linear, nonlinear, and conditional causal mechanisms. Trained on a diverse corpus of synthetic data (from linear, mixed, and nonlinear models) and 10 real-world biological datasets with rigorously validated causal relationships, our model ensures robustness and generalizability. Empirical evaluation demonstrates that CALM significantly outperforms existing methods in both simulation studies, achieving over 91% accuracy, and in a real-world application identifying causal factors in Hepatitis C virus progression. This work represents a significant step towards accurate and generalizable causal discovery by successfully adapting the pattern recognition capabilities of language models to the intricacies of tabular data.

---

### Harnessing Self-Supervised Deep Learning and Geostationary Remote Sensing for Advancing Wildfire and Associated Air Quality Monitoring: Improved Smoke and Fire Front Masking using GOES and TEMPO Radiance Data

Authors: Nicholas LaHaye, Thilanka Munashinge, Hugo Lee, Xiaohua Pan, Gonzalo Gonzalez Abad, Hazem Mahmoud, Jennifer Wei
Published: 2025-10-10
Categories: cs.LG, cs.AI, cs.CV
arXiv: https://arxiv.org/abs/2510.09845v1

This work demonstrates the possibilities for improving wildfire and air quality management in the western United States by leveraging the unprecedented hourly data from NASA's TEMPO satellite mission and advances in self-supervised deep learning. Here we demonstrate the efficacy of deep learning for mapping the near real-time hourly spread of wildfire fronts and smoke plumes using an innovative self-supervised deep learning-system: successfully distinguishing smoke plumes from clouds using GOES-18 and TEMPO data, strong agreement across the smoke and fire masks generated from different sensing modalities as well as significant improvement over operational products for the same cases.

---

### Exploration of Incremental Synthetic Non-Morphed Images for Single Morphing Attack Detection

Authors: David Benavente-Rios, Juan Ruiz Rodriguez, Gustavo Gatica
Published: 2025-10-10
Categories: cs.CV, cs.CR, cs.LG, eess.IV
arXiv: https://arxiv.org/abs/2510.09836v1

This paper investigates the use of synthetic face data to enhance Single-Morphing Attack Detection (S-MAD), addressing the limitations of availability of large-scale datasets of bona fide images due to privacy concerns. Various morphing tools and cross-dataset evaluation schemes were utilized to conduct this study. An incremental testing protocol was implemented to assess the generalization capabilities as more and more synthetic images were added. The results of the experiments show that generalization can be improved by carefully incorporating a controlled number of synthetic images into existing datasets or by gradually adding bona fide images during training. However, indiscriminate use of synthetic data can lead to sub-optimal performance. Evenmore, the use of only synthetic data (morphed and non-morphed images) achieves the highest Equal Error Rate (EER), which means in operational scenarios the best option is not relying only on synthetic data for S-MAD.

---

### Post Processing of image segmentation using Conditional Random Fields

Authors: Aashish Dhawan, Pankaj Bodani, Vishal Garg
Published: 2025-10-10
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.09833v1

The output of image the segmentation process is usually not very clear due to low quality features of Satellite images. The purpose of this study is to find a suitable Conditional Random Field (CRF) to achieve better clarity in a segmented image. We started with different types of CRFs and studied them as to why they are or are not suitable for our purpose. We evaluated our approach on two different datasets - Satellite imagery having low quality features and high quality Aerial photographs. During the study we experimented with various CRFs to find which CRF gives the best results on images and compared our results on these datasets to show the pitfalls and potentials of different approaches.

---

### An Exploration of Non-Euclidean Gradient Descent: Muon and its Many Variants

Authors: Michael Crawshaw, Chirag Modi, Mingrui Liu, Robert M. Gower
Published: 2025-10-10
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2510.09827v1

To define a steepest descent method over a neural network, we need to choose a norm for each layer, a way to aggregate these norms across layers, and whether to use normalization. We systematically explore different alternatives for aggregating norms across layers, both formalizing existing combinations of Adam and the recently proposed Muon as a type of non-Euclidean gradient descent, and deriving new variants of the Muon optimizer. Through a comprehensive experimental evaluation of the optimizers within our framework, we find that Muon is sensitive to the choice of learning rate, whereas a new variant we call MuonMax is significantly more robust. We then show how to combine any non-Euclidean gradient method with model based momentum (known as Momo). The new Momo variants of Muon are significantly more robust to hyperparameter tuning, and often achieve a better validation score. Thus for new tasks, where the optimal hyperparameters are not known, we advocate for using Momo in combination with MuonMax to save on costly hyperparameter tuning.

---

### Decomposer Networks: Deep Component Analysis and Synthesis

Authors: Mohsen Joneidi
Published: 2025-10-10
Categories: cs.LG, cs.CV, cs.IT, cs.NE
arXiv: https://arxiv.org/abs/2510.09825v1

We propose the Decomposer Networks (DecompNet), a semantic autoencoder that factorizes an input into multiple interpretable components. Unlike classical autoencoders that compress an input into a single latent representation, the Decomposer Network maintains N parallel branches, each assigned a residual input defined as the original signal minus the reconstructions of all other branches. By unrolling a Gauss--Seidel style block-coordinate descent into a differentiable network, DecompNet enforce explicit competition among components, yielding parsimonious, semantically meaningful representations. We situate our model relative to linear decomposition methods (PCA, NMF), deep unrolled optimization, and object-centric architectures (MONet, IODINE, Slot Attention), and highlight its novelty as the first semantic autoencoder to implement an all-but-one residual update rule.

---

### Task-Aware Resolution Optimization for Visual Large Language Models

Authors: Weiqing Luo, Zhen Tan, Yifan Li, Xinyu Zhao, Kwonjoon Lee, Behzad Dariush, Tianlong Chen
Published: 2025-10-10
Categories: cs.CV, cs.CL
arXiv: https://arxiv.org/abs/2510.09822v1

Real-world vision-language applications demand varying levels of perceptual granularity. However, most existing visual large language models (VLLMs), such as LLaVA, pre-assume a fixed resolution for downstream tasks, which leads to subpar performance. To address this problem, we first conduct a comprehensive and pioneering investigation into the resolution preferences of different vision-language tasks, revealing a correlation between resolution preferences with image complexity, and uncertainty variance of the VLLM at different image input resolutions. Building on this insight, we propose an empirical formula to determine the optimal resolution for a given vision-language task, combining these two factors. Second, based on rigorous experiments, we propose a novel parameter-efficient fine-tuning technique to extend the visual input resolution of pre-trained VLLMs to the identified optimal resolution. Extensive experiments on various vision-language tasks validate the effectiveness of our method.

---

### Cross-Sensor Touch Generation

Authors: Samanta Rodriguez, Yiming Dou, Miquel Oller, Andrew Owens, Nima Fazeli
Published: 2025-10-10
Categories: cs.RO, cs.CV
arXiv: https://arxiv.org/abs/2510.09817v1

Today's visuo-tactile sensors come in many shapes and sizes, making it challenging to develop general-purpose tactile representations. This is because most models are tied to a specific sensor design. To address this challenge, we propose two approaches to cross-sensor image generation. The first is an end-to-end method that leverages paired data (Touch2Touch). The second method builds an intermediate depth representation and does not require paired data (T2D2: Touch-to-Depth-to-Touch). Both methods enable the use of sensor-specific models across multiple sensors via the cross-sensor touch generation process. Together, these models offer flexible solutions for sensor translation, depending on data availability and application needs. We demonstrate their effectiveness on downstream tasks such as in-hand pose estimation and behavior cloning, successfully transferring models trained on one sensor to another. Project page: https://samantabelen.github.io/cross_sensor_touch_generation.

---

### A mathematical theory for understanding when abstract representations emerge in neural networks

Authors: Bin Wang, W. Jeffrey Johnston, Stefano Fusi
Published: 2025-10-10
Categories: q-bio.NC, math.OC, physics.bio-ph, physics.data-an, stat.ML
arXiv: https://arxiv.org/abs/2510.09816v2

Recent experiments in neuroscience reveal that task-relevant variables are often encoded in approximately orthogonal subspaces of neural population activity. These disentangled, or abstract, representations have been observed in multiple brain areas and across different species. These representations have been shown to support out of distribution generalization and rapid learning of novel tasks. The mechanisms by which these representations emerge remain poorly understood, especially in the case of supervised task behavior. Here, we show mathematically that abstract representations of latent variables are guaranteed to appear in the hidden layer of feedforward nonlinear networks when they are trained on tasks that depend directly on these latent variables. These learned abstract representations reflect the semantics of the input stimuli. To show this, we reformulate the usual optimization over the network weights into a mean field optimization problem over the distribution of neural preactivations. We then apply this framework to finite-width ReLU networks and show that the hidden layer of these networks will exhibit an abstract representation at all global minima of the task objective. Finally, we extend our findings to two broad families of activation functions as well as deep feedforward architectures. Together, our results provide an explanation for the widely observed abstract representations in both the brain and artificial neural networks. In addition, the general framework that we develop here provides a mathematically tractable toolkit for understanding the emergence of different kinds of representations in task-optimized, feature-learning network models.

---

### Towards Understanding Ambiguity Resolution in Multimodal Inference of Meaning

Authors: Yufei Wang, Adriana Kovashka, Loretta Fernández, Marc N. Coutanche, Seth Wiener
Published: 2025-10-10
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2510.09815v1

We investigate a new setting for foreign language learning, where learners infer the meaning of unfamiliar words in a multimodal context of a sentence describing a paired image. We conduct studies with human participants using different image-text pairs. We analyze the features of the data (i.e., images and texts) that make it easier for participants to infer the meaning of a masked or unfamiliar word, and what language backgrounds of the participants correlate with success. We find only some intuitive features have strong correlations with participant performance, prompting the need for further investigating of predictive features for success in these tasks. We also analyze the ability of AI systems to reason about participant performance, and discover promising future directions for improving this reasoning ability.

---

### Temporal Lifting as Latent-Space Regularization for Continuous-Time Flow Models in AI Systems

Authors: Jeffrey Camlin
Published: 2025-10-10
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2510.09805v2

We present a latent-space formulation of adaptive temporal lifting for continuous-time dynamical systems. The method introduces a smooth monotone mapping $t \mapsto τ(t)$ that regularizes near-singular behavior of the underlying flow while preserving its conservation laws. In the lifted coordinate, trajectories such as those of the incompressible Navier-Stokes equations on the torus $\mathbb{T}^3$ become globally smooth. From the standpoint of machine-learning dynamics, temporal lifting acts as a continuous-time normalization operator that can stabilize physics-informed neural networks and other latent-flow architectures used in AI systems. The framework links analytic regularity theory with representation-learning methods for stiff or turbulent processes.

---

### How can we assess human-agent interactions? Case studies in software agent design

Authors: Valerie Chen, Rohit Malhotra, Xingyao Wang, Juan Michelini, Xuhui Zhou, Aditya Bharat Soni, Hoang H. Tran, Calvin Smith, Ameet Talwalkar, Graham Neubig
Published: 2025-10-10
Categories: cs.AI
arXiv: https://arxiv.org/abs/2510.09801v3

While benchmarks measure the accuracy of LLM-powered agents, they mostly assume full automation, failing to represent the collaborative nature of real-world use cases. In this paper, we make two major steps towards the rigorous assessment of human-agent interactions. First, we propose PULSE, a framework for more efficient human-centric evaluation of agent designs, which comprises collecting user feedback, training an ML model to predict user satisfaction, and computing results by combining human satisfaction ratings with model-generated pseudo-labels. Second, we deploy PULSE in software engineering -- one of the highest-impact, real-world domains for LLM agents -- via a large-scale web platform built around the open-source agent OpenHands. Across 15k users, we evaluate how three agent design decisions impact developer satisfaction rates. We also show how PULSE can lead to more robust conclusions about agent design, reducing confidence intervals by 40\% compared to a standard A/B test. Finally, we find substantial discrepancies between in-the-wild results with benchmark performance (e.g., the anti-correlation between claude-sonnet-4 and gpt-5), underscoring the limitations of benchmark-driven evaluation. Our framework PULSE provides guidance for future evaluations, and our findings identify opportunities for better software agent designs.

---

### Distributed clustering in partially overlapping feature spaces

Authors: Alessio Maritan, Luca Schenato
Published: 2025-10-10
Categories: cs.DS, cs.DC, cs.LG
arXiv: https://arxiv.org/abs/2510.09799v1

We introduce and address a novel distributed clustering problem where each participant has a private dataset containing only a subset of all available features, and some features are included in multiple datasets. This scenario occurs in many real-world applications, such as in healthcare, where different institutions have complementary data on similar patients. We propose two different algorithms suitable for solving distributed clustering problems that exhibit this type of feature space heterogeneity. The first is a federated algorithm in which participants collaboratively update a set of global centroids. The second is a one-shot algorithm in which participants share a statistical parametrization of their local clusters with the central server, who generates and merges synthetic proxy datasets. In both cases, participants perform local clustering using algorithms of their choice, which provides flexibility and personalized computational costs. Pretending that local datasets result from splitting and masking an initial centralized dataset, we identify some conditions under which the proposed algorithms are expected to converge to the optimal centralized solution. Finally, we test the practical performance of the algorithms on three public datasets.

---

### A Unified Framework for Lifted Training and Inversion Approaches

Authors: Xiaoyu Wang, Alexandra Valavanis, Azhir Mahmood, Andreas Mang, Martin Benning, Audrey Repetti
Published: 2025-10-10
Categories: cs.LG, math.NA, math.OC, stat.ML
arXiv: https://arxiv.org/abs/2510.09796v2

The training of deep neural networks predominantly relies on a combination of gradient-based optimisation and back-propagation for the computation of the gradient. While incredibly successful, this approach faces challenges such as vanishing or exploding gradients, difficulties with non-smooth activations, and an inherently sequential structure that limits parallelisation. Lifted training methods offer an alternative by reformulating the nested optimisation problem into a higher-dimensional, constrained optimisation problem where the constraints are no longer enforced directly but penalised with penalty terms. This chapter introduces a unified framework that encapsulates various lifted training strategies, including the Method of Auxiliary Coordinates, Fenchel Lifted Networks, and Lifted Bregman Training, and demonstrates how diverse architectures, such as Multi-Layer Perceptrons, Residual Neural Networks, and Proximal Neural Networks fit within this structure. By leveraging tools from convex optimisation, particularly Bregman distances, the framework facilitates distributed optimisation, accommodates non-differentiable proximal activations, and can improve the conditioning of the training landscape. We discuss the implementation of these methods using block-coordinate descent strategies, including deterministic implementations enhanced by accelerated and adaptive optimisation techniques, as well as implicit stochastic gradient methods. Furthermore, we explore the application of this framework to inverse problems, detailing methodologies for both the training of specialised networks (e.g., unrolled architectures) and the stable inversion of pre-trained networks. Numerical results on standard imaging tasks validate the effectiveness and stability of the lifted Bregman approach compared to conventional training, particularly for architectures employing proximal activations.

---

### Causality $\neq$ Decodability, and Vice Versa: Lessons from Interpreting Counting ViTs

Authors: Lianghuan Huang, Yingshan Chang
Published: 2025-10-10
Categories: cs.LG, cs.CV
arXiv: https://arxiv.org/abs/2510.09794v1

Mechanistic interpretability seeks to uncover how internal components of neural networks give rise to predictions. A persistent challenge, however, is disentangling two often conflated notions: decodability--the recoverability of information from hidden states--and causality--the extent to which those states functionally influence outputs. In this work, we investigate their relationship in vision transformers (ViTs) fine-tuned for object counting. Using activation patching, we test the causal role of spatial and CLS tokens by transplanting activations across clean-corrupted image pairs. In parallel, we train linear probes to assess the decodability of count information at different depths. Our results reveal systematic mismatches: middle-layer object tokens exert strong causal influence despite being weakly decodable, whereas final-layer object tokens support accurate decoding yet are functionally inert. Similarly, the CLS token becomes decodable in mid-layers but only acquires causal power in the final layers. These findings highlight that decodability and causality reflect complementary dimensions of representation--what information is present versus what is used--and that their divergence can expose hidden computational circuits.

---

### Principled Operator Learning in Ocean Dynamics: The Role of Temporal Structure

Authors: Vahidreza Jahanmard, Ali Ramezani-Kebrya, Robinson Hordoir
Published: 2025-10-10
Categories: cs.LG, physics.ao-ph
arXiv: https://arxiv.org/abs/2510.09792v1

Neural operators are becoming the default tools to learn solutions to governing partial differential equations (PDEs) in weather and ocean forecasting applications. Despite early promising achievements, significant challenges remain, including long-term prediction stability and adherence to physical laws, particularly for high-frequency processes. In this paper, we take a step toward addressing these challenges in high-resolution ocean prediction by incorporating temporal Fourier modes, demonstrating how this modification enhances physical fidelity. This study compares the standard Fourier Neural Operator (FNO) with its variant, FNOtD, which has been modified to internalize the dispersion relation while learning the solution operator for ocean PDEs. The results demonstrate that entangling space and time in the training of integral kernels enables the model to capture multiscale wave propagation and effectively learn ocean dynamics. FNOtD substantially improves long-term prediction stability and consistency with underlying physical dynamics in challenging high-frequency settings compared to the standard FNO. It also provides competitive predictive skill relative to a state-of-the-art numerical ocean model, while requiring significantly lower computational cost.

---

### Mapping Semantic & Syntactic Relationships with Geometric Rotation

Authors: Michael Freenor, Lauren Alvarez
Published: 2025-10-10
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.09790v2

Understanding how language and embedding models encode semantic relationships is fundamental to model interpretability. While early word embeddings exhibited intuitive vector arithmetic (''king'' - ''man'' + ''woman'' = ''queen''), modern high-dimensional text representations lack straightforward interpretable geometric properties. We introduce Rotor-Invariant Shift Estimation (RISE), a geometric approach that represents semantic-syntactic transformations as consistent rotational operations in embedding space, leveraging the manifold structure of modern language representations. RISE operations have the ability to operate across both languages and models without reducing performance, suggesting the existence of analogous cross-lingual geometric structure. We compare and evaluate RISE using two baseline methods, three embedding models, three datasets, and seven morphologically diverse languages in five major language groups. Our results demonstrate that RISE consistently maps discourse-level semantic-syntactic transformations with distinct grammatical features (e.g., negation and conditionality) across languages and models. This work provides the first demonstration that discourse-level semantic-syntactic transformations correspond to consistent geometric operations in multilingual embedding spaces, empirically supporting the linear representation hypothesis at the sentence level.

---

### Combined Representation and Generation with Diffusive State Predictive Information Bottleneck

Authors: Richard John, Yunrui Qiu, Lukas Herron, Pratyush Tiwary
Published: 2025-10-10
Categories: cs.LG, cond-mat.stat-mech, q-bio.QM
arXiv: https://arxiv.org/abs/2510.09784v1

Generative modeling becomes increasingly data-intensive in high-dimensional spaces. In molecular science, where data collection is expensive and important events are rare, compression to lower-dimensional manifolds is especially important for various downstream tasks, including generation. We combine a time-lagged information bottleneck designed to characterize molecular important representations and a diffusion model in one joint training objective. The resulting protocol, which we term Diffusive State Predictive Information Bottleneck (D-SPIB), enables the balancing of representation learning and generation aims in one flexible architecture. Additionally, the model is capable of combining temperature information from different molecular simulation trajectories to learn a coherent and useful internal representation of thermodynamics. We benchmark D-SPIB on multiple molecular tasks and showcase its potential for exploring physical conditions outside the training set.

---

### Large Language Models for Imbalanced Classification: Diversity makes the difference

Authors: Dang Nguyen, Sunil Gupta, Kien Do, Thin Nguyen, Taylor Braund, Alexis Whitton, Svetha Venkatesh
Published: 2025-10-10
Categories: cs.LG, cs.AI, stat.ML
arXiv: https://arxiv.org/abs/2510.09783v2

Oversampling is one of the most widely used approaches for addressing imbalanced classification. The core idea is to generate additional minority samples to rebalance the dataset. Most existing methods, such as SMOTE, require converting categorical variables into numerical vectors, which often leads to information loss. Recently, large language model (LLM)-based methods have been introduced to overcome this limitation. However, current LLM-based approaches typically generate minority samples with limited diversity, reducing robustness and generalizability in downstream classification tasks. To address this gap, we propose a novel LLM-based oversampling method designed to enhance diversity. First, we introduce a sampling strategy that conditions synthetic sample generation on both minority labels and features. Second, we develop a new permutation strategy for fine-tuning pre-trained LLMs. Third, we fine-tune the LLM not only on minority samples but also on interpolated samples to further enrich variability. Extensive experiments on 10 tabular datasets demonstrate that our method significantly outperforms eight SOTA baselines. The generated synthetic samples are both realistic and diverse. Moreover, we provide theoretical analysis through an entropy-based perspective, proving that our method encourages diversity in the generated samples.

---

### The Geometry of Reasoning: Flowing Logics in Representation Space

Authors: Yufa Zhou, Yixiao Wang, Xunjian Yin, Shuyan Zhou, Anru R. Zhang
Published: 2025-10-10
Categories: cs.AI, cs.CL, cs.LG, cs.LO
arXiv: https://arxiv.org/abs/2510.09782v2

We study how large language models (LLMs) ``think'' through their representation space. We propose a novel geometric framework that models an LLM's reasoning as flows -- embedding trajectories evolving where logic goes. We disentangle logical structure from semantics by employing the same natural deduction propositions with varied semantic carriers, allowing us to test whether LLMs internalize logic beyond surface form. This perspective connects reasoning with geometric quantities such as position, velocity, and curvature, enabling formal analysis in representation and concept spaces. Our theory establishes: (1) LLM reasoning corresponds to smooth flows in representation space, and (2) logical statements act as local controllers of these flows' velocities. Using learned representation proxies, we design controlled experiments to visualize and quantify reasoning flows, providing empirical validation of our theoretical framework. Our findings indicate that training solely via next-token prediction can lead LLMs to internalize logical invariants as higher-order geometry in representation space, challenging the ``stochastic parrot'' argument. Experiments across Qwen and LLaMA model families further suggest the presence of a general, possibly universal, representational law underlying machine understanding and human linguistic regularities, largely independent of specific training recipes or model architectures. Our work serves as both a conceptual foundation and practical tools for studying reasoning phenomena, offering a new lens for interpretability and formal analysis of LLMs' behavior.

---

### Building a Foundational Guardrail for General Agentic Systems via Synthetic Data

Authors: Yue Huang, Hang Hua, Yujun Zhou, Pengcheng Jing, Manish Nagireddy, Inkit Padhi, Greta Dolcetti, Zhangchen Xu, Subhajit Chaudhury, Ambrish Rawat, Liubov Nedoshivina, Pin-Yu Chen et al.
Published: 2025-10-10
Categories: cs.LG, cs.AI, cs.CL
arXiv: https://arxiv.org/abs/2510.09781v1

While LLM agents can plan multi-step tasks, intervening at the planning stage-before any action is executed-is often the safest way to prevent harm, since certain risks can lead to severe consequences once carried out. However, existing guardrails mostly operate post-execution, which is difficult to scale and leaves little room for controllable supervision at the plan level. To address this challenge, we highlight three critical gaps in current research: data gap, model gap, and evaluation gap. To close the data gap, we introduce AuraGen, a controllable engine that (i) synthesizes benign trajectories, (ii) injects category-labeled risks with calibrated difficulty, and (iii) filters outputs via an automated reward model, producing large and reliable corpora for pre-execution safety. To close the guardian model gap, we propose a foundational guardrail Safiron, combining a cross-planner adapter with a compact guardian model. The adapter unifies different input formats, while Safiron flags risky cases, assigns risk types, and generates rationales; trained in two stages with a broadly explored data recipe, Safiron achieves robust transfer across settings. To close the evaluation gap, we release Pre-Exec Bench, a realistic benchmark covering diverse tools and branching trajectories, which measures detection, fine-grained categorization, explanation, and cross-planner generalization in human-verified scenarios. Extensive experiments demonstrate consistent gains of the proposed guardrail over strong baselines on Pre-Exec Bench, and ablations further distill actionable practices, providing a practical template for safer agentic systems.

---