# Research Papers — 2025-10 · bundle page 102 of 217

> Full-text markdown of papers #5051–5100 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/102
[← Previous bundle](https://aipod.dev/papers/archive/2025-10/page/101.md)
[Next bundle →](https://aipod.dev/papers/archive/2025-10/page/103.md)

### Backdoor Unlearning by Linear Task Decomposition

Authors: Amel Abdelraheem, Alessandro Favero, Gerome Bovet, Pascal Frossard
Published: 2025-10-16
Categories: cs.LG, cs.CV
arXiv: https://arxiv.org/abs/2510.14845v1

Foundation models have revolutionized computer vision by enabling broad generalization across diverse tasks. Yet, they remain highly susceptible to adversarial perturbations and targeted backdoor attacks. Mitigating such vulnerabilities remains an open challenge, especially given that the large-scale nature of the models prohibits retraining to ensure safety. Existing backdoor removal approaches rely on costly fine-tuning to override the harmful behavior, and can often degrade performance on other unrelated tasks. This raises the question of whether backdoors can be removed without compromising the general capabilities of the models. In this work, we address this question and study how backdoors are encoded in the model weight space, finding that they are disentangled from other benign tasks. Specifically, this separation enables the isolation and erasure of the backdoor's influence on the model with minimal impact on clean performance. Building on this insight, we introduce a simple unlearning method that leverages such disentanglement. Through extensive experiments with CLIP-based models and common adversarial triggers, we show that, given the knowledge of the attack, our method achieves approximately perfect unlearning, while retaining, on average, 96% of clean accuracy. Additionally, we demonstrate that even when the attack and its presence are unknown, our method successfully unlearns backdoors by proper estimation using reverse-engineered triggers. Overall, our method consistently yields better unlearning and clean accuracy tradeoffs when compared to present state-of-the-art defenses.

---

### Provable Unlearning with Gradient Ascent on Two-Layer ReLU Neural Networks

Authors: Odelia Melamed, Gilad Yehudai, Gal Vardi
Published: 2025-10-16
Categories: cs.LG, cs.CR, cs.NE, stat.ML
arXiv: https://arxiv.org/abs/2510.14844v1

Machine Unlearning aims to remove specific data from trained models, addressing growing privacy and ethical concerns. We provide a theoretical analysis of a simple and widely used method - gradient ascent - used to reverse the influence of a specific data point without retraining from scratch. Leveraging the implicit bias of gradient descent towards solutions that satisfy the Karush-Kuhn-Tucker (KKT) conditions of a margin maximization problem, we quantify the quality of the unlearned model by evaluating how well it satisfies these conditions w.r.t. the retained data. To formalize this idea, we propose a new success criterion, termed \textbf{$(ε, δ, τ)$-successful} unlearning, and show that, for both linear models and two-layer neural networks with high dimensional data, a properly scaled gradient-ascent step satisfies this criterion and yields a model that closely approximates the retrained solution on the retained data. We also show that gradient ascent performs successful unlearning while still preserving generalization in a synthetic Gaussian-mixture setting.

---

### Boosting Instruction Following at Scale

Authors: Ben Elder, Evelyn Duesterwald, Vinod Muthusamy
Published: 2025-10-16
Categories: cs.AI
arXiv: https://arxiv.org/abs/2510.14842v1

A typical approach developers follow to influence an LLM's behavior in an application is through careful manipulation of the prompt, such as by adding or modifying instructions. However, merely adding more instructions provides little assurance that they will actually be followed. We introduce Instruction Boosting as a post-generation method to increase the reliability of LLM prompt instructions. We show that Instruction Boosting improves the instruction following rate by up to 7 points for two instructions and up to 4 points for ten instructions. To demonstrate these results we introduce SCALEDIF, a benchmark with a scaled instruction volume of up to ten instructions per data sample. We also present an analysis of the commonly observed trend that performance degrades as more instructions are added. We show that an important factor contributing to this trend is the degree of tension and conflict that arises as the number of instructions is increased. We contribute a quantitative conflict scoring tool that explains the observed performance trends and provides feedback to developers on the impact that additional prompt instructions have on a model's performance.

---

### Reinforcement Learning with Stochastic Reward Machines

Authors: Jan Corazza, Ivan Gavran, Daniel Neider
Published: 2025-10-16
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2510.14837v1

Reward machines are an established tool for dealing with reinforcement learning problems in which rewards are sparse and depend on complex sequences of actions. However, existing algorithms for learning reward machines assume an overly idealized setting where rewards have to be free of noise. To overcome this practical limitation, we introduce a novel type of reward machines, called stochastic reward machines, and an algorithm for learning them. Our algorithm, based on constraint solving, learns minimal stochastic reward machines from the explorations of a reinforcement learning agent. This algorithm can easily be paired with existing reinforcement learning algorithms for reward machines and guarantees to converge to an optimal policy in the limit. We demonstrate the effectiveness of our algorithm in two case studies and show that it outperforms both existing methods and a naive approach for handling noisy reward functions.

---

### QDepth-VLA: Quantized Depth Prediction as Auxiliary Supervision for Vision-Language-Action Models

Authors: Yixuan Li, Yuhui Chen, Mingcai Zhou, Haoran Li, Zhengtao Zhang, Dongbin Zhao
Published: 2025-10-16
Categories: cs.CV, cs.RO
arXiv: https://arxiv.org/abs/2510.14836v3

Spatial perception and reasoning are crucial for Vision-Language-Action (VLA) models to accomplish fine-grained manipulation tasks. However, existing approaches often lack the ability to understand and reason over the essential 3D structures necessary for precise control. To address this limitation, we propose QDepth-VLA, a general framework that augments VLA models with an auxiliary depth prediction task. A dedicated depth expert is designed to predict quantized latent tokens of depth maps obtained from a VQ-VAE encoder, enabling the model to learn depth-aware representations that capture critical geometric cues. Experimental results on the simulation benchmarks and real-world tasks demonstrate that QDepth-VLA yields strong spatial reasoning and competitive performance on manipulation tasks.

---

### Intelligent Dynamic Handover via AI-assisted Signal Quality Prediction in 6G Multi-RAT Networks

Authors: Maria Lamprini A. Bartsioka, Anastasios Giannopoulos, Sotirios Spantideas
Published: 2025-10-16
Categories: cs.LG, cs.NI
arXiv: https://arxiv.org/abs/2510.14832v1

The emerging paradigm of 6G multiple Radio Access Technology (multi-RAT) networks, where cellular and Wireless Fidelity (WiFi) transmitters coexist, requires mobility decisions that remain reliable under fast channel dynamics, interference, and heterogeneous coverage. Handover in multi-RAT deployments is still highly reactive and event-triggered, relying on instantaneous measurements and threshold events. This work proposes a Machine Learning (ML)-assisted Predictive Conditional Handover (P-CHO) framework based on a model-driven and short-horizon signal quality forecasts. We present a generalized P-CHO sequence workflow orchestrated by a RAT Steering Controller, which standardizes data collection, parallel per-RAT predictions, decision logic with hysteresis-based conditions, and CHO execution. Considering a realistic multi-RAT environment, we train RAT-aware Long Short Term Memory (LSTM) networks to forecast the signal quality indicators of mobile users along randomized trajectories. The proposed P-CHO models are trained and evaluated under different channel models for cellular and IEEE 802.11 WiFi integrated coverage. We study the impact of hyperparameter tuning of LSTM models under different system settings, and compare direct multi-step versus recursive P-CHO variants. Comparisons against baseline predictors are also carried out. Finally, the proposed P-CHO is tested under soft and hard handover settings, showing that hysteresis-enabled P-CHO scheme is able to reduce handover failures and ping-pong events. Overall, the proposed P-CHO framework can enable accurate, low-latency, and proactive handovers suitable for ML-assisted handover steering in 6G multi-RAT deployments.

---

### Scaling Tumor Segmentation: Best Lessons from Real and Synthetic Data

Authors: Qi Chen, Xinze Zhou, Chen Liu, Hao Chen, Wenxuan Li, Zekun Jiang, Ziyan Huang, Yuxuan Zhao, Dexin Yu, Junjun He, Yefeng Zheng, Ling Shao et al.
Published: 2025-10-16
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.14831v2

AI for tumor segmentation is limited by the lack of large, voxel-wise annotated datasets, which are hard to create and require medical experts. In our proprietary JHH dataset of 3,000 annotated pancreatic tumor scans, we found that AI performance stopped improving after 1,500 scans. With synthetic data, we reached the same performance using only 500 real scans. This finding suggests that synthetic data can steepen data scaling laws, enabling more efficient model training than real data alone. Motivated by these lessons, we created AbdomenAtlas 2.0--a dataset of 10,135 CT scans with a total of 15,130 tumor instances per-voxel manually annotated in six organs (pancreas, liver, kidney, colon, esophagus, and uterus) and 5,893 control scans. Annotated by 23 expert radiologists, it is several orders of magnitude larger than existing public tumor datasets. While we continue expanding the dataset, the current version of AbdomenAtlas 2.0 already provides a strong foundation--based on lessons from the JHH dataset--for training AI to segment tumors in six organs. It achieves notable improvements over public datasets, with a +7% DSC gain on in-distribution tests and +16% on out-of-distribution tests.

---

### RL-100: Performant Robotic Manipulation with Real-World Reinforcement Learning

Authors: Kun Lei, Huanyu Li, Dongjie Yu, Zhenyu Wei, Lingxiao Guo, Zhennan Jiang, Ziyu Wang, Shiyu Liang, Huazhe Xu
Published: 2025-10-16
Categories: cs.RO, cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2510.14830v4

Real-world robotic manipulation in homes and factories demands reliability, efficiency, and robustness that approach or surpass those of skilled human operators. We present RL-100, a real-world reinforcement learning framework built on diffusion visuomotor policies. RL-100 unifies imitation and reinforcement learning under a single clipped PPO surrogate objective applied within the denoising process, yielding conservative and stable improvements across offline and online stages. To meet deployment latency requirements, a lightweight consistency distillation method compresses multi-step diffusion into a one-step controller for high-frequency control. The framework is task-, embodiment-, and representation-agnostic, and supports both single-action and action-chunking control. We evaluate RL-100 on eight diverse real-robot tasks, from dynamic pushing and agile bowling to pouring, cloth folding, unscrewing, multi-stage juicing, and long-horizon box folding. RL-100 attains 100 percent success across evaluated trials, for a total of 1000 out of 1000 episodes, including up to 250 out of 250 consecutive trials on one task. It matches or surpasses expert teleoperators in time to completion. Without retraining, a single policy attains approximately 90 percent zero-shot success under environmental and dynamics shifts, adapts in a few-shot regime to significant task variations (86.7 percent), and remains robust to aggressive human perturbations (about 96 percent). Notably, our juicing robot served random customers continuously for about seven hours without failure when deployed zero-shot in a shopping mall. These results suggest a practical path to deployment-ready robot learning by starting from human priors, aligning training objectives with human-grounded metrics, and reliably extending performance beyond human demonstrations.

---

### RoboGPT-R1: Enhancing Robot Task Planning with Reinforcement Learning

Authors: Jinrui Liu, Bingyan Nie, Boyu Li, Yaran Chen, Yuze Wang, Shunsen He, Haoran Li
Published: 2025-10-16
Categories: cs.AI, cs.RO
arXiv: https://arxiv.org/abs/2510.14828v3

Improving the reasoning capabilities of embodied agents is crucial for robots to complete complex human instructions in long-view manipulation tasks successfully. Despite the success of large language models and vision language models based on Supervised Fine-Tuning (SFT) in planning tasks, they continue facing challenges in performing long-horizon manipulation tasks in complex real-world environments, owing to their restricted common sense and reasoning capabilities. Considering that aligning general-purpose vision language models to robotic planning tasks via supervised fine-tuning suffers from poor generalization and insufficient physical understanding, we propose RoboGPT-R1, a two-stage fine-tuning framework for embodied planning. In this framework, supervised training acquires foundational knowledge through expert sequences, followed by RL to address the model's shortcomings in visual-spatial understanding and reasoning. To achieve physical understanding and action sequence consistency in multi-step reasoning tasks, we design a rule-based reward function that simultaneously considers long-horizon performance and action constraint in the environment. The reasoning model, trained on Qwen2.5-VL-3B, significantly outperforms the larger-scale model, GPT-4o-mini, by 21.33% and surpasses other work trained on Qwen2.5-VL-7B by 20.33% on the EmbodiedBench benchmark.

---

### To Infinity and Beyond: Tool-Use Unlocks Length Generalization in State Space Models

Authors: Eran Malach, Omid Saremi, Sinead Williamson, Arwen Bradley, Aryo Lotfi, Emmanuel Abbe, Josh Susskind, Etai Littwin
Published: 2025-10-16
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.14826v1

State Space Models (SSMs) have become the leading alternative to Transformers for sequence modeling. Their primary advantage is efficiency in long-context and long-form generation, enabled by fixed-size memory and linear scaling of computational complexity. We begin this work by showing a simple theoretical result stating that SSMs cannot accurately solve any ``truly long-form'' generation problem (in a sense we formally define), undermining their main competitive advantage. However, we show that this limitation can be mitigated by allowing SSMs interactive access to external tools. In fact, we show that given the right choice of tool access and problem-dependent training data, SSMs can learn to solve any tractable problem and generalize to arbitrary problem length/complexity (i.e., achieve length generalization). Following our theoretical finding, we demonstrate that tool-augmented SSMs achieve remarkable length generalization on a variety of arithmetic, reasoning, and coding tasks. These findings highlight SSMs as a potential efficient alternative to Transformers in interactive tool-based and agentic settings.

---

### Programmatic Representation Learning with Language Models

Authors: Gabriel Poesia, Georgia Gabriela Sampaio
Published: 2025-10-16
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.14825v1

Classical models for supervised machine learning, such as decision trees, are efficient and interpretable predictors, but their quality is highly dependent on the particular choice of input features. Although neural networks can learn useful representations directly from raw data (e.g., images or text), this comes at the expense of interpretability and the need for specialized hardware to run them efficiently. In this paper, we explore a hypothesis class we call Learned Programmatic Representations (LeaPR) models, which stack arbitrary features represented as code (functions from data points to scalars) and decision tree predictors. We synthesize feature functions using Large Language Models (LLMs), which have rich prior knowledge in a wide range of domains and a remarkable ability to write code using existing domain-specific libraries. We propose two algorithms to learn LeaPR models from supervised data. First, we design an adaptation of FunSearch to learn features rather than directly generate predictors. Then, we develop a novel variant of the classical ID3 algorithm for decision tree learning, where new features are generated on demand when splitting leaf nodes. In experiments from chess position evaluation to image and text classification, our methods learn high-quality, neural network-free predictors often competitive with neural networks. Our work suggests a flexible paradigm for learning interpretable representations end-to-end where features and predictions can be readily inspected and understood.

---

### Supervised Fine-Tuning or Contrastive Learning? Towards Better Multimodal LLM Reranking

Authors: Ziqi Dai, Xin Zhang, Mingxin Li, Yanzhao Zhang, Dingkun Long, Pengjun Xie, Meishan Zhang, Wenjie Li, Min Zhang
Published: 2025-10-16
Categories: cs.CL, cs.CV, cs.IR
arXiv: https://arxiv.org/abs/2510.14824v1

In information retrieval, training reranking models mainly focuses on two types of objectives: metric learning (e.g. contrastive loss to increase the predicted scores on relevant query-document pairs) and classification (binary label prediction of relevance vs. irrelevance). For BERT-style encoders, various studies have shown that contrastive learning (CL) can be more effective than discriminative (classification) learning. However, for large language models (LLMs), classification via supervised fine-tuning (SFT), which predicts ''yes'' (resp. ''no'') token for relevant (resp. irrelevant) pairs, appears more promising as it aligns well with the generative nature of LLMs. This divergence raises a central question: which objective is intrinsically better suited to LLM-based reranking, and what mechanism underlies the difference? In this work, we conduct a comprehensive comparison and analysis between CL and SFT for reranking, taking the universal multimodal retrieval (UMR) as the experimental playground. We first decompose the objectives into two components: weight, which controls the magnitude of those updates, and direction, which guides the model updates, then present a unified framework for understanding their interactions. Through probing experiments, we find that SFT provides a substantially stronger weighting scheme than CL, whereas the preferred scoring direction shows no clear winner. Taken together, these results point to a consistent advantage of SFT over CL for LLM reranking. To further validate our findings, we conduct large-scale training with SFT and present new state-of-the-art rerankers on the MRB benchmark. We also provide ablations on SFT settings and expect our findings to benefit future research and applications in this area.

---

### FraQAT: Quantization Aware Training with Fractional bits

Authors: Luca Morreale, Alberto Gil C. P. Ramos, Malcolm Chadwick, Mehid Noroozi, Ruchika Chavhan, Abhinav Mehrotra, Sourav Bhattacharya
Published: 2025-10-16
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.14823v1

State-of-the-art (SOTA) generative models have demonstrated impressive capabilities in image synthesis or text generation, often with a large capacity model. However, these large models cannot be deployed on smartphones due to the limited availability of on-board memory and computations. Quantization methods lower the precision of the model parameters, allowing for efficient computations, \eg, in \INT{8}. Although aggressive quantization addresses efficiency and memory constraints, preserving the quality of the model remains a challenge. To retain quality in previous aggressive quantization, we propose a new fractional bits quantization (\short) approach. The novelty is a simple yet effective idea: we progressively reduce the model's precision from 32 to 4 bits per parameter, and exploit the fractional bits during optimization to maintain high generation quality. We show that the \short{} yields improved quality on a variety of diffusion models, including SD3.5-Medium, Sana, \pixart, and FLUX.1-schnell, while achieving $4-7\%$ lower FiD than standard QAT. Finally, we deploy and run Sana on a Samsung S25U, which runs on the Qualcomm SM8750-AB Snapdragon 8 Elite Hexagon Tensor Processor (HTP).

---

### Capturing Context-Aware Route Choice Semantics for Trajectory Representation Learning

Authors: Ji Cao, Yu Wang, Tongya Zheng, Jie Song, Qinghong Guo, Zujie Ren, Canghong Jin, Gang Chen, Mingli Song
Published: 2025-10-16
Categories: cs.CV, cs.LG
arXiv: https://arxiv.org/abs/2510.14819v3

Trajectory representation learning (TRL) aims to encode raw trajectory data into low-dimensional embeddings for downstream tasks such as travel time estimation, mobility prediction, and trajectory similarity analysis. From a behavioral perspective, a trajectory reflects a sequence of route choices within an urban environment. However, most existing TRL methods ignore this underlying decision-making process and instead treat trajectories as static, passive spatiotemporal sequences, thereby limiting the semantic richness of the learned representations. To bridge this gap, we propose CORE, a TRL framework that integrates context-aware route choice semantics into trajectory embeddings. CORE first incorporates a multi-granular Environment Perception Module, which leverages large language models (LLMs) to distill environmental semantics from point of interest (POI) distributions, thereby constructing a context-enriched road network. Building upon this backbone, CORE employs a Route Choice Encoder with a mixture-of-experts (MoE) architecture, which captures route choice patterns by jointly leveraging the context-enriched road network and navigational factors. Finally, a Transformer encoder aggregates the route-choice-aware representations into a global trajectory embedding. Extensive experiments on 4 real-world datasets across 6 downstream tasks demonstrate that CORE consistently outperforms 15 state-of-the-art TRL methods, achieving an average improvement of 9.20\% over the best-performing baseline. Our code is available at https://github.com/caoji2001/CORE.

---

### Tackling Time-Series Forecasting Generalization via Mitigating Concept Drift

Authors: Zhiyuan Zhao, Haoxin Liu, B. Aditya Prakash
Published: 2025-10-16
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.14814v2

Time-series forecasting finds broad applications in real-world scenarios. Due to the dynamic nature of time series data, it is important for time-series forecasting models to handle potential distribution shifts over time. In this paper, we initially identify two types of distribution shifts in time series: concept drift and temporal shift. We acknowledge that while existing studies primarily focus on addressing temporal shift issues in time series forecasting, designing proper concept drift methods for time series forecasting has received comparatively less attention. Motivated by the need to address potential concept drift, while conventional concept drift methods via invariant learning face certain challenges in time-series forecasting, we propose a soft attention mechanism that finds invariant patterns from both lookback and horizon time series. Additionally, we emphasize the critical importance of mitigating temporal shifts as a preliminary to addressing concept drift. In this context, we introduce ShifTS, a method-agnostic framework designed to tackle temporal shift first and then concept drift within a unified approach. Extensive experiments demonstrate the efficacy of ShifTS in consistently enhancing the forecasting accuracy of agnostic models across multiple datasets, and outperforming existing concept drift, temporal shift, and combined baselines.

---

### SHUFFLESPARSE: Learned Shuffles for Structured Sparse Networks

Authors: Abhishek Tyagi, Arjun Iyer, Liam Young, William H Renninger, Christopher Kanan, Yuhao Zhu
Published: 2025-10-16
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.14812v2

Structured weight sparsity accelerates training and inference on modern GPUs, but it trails unstructured dynamic sparse training (DST) in accuracy especially at extreme sparsity. We pinpoint the reason for this difference in performance to a lack of expressivity: a dense layer can implement any pattern of non-zero weights, whereas structured patterns are restricted to only a small set of weight configurations. We introduce SHUFFLESPARSE, a single permutation primitive that applies uniformly across DST-from-scratch and one-shot pruning, and across N:M, block etc. We close most of this gap by learning a single permutation matrix jointly with the structured weight matrix. When used on three different types of structures (block, N:M, and diagonal), SHUFFLESPARSE is able to reduce the structured-vs-unstructured accuracy gap on ViT-B16 (ImageNet-1K) and GPT-2 (WikiText-103) at 90-95% sparsity, while adding minimal inference overhead (< 8.7% inference overhead) and preserving any training acceleration the host structure provides. The same permutation formulation transfers to one-shot 2:4 pruning of pretrained LLMs, where it improves zero-shot accuracy by 4.6 points on LLaMA-2 7B. Together, these results establish learned permutations as a general tool for recovering unstructured-level accuracy from structured sparsity patterns.

---

### Rethinking Hebbian Principle: Low-Dimensional Structural Projection for Unsupervised Learning

Authors: Shikuang Deng, Jiayuan Zhang, Yuhang Wu, Ting Chen, Shi Gu
Published: 2025-10-16
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.14810v2

Hebbian learning is a biological principle that intuitively describes how neurons adapt their connections through repeated stimuli. However, when applied to machine learning, it suffers serious issues due to the unconstrained updates of the connections and the lack of accounting for feedback mediation. Such shortcomings limit its effective scaling to complex network architectures and tasks. To this end, here we introduce the Structural Projection Hebbian Representation (SPHeRe), a novel unsupervised learning method that integrates orthogonality and structural information preservation through a local auxiliary nonlinear block. The loss for structural information preservation backpropagates to the input through an auxiliary lightweight projection that conceptually serves as feedback mediation while the orthogonality constraints account for the boundedness of updating magnitude. Extensive experimental results show that SPHeRe achieves SOTA performance among unsupervised synaptic plasticity approaches on standard image classification benchmarks, including CIFAR-10, CIFAR-100, and Tiny-ImageNet. Furthermore, the method exhibits strong effectiveness in continual learning and transfer learning scenarios, and image reconstruction tasks show the robustness and generalizability of the extracted features. This work demonstrates the competitiveness and potential of Hebbian unsupervised learning rules within modern deep learning frameworks, demonstrating the possibility of efficient and biologically inspired learning algorithms without the strong dependence on strict backpropagation. Our code is available at https://github.com/brain-intelligence-lab/SPHeRe.

---

### Agentic NL2SQL to Reduce Computational Costs

Authors: Dominik Jehle, Lennart Purucker, Frank Hutter
Published: 2025-10-16
Categories: cs.AI, cs.LG
arXiv: https://arxiv.org/abs/2510.14808v1

Translating natural language queries into SQL queries (NL2SQL or Text-to-SQL) has recently been empowered by large language models (LLMs). Using LLMs to perform NL2SQL methods on a large collection of SQL databases necessitates processing large quantities of meta-information about the databases, which in turn results in lengthy prompts with many tokens and high processing costs. To address this challenge, we introduce Datalake Agent, an agentic system designed to enable an LLM to solve NL2SQL tasks more efficiently. Instead of utilizing direct solvers for NL2SQL that call the LLM once with all meta-information in the prompt, the Datalake Agent employs an interactive loop to reduce the utilized meta-information. Within the loop, the LLM is used in a reasoning framework that selectively requests only the necessary information to solve a table question answering task. We evaluate the Datalake Agent on a collection of 23 databases with 100 table question answering tasks. The Datalake Agent reduces the tokens used by the LLM by up to 87\% and thus allows for substantial cost reductions while maintaining competitive performance.

---

### Beyond the Sampled Token: Preserving Candidate Support in RLVR

Authors: Ruotian Peng, Yi Ren, Zhouliang Yu, Weiyang Liu, Yandong Wen
Published: 2025-10-16
Categories: cs.AI
arXiv: https://arxiv.org/abs/2510.14807v3

We revisit exploration collapse in reinforcement learning with verifiable rewards (RLVR), from the perspective of the \emph{candidate distribution} for next-token prediction. We formally show that as probability concentrates on the top-$1$ candidate, the expected number of distinct responses collapses to one regardless of the sampling budget $K$. This theoretical implication is further verified by our empirical tracking of top-$N$ candidate probabilities during training, where the top-$1$ candidate progressively dominates while plausible alternatives are suppressed. These findings suggest a key desideratum for effective exploration: \emph{preserving non-negligible probability mass on the top-$N$ candidates}. To this end, we propose Candidate-aware Support Preservation (CaSP), with two complementary designs. Specifically, CaSP redistributes positive gradients among top-$N$ candidates for correct responses, and applies a stronger penalty to the top-$1$ candidate for incorrect responses. Unlike many exploration-oriented methods that improve pass@$K$ at the cost of pass@1, CaSP improves pass@$K$ across the full $K$ spectrum. These gains generalize to 6 math, 2 logical-reasoning, and 2 coding benchmarks, and scales to 32B-parameter models and sampling budgets up to $K=1024$, positioning it as a principled, candidate-level approach for RLVR exploration.

---

### Scaling Artificial Intelligence for Multi-Tumor Early Detection with More Reports, Fewer Masks

Authors: Pedro R. A. S. Bassi, Xinze Zhou, Wenxuan Li, Szymon Płotka, Jieneng Chen, Qi Chen, Zheren Zhu, Jakub Prządo, Ibrahim E. Hamacı, Sezgin Er, Yuhan Wang, Ashwin Kumar et al.
Published: 2025-10-16
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2510.14803v1

Early tumor detection save lives. Each year, more than 300 million computed tomography (CT) scans are performed worldwide, offering a vast opportunity for effective cancer screening. However, detecting small or early-stage tumors on these CT scans remains challenging, even for experts. Artificial intelligence (AI) models can assist by highlighting suspicious regions, but training such models typically requires extensive tumor masks--detailed, voxel-wise outlines of tumors manually drawn by radiologists. Drawing these masks is costly, requiring years of effort and millions of dollars. In contrast, nearly every CT scan in clinical practice is already accompanied by medical reports describing the tumor's size, number, appearance, and sometimes, pathology results--information that is rich, abundant, and often underutilized for AI training. We introduce R-Super, which trains AI to segment tumors that match their descriptions in medical reports. This approach scales AI training with large collections of readily available medical reports, substantially reducing the need for manually drawn tumor masks. When trained on 101,654 reports, AI models achieved performance comparable to those trained on 723 masks. Combining reports and masks further improved sensitivity by +13% and specificity by +8%, surpassing radiologists in detecting five of the seven tumor types. Notably, R-Super enabled segmentation of tumors in the spleen, gallbladder, prostate, bladder, uterus, and esophagus, for which no public masks or AI models previously existed. This study challenges the long-held belief that large-scale, labor-intensive tumor mask creation is indispensable, establishing a scalable and accessible path toward early detection across diverse tumor types. We plan to release our trained models, code, and dataset at https://github.com/MrGiovanni/R-Super

---

### Morphology-Aware Prognostic model for Five-Year Survival Prediction in Colorectal Cancer from H&E Whole Slide Images

Authors: Usama Sajjad, Abdul Rehman Akbar, Ziyu Su, Deborah Knight, Wendy L. Frankel, Metin N. Gurcan, Wei Chen, Muhammad Khalid Khan Niazi
Published: 2025-10-16
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2510.14800v1

Colorectal cancer (CRC) remains the third most prevalent malignancy globally, with approximately 154,000 new cases and 54,000 projected deaths anticipated for 2025. The recent advancement of foundation models in computational pathology has been largely propelled by task agnostic methodologies that can overlook organ-specific crucial morphological patterns that represent distinct biological processes that can fundamentally influence tumor behavior, therapeutic response, and patient outcomes. The aim of this study is to develop a novel, interpretable AI model, PRISM (Prognostic Representation of Integrated Spatial Morphology), that incorporates a continuous variability spectrum within each distinct morphology to characterize phenotypic diversity and reflecting the principle that malignant transformation occurs through incremental evolutionary processes rather than abrupt phenotypic shifts. PRISM is trained on 8.74 million histological images extracted from surgical resection specimens of 424 patients with stage III CRC. PRISM achieved superior prognostic performance for five-year OS (AUC = 0.70 +- 0.04; accuracy = 68.37% +- 4.75%; HR = 3.34, 95% CI = 2.28-4.90; p < 0.0001), outperforming existing CRC-specific methods by 15% and AI foundation models by ~23% accuracy. It showed sex-agnostic robustness (AUC delta = 0.02; accuracy delta = 0.15%) and stable performance across clinicopathological subgroups, with minimal accuracy fluctuation (delta = 1.44%) between 5FU/LV and CPT-11/5FU/LV regimens, replicating the Alliance cohort finding of no survival difference between treatments.

---

### MSPL: Multi-Step Pseudo-Labeling for Open-Vocabulary Object Detection

Authors: Hojun Choi, Youngsun Lim, Jaeyo Shin, Hyunjung Shim
Published: 2025-10-16
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.14792v4

Open-vocabulary object detection (OVD) aims to recognize and localize object categories beyond the training set. Recent approaches leverage vision-language models to generate pseudo-labels using image-text alignment, allowing detectors to generalize to unseen classes without explicit supervision. However, these methods depend heavily on single-step image-text matching, neglecting the intermediate reasoning steps crucial for interpreting semantically complex visual contexts, such as crowding or occlusion. In this paper, we introduce MSPL, a framework that incorporates multi-step visual reasoning into the pseudo-labeling process for OVD. It decomposes complex scene understanding into three interpretable steps-object localization, category recognition, and background grounding-where these intermediate reasoning states serve as rich supervision sources. Extensive experiments on standard OVD evaluation protocols demonstrate that MSPL achieves state-of-the-art performance with superior pseudo-labeling efficiency, outperforming the strong baseline by 9.4 AP50 for novel classes on OV-COCO and improving box and mask APr by 3.2 and 2.2, respectively, on OV-LVIS. Code and models are available at https://github.com/hchoi256/mspl.

---

### Active Jammer Localization via Acquisition-Aware Path Planning

Authors: Luis González-Gudiño, Mariona Jaramillo-Civill, Pau Closas, Tales Imbiriba
Published: 2025-10-16
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.14790v1

We propose an active jammer localization framework that combines Bayesian optimization with acquisition-aware path planning. Unlike passive crowdsourced methods, our approach adaptively guides a mobile agent to collect high-utility Received Signal Strength measurements while accounting for urban obstacles and mobility constraints. For this, we modified the A* algorithm, A-UCB*, by incorporating acquisition values into trajectory costs, leading to high-acquisition planned paths. Simulations on realistic urban scenarios show that the proposed method achieves accurate localization with fewer measurements compared to uninformed baselines, demonstrating consistent performance under different environments.

---

### Cross-Scenario Unified Modeling of User Interests at Billion Scale

Authors: Manjie Xu, Cheng Chen, Xin Jia, Jingyi Zhou, Yongji Wu, Zejian Wang, Chi Zhang, Kai Zuo, Yibo Chen, Xu Tang, Yao Hu, Yixin Zhu
Published: 2025-10-16
Categories: cs.IR, cs.AI
arXiv: https://arxiv.org/abs/2510.14788v2

User interests on content platforms are inherently diverse, manifesting through complex behavioral patterns across heterogeneous scenarios such as search, feed browsing, and content discovery. Traditional recommendation systems typically prioritize business metric optimization within isolated specific scenarios, neglecting cross-scenario behavioral signals and struggling to integrate advanced techniques like LLMs at billion-scale deployments, which finally limits their ability to capture holistic user interests across platform touchpoints. We propose RED-Rec, an LLM-enhanced hierarchical Recommender Engine for Diversified scenarios, tailored for industry-level content recommendation systems. RED-Rec unifies user interest representations across multiple behavioral contexts by aggregating and synthesizing actions from varied scenarios, resulting in comprehensive item and user modeling. At its core, a two-tower LLM-powered framework enables nuanced, multifaceted representations with deployment efficiency, and a scenario-aware dense mixing and querying policy effectively fuses diverse behavioral signals to capture cross-scenario user intent patterns and express fine-grained, context-specific intents during serving. We validate RED-Rec through online A/B testing on hundreds of millions of users in RedNote through online A/B testing, showing substantial performance gains in both content recommendation and advertisement targeting tasks. We further introduce a million-scale sequential recommendation dataset, RED-MMU, for comprehensive offline training and evaluation. Our work advances unified user modeling, unlocking deeper personalization and fostering more meaningful user engagement in large-scale UGC platforms.

---

### Causal Discovery for Linear DAGs with Dependent Latent Variables via Higher-order Cumulants

Authors: Ming Cai, Penggang Gao, Hisayuki Hara
Published: 2025-10-16
Categories: cs.LG, stat.ML
arXiv: https://arxiv.org/abs/2510.14780v1

This paper addresses the problem of estimating causal directed acyclic graphs in linear non-Gaussian acyclic models with latent confounders (LvLiNGAM). Existing methods assume mutually independent latent confounders or cannot properly handle models with causal relationships among observed variables. We propose a novel algorithm that identifies causal DAGs in LvLiNGAM, allowing causal structures among latent variables, among observed variables, and between the two. The proposed method leverages higher-order cumulants of observed data to identify the causal structure. Extensive simulations and experiments with real-world data demonstrate the validity and practical utility of the proposed algorithm.

---

### Leveraging Code Cohesion Analysis to Identify Source Code Supply Chain Attacks

Authors: Maor Reuben, Ido Mendel, Or Feldman, Moshe Kravchik, Mordehai Guri, Rami Puzis
Published: 2025-10-16
Categories: cs.SE, cs.LG
arXiv: https://arxiv.org/abs/2510.14778v2

Supply chain attacks significantly threaten software security with malicious code injections within legitimate projects. Such attacks are very rare but may have a devastating impact. Detecting spurious code injections using automated tools is further complicated as it often requires deciphering the intention of both the inserted code and its context. In this study, we propose an unsupervised approach for highlighting spurious code injections by quantifying cohesion disruptions in the source code. Using a name-prediction-based cohesion (NPC) metric, we analyze how function cohesion changes when malicious code is introduced compared to natural cohesion fluctuations. An analysis of 54,707 functions over 369 open-source C++ repositories reveals that code injection reduces cohesion and shifts naming patterns toward shorter, less descriptive names compared to genuine function updates. Considering the sporadic nature of real supply-chain attacks, we evaluate the proposed method with extreme test-set imbalance and show that monitoring high-cohesion functions with NPC can effectively detect functions with injected code, achieving a Precision@100 of 36.41% at a 1:1,000 ratio and 12.47% at 1:10,000. These results suggest that automated cohesion measurements, in general, and name-prediction-based cohesion, in particular, may help identify supply chain attacks, improving source code integrity.

---

### Finding Answers in Thought Matters: Revisiting Evaluation on Large Language Models with Reasoning

Authors: Hwiyeol Jo, Joosung Lee, Jaehone Lee, Sang-Woo Lee, Joonsuk Park, Kang Min Yoo
Published: 2025-10-16
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2510.14773v1

Evaluating generative models, such as large language models (LLMs), commonly involves question-answering tasks where the final answer is selected based on probability of answer choices. On the other hand, for models requiring reasoning, the method of answer extraction plays a critical role. Our research reveals that the performance of reasoning models and their final answer distributions are highly sensitive to the answer extraction algorithm employed. In order to mitigate this, we propose a basic framework: Answer Regeneration. The method uses an additional model inference, providing the prior input and output prefaced by the prompt "Answer:". The final answer is then selected or extracted from the regenerated output. We show that this extraction-rule-agnostic approach exhibits improved performance and enhanced robustness. Furthermore, we have applied this framework to general math problems and open-ended question answering tasks. Our analysis and this framework could offer a more reliable results for model evaluation.

---

### MoCom: Motion-based Inter-MAV Visual Communication Using Event Vision and Spiking Neural Networks

Authors: Zhang Nengbo, Hann Woei Ho, Ye Zhou
Published: 2025-10-16
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.14770v1

Reliable communication in Micro Air Vehicle (MAV) swarms is challenging in environments, where conventional radio-based methods suffer from spectrum congestion, jamming, and high power consumption. Inspired by the waggle dance of honeybees, which efficiently communicate the location of food sources without sound or contact, we propose a novel visual communication framework for MAV swarms using motion-based signaling. In this framework, MAVs convey information, such as heading and distance, through deliberate flight patterns, which are passively captured by event cameras and interpreted using a predefined visual codebook of four motion primitives: vertical (up/down), horizontal (left/right), left-to-up-to-right, and left-to-down-to-right, representing control symbols (``start'', ``end'', ``1'', ``0''). To decode these signals, we design an event frame-based segmentation model and a lightweight Spiking Neural Network (SNN) for action recognition. An integrated decoding algorithm then combines segmentation and classification to robustly interpret MAV motion sequences. Experimental results validate the framework's effectiveness, which demonstrates accurate decoding and low power consumption, and highlights its potential as an energy-efficient alternative for MAV communication in constrained environments.

---

### Inpainting the Red Planet: Diffusion Models for the Reconstruction of Martian Environments in Virtual Reality

Authors: Giuseppe Lorenzo Catalano, Agata Marta Soccini
Published: 2025-10-16
Categories: cs.CV, cs.AI, cs.GR
arXiv: https://arxiv.org/abs/2510.14765v2

Space exploration increasingly relies on Virtual Reality for several tasks, such as mission planning, multidisciplinary scientific analysis, and astronaut training. A key factor for the reliability of the simulations is having accurate 3D representations of planetary terrains. Extraterrestrial heightmaps derived from satellite imagery often contain missing values due to acquisition and transmission constraints. Mars is among the most studied planets beyond Earth, and its extensive terrain datasets make the Martian surface reconstruction a valuable task, although many areas remain unmapped. Deep learning algorithms can support void-filling tasks; however, whereas Earth's comprehensive datasets enables the use of conditional methods, such approaches cannot be applied to Mars. Current approaches rely on simpler interpolation techniques which, however, often fail to preserve geometric coherence. In this work, we propose a method for reconstructing the surface of Mars based on an unconditional diffusion model. Training was conducted on an augmented dataset of 12000 Martian heightmaps derived from NASA's HiRISE survey. A non-homogeneous rescaling strategy captures terrain features across multiple scales before resizing to a fixed 128x128 model resolution. We compared our method against established void-filling and inpainting techniques, including Inverse Distance Weighting, kriging, and Navier-Stokes algorithm, on an evaluation set of 1000 samples. Results show that our approach consistently outperforms these methods in terms of reconstruction accuracy (4-15% on RMSE) and perceptual similarity (29-81% on LPIPS) with the original data.

---

### COIG-Writer: A High-Quality Dataset for Chinese Creative Writing with Thought Processes

Authors: Yunwen Li, Shuangshuang Ying, Xingwei Qu, Xin Li, Sheng Jin, Minghao Liu, Zhoufutu Wen, Tianyu Zheng, Xeron Du, Qiguang Chen, Jiajun Shi, Wangchunshu Zhou et al.
Published: 2025-10-16
Categories: cs.CL, cs.AI
arXiv: https://arxiv.org/abs/2510.14763v1

Large language models exhibit systematic deficiencies in creative writing, particularly in non-English contexts where training data is scarce and lacks process-level supervision. We present COIG-Writer, a novel Chinese creative writing dataset that captures both diverse outputs and their underlying thought processes through systematic reverse-engineering of high-quality texts. Unlike existing datasets that provide only input-output pairs, COIG-Writer comprises 1,665 meticulously curated triplets spanning 51 genres, each containing: (1) a reverse-engineered prompt, (2) detailed creative reasoning documenting decision-making processes, and (3) the final text. Through comprehensive experiments, we identify a two-component model of creative writing: narrative logic (provided by process supervision) and linguistic expression (maintained by general-purpose data). Our findings reveal three critical insights: (1) Process supervision is highly effective but requires stabilization with general data. A ratio of at least one creative sample to twelve general samples is needed to achieve optimal performance; below this threshold, the win rate progressively degrades (from 62.75% down to 35.78%)., (2) creative capabilities are culturally-bound with no cross-lingual transfer (89.26pp gap between Chinese and English performance), and (3) lexical diversity inversely correlates with creative quality (TTR paradox), suggesting high diversity signals compensatory behavior for logical deficiencies. These findings establish that creative excellence emerges from the interaction between logical scaffolding and linguistic grounding, analogous to how mathematical reasoning enhances but cannot replace linguistic competence in foundation models.

---

### Pluto: A Benchmark for Evaluating Efficiency of LLM-generated Hardware Code

Authors: Manar Abdelatty, Maryam Nouh, Jacob K. Rosenstein, Sherief Reda
Published: 2025-10-16
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.14756v1

Large Language Models (LLMs) are increasingly used to automate hardware design tasks, including the generation of Verilog code. While early benchmarks focus primarily on functional correctness, efficient hardware design demands additional optimization for synthesis metrics such as area, delay, and power. Existing benchmarks fall short in evaluating these aspects comprehensively: they often lack optimized baselines or testbenches for verification. To address these gaps, we present Pluto, a benchmark and evaluation framework designed to assess the efficiency of LLM-generated Verilog designs. Pluto presents a comprehensive evaluation set of 114 problems with self-checking testbenches and multiple Pareto-optimal reference implementations. Experimental results show that state-of-the-art LLMs can achieve high functional correctness, reaching 78.3\% at pass@1, but their synthesis efficiency still lags behind expert-crafted implementations, with area efficiency of 63.8\%, delay efficiency of 65.9\%, and power efficiency of 64.0\% at eff@1. This highlights the need for efficiency-aware evaluation frameworks such as Pluto to drive progress in hardware-focused LLM research.

---

### LightQANet: Quantized and Adaptive Feature Learning for Low-Light Image Enhancement

Authors: Xu Wu, Zhihui Lai, Xianxu Hou, Jie Zhou, Ya-nan Zhang, Linlin Shen
Published: 2025-10-16
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.14753v1

Low-light image enhancement (LLIE) aims to improve illumination while preserving high-quality color and texture. However, existing methods often fail to extract reliable feature representations due to severely degraded pixel-level information under low-light conditions, resulting in poor texture restoration, color inconsistency, and artifact. To address these challenges, we propose LightQANet, a novel framework that introduces quantized and adaptive feature learning for low-light enhancement, aiming to achieve consistent and robust image quality across diverse lighting conditions. From the static modeling perspective, we design a Light Quantization Module (LQM) to explicitly extract and quantify illumination-related factors from image features. By enforcing structured light factor learning, LQM enhances the extraction of light-invariant representations and mitigates feature inconsistency across varying illumination levels. From the dynamic adaptation perspective, we introduce a Light-Aware Prompt Module (LAPM), which encodes illumination priors into learnable prompts to dynamically guide the feature learning process. LAPM enables the model to flexibly adapt to complex and continuously changing lighting conditions, further improving image enhancement. Extensive experiments on multiple low-light datasets demonstrate that our method achieves state-of-the-art performance, delivering superior qualitative and quantitative results across various challenging lighting scenarios.

---

### Beyond Multi-Token Prediction: Pretraining LLMs with Future Summaries

Authors: Divyat Mahajan, Sachin Goyal, Badr Youbi Idrissi, Mohammad Pezeshki, Ioannis Mitliagkas, David Lopez-Paz, Kartik Ahuja
Published: 2025-10-16
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2510.14751v2

Next-token prediction (NTP) has driven the success of large language models (LLMs), but it struggles with long-horizon reasoning, planning, and creative writing, with these limitations largely attributed to teacher-forced training. Multi-token prediction (MTP) partially mitigates these issues by predicting several future tokens at once, but it mostly captures short-range dependencies and offers limited improvement. We propose future summary prediction (FSP), which trains an auxiliary head to predict a compact representation of the long-term future, preserving information relevant for long-form generations. We explore two variants of FSP: handcrafted summaries, for example, a bag of words summary of the future sequence, and learned summaries, which use embeddings produced by a reverse language model trained from right-to-left order. Large-scale pretraining experiments (3B and 8B-parameter models) demonstrate that FSP provides improvements over both NTP and MTP across math, reasoning, and coding benchmarks.

---

### DEXTER: Diffusion-Guided EXplanations with TExtual Reasoning for Vision Models

Authors: Simone Carnemolla, Matteo Pennisi, Sarinda Samarasinghe, Giovanni Bellitto, Simone Palazzo, Daniela Giordano, Mubarak Shah, Concetto Spampinato
Published: 2025-10-16
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2510.14741v2

Understanding and explaining the behavior of machine learning models is essential for building transparent and trustworthy AI systems. We introduce DEXTER, a data-free framework that employs diffusion models and large language models to generate global, textual explanations of visual classifiers. DEXTER operates by optimizing text prompts to synthesize class-conditional images that strongly activate a target classifier. These synthetic samples are then used to elicit detailed natural language reports that describe class-specific decision patterns and biases. Unlike prior work, DEXTER enables natural language explanation about a classifier's decision process without access to training data or ground-truth labels. We demonstrate DEXTER's flexibility across three tasks-activation maximization, slice discovery and debiasing, and bias explanation-each illustrating its ability to uncover the internal mechanisms of visual classifiers. Quantitative and qualitative evaluations, including a user study, show that DEXTER produces accurate, interpretable outputs. Experiments on ImageNet, Waterbirds, CelebA, and FairFaces confirm that DEXTER outperforms existing approaches in global model explanation and class-level bias reporting. Code is available at https://github.com/perceivelab/dexter.

---

### AutoRubric: Rubric-Based Generative Rewards for Faithful Multimodal Reasoning

Authors: Mengzhao Jia, Zhihan Zhang, Ignacio Cases, Zheyuan Liu, Meng Jiang, Peng Qi
Published: 2025-10-16
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.14738v2

Multimodal large language models (MLLMs) have rapidly advanced from perception tasks to complex multi-step reasoning, yet reinforcement learning with verifiable rewards (RLVR) often leads to spurious reasoning since only the final-answer correctness is rewarded. To address this limitation, we propose AutoRubric, a framework that integrates RLVR with process-level supervision through automatically collected rubric-based generative rewards. Our key innovation lies in a scalable self-aggregation method that distills consistent reasoning checkpoints from successful trajectories, enabling problem-specific rubric construction without human annotation or stronger teacher models. By jointly leveraging rubric-based and outcome rewards, AutoRubric achieves state-of-the-art performance on six multimodal reasoning benchmarks and substantially improves reasoning faithfulness in dedicated evaluations.

---

### Free-Grained Hierarchical Visual Recognition

Authors: Seulki Park, Zilin Wang, Stella X. Yu
Published: 2025-10-16
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.14737v3

Hierarchical image recognition seeks to predict class labels along a semantic taxonomy, from broad categories to specific ones, typically under the tidy assumption that every training image is fully annotated along its taxonomy path. Reality is messier: A distant bird may be labeled only bird, while a clear close-up may justify bald eagle. We introduce free-grain training, where labels may appear at any level of the taxonomy and models must learn consistent hierarchical predictions from incomplete, mixed-granularity supervision. We build benchmark datasets with varying label granularity and show that existing hierarchical methods deteriorate sharply in this setting. To make up for missing supervision, we propose two simple solutions: One adds broad text-based supervision that captures visual attributes, and the other treats missing labels at specific taxonomy levels as a semi-supervised learning problem. We also study free-grained inference, where the model chooses how deep to predict, returning a reliable coarse label when a fine-grained one is uncertain. Together, our task, datasets, and methods move hierarchical recognition closer to the way labels arise in the real world.

---

### The Pursuit of Diversity: Multi-Objective Testing of Deep Reinforcement Learning Agents

Authors: Antony Bartlett, Cynthia Liem, Annibale Panichella
Published: 2025-10-16
Categories: cs.LG
arXiv: https://arxiv.org/abs/2510.14727v1

Testing deep reinforcement learning (DRL) agents in safety-critical domains requires discovering diverse failure scenarios. Existing tools such as INDAGO rely on single-objective optimization focused solely on maximizing failure counts, but this does not ensure discovered scenarios are diverse or reveal distinct error types. We introduce INDAGO-Nexus, a multi-objective search approach that jointly optimizes for failure likelihood and test scenario diversity using multi-objective evolutionary algorithms with multiple diversity metrics and Pareto front selection strategies. We evaluated INDAGO-Nexus on three DRL agents: humanoid walker, self-driving car, and parking agent. On average, INDAGO-Nexus discovers up to 83% and 40% more unique failures (test effectiveness) than INDAGO in the SDC and Parking scenarios, respectively, while reducing time-to-failure by up to 67% across all agents.

---

### Cross-Layer Feature Self-Attention Module for Multi-Scale Object Detection

Authors: Dingzhou Xie, Rushi Lan, Cheng Pang, Enhao Ning, Jiahao Zeng, Wei Zheng
Published: 2025-10-16
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.14726v1

Recent object detection methods have made remarkable progress by leveraging attention mechanisms to improve feature discriminability. However, most existing approaches are confined to refining single-layer or fusing dual-layer features, overlooking the rich inter-layer dependencies across multi-scale representations. This limits their ability to capture comprehensive contextual information essential for detecting objects with large scale variations. In this paper, we propose a novel Cross-Layer Feature Self-Attention Module (CFSAM), which holistically models both local and global dependencies within multi-scale feature maps. CFSAM consists of three key components: a convolutional local feature extractor, a Transformer-based global modeling unit that efficiently captures cross-layer interactions, and a feature fusion mechanism to restore and enhance the original representations. When integrated into the SSD300 framework, CFSAM significantly boosts detection performance, achieving 78.6% mAP on PASCAL VOC (vs. 75.5% baseline) and 52.1% mAP on COCO (vs. 43.1% baseline), outperforming existing attention modules. Moreover, the module accelerates convergence during training without introducing substantial computational overhead. Our work highlights the importance of explicit cross-layer attention modeling in advancing multi-scale object detection.

---

### Tawa: Automatic Warp Specialization for Modern GPUs with Asynchronous References

Authors: Hongzheng Chen, Bin Fan, Alexander Collins, Bastian Hagedorn, Evghenii Gaburov, Masahiro Masuda, Matthew Brookhart, Chris Sullivan, Jason Knight, Zhiru Zhang, Vinod Grover
Published: 2025-10-16
Categories: cs.LG, cs.AR, cs.PL
arXiv: https://arxiv.org/abs/2510.14719v2

Modern GPUs feature specialized hardware units that enable high-performance, asynchronous dataflow execution. However, the conventional SIMT programming model is fundamentally misaligned with this task-parallel hardware, creating a significant programmability gap. While hardware-level warp specialization is the key to unlocking peak performance, it forces developers to manually orchestrate complex, low-level communication and software pipelines--a process that is labor-intensive, error-prone, and unsustainable. To address this challenge, we present Tawa, an automated compiler that systematically generates high-performance, warp-specialized code from a high-level, tile-based program. Central to our approach is a novel IR abstraction, asynchronous references (aref), which expresses warp-level communication without exposing low-level hardware details. Using this abstraction, Tawa automatically partitions programs into producer-consumer roles and manages the intricate dataflow pipeline, relieving developers of invasive kernel rewriting. Evaluation on NVIDIA H100 GPUs across representative LLM kernels shows that Tawa delivers high hardware utilization, achieving up to 1.1$\times$ speedup over highly optimized cuBLAS GEMM kernels. For attention workloads, Tawa attains 1.2$\times$ speedup over Triton and matches the performance of the hand-optimized CUTLASS C++ FlashAttention-3 kernel with far less programming effort.

---

### Telling Speculative Stories to Help Humans Imagine the Harms of Healthcare AI

Authors: Xingmeng Zhao, Tongnian Wang, Dan Schumacher, Veronica Rammouz, Anthony Rios
Published: 2025-10-16
Categories: cs.CL
arXiv: https://arxiv.org/abs/2510.14718v3

Artificial intelligence (AI) is rapidly transforming healthcare, enabling fast development of tools like stress monitors, wellness trackers, and mental health chatbots. However, rapid and low-barrier development can introduce risks of bias, privacy violations, and unequal access, especially when systems ignore real-world contexts and diverse user needs. Many recent methods use AI to detect risks automatically, but this can reduce human engagement in understanding how harms arise and who they affect. We present a human-centered framework that generates user stories and supports multi-agent discussions to help people think creatively about potential benefits and harms before deployment. In a user study, participants who read stories recognized a broader range of harms, distributing their responses more evenly across all 17 harm types. In contrast, those who did not read stories focused primarily on privacy and well-being (79.1%). Our findings show that storytelling helped participants speculate about a broader range of harms and benefits and think more creatively about AI's impact on users.

---

### Seesaw: Accelerating Training by Balancing Learning Rate and Batch Size Scheduling

Authors: Alexandru Meterez, Depen Morwani, Jingfeng Wu, Costin-Andrei Oncescu, Cengiz Pehlevan, Sham Kakade
Published: 2025-10-16
Categories: cs.LG, cs.AI, math.OC, stat.ML
arXiv: https://arxiv.org/abs/2510.14717v2

Increasing the batch size during training -- a ''batch ramp'' -- is a promising strategy to accelerate large language model pretraining. While for SGD, doubling the batch size can be equivalent to halving the learning rate, the optimal strategy for adaptive optimizers like Adam is less clear. As a result, any batch-ramp scheduling, if used at all, is typically tuned heuristically. This work develops a principled framework for batch-size scheduling and introduces Seesaw: whenever a standard scheduler would halve the learning rate, Seesaw instead multiplies it by $1/\sqrt{2}$ and doubles the batch size, preserving loss dynamics while reducing serial steps. Theoretically, we provide, to our knowledge, the first finite-sample proof of equivalence between learning-rate decay and batch-size ramp-up for SGD on noisy linear regression, and we extend this equivalence to normalized SGD, a tractable proxy for Adam, under a variance-dominated regime observed in practice. Empirically, on 150M/300M/600M-parameter models trained at Chinchilla scale using a constant (critical) batch size, Seesaw matches cosine decay at equal FLOPs while reducing wall-clock time by $\approx 36\%$, approaching the theoretical limit implied by our analysis.

---

### Camera Movement Classification in Historical Footage: A Comparative Study of Deep Video Models

Authors: Tingyu Lin, Armin Dadras, Florian Kleber, Robert Sablatnig
Published: 2025-10-16
Categories: cs.CV, cs.AI, eess.IV
arXiv: https://arxiv.org/abs/2510.14713v1

Camera movement conveys spatial and narrative information essential for understanding video content. While recent camera movement classification (CMC) methods perform well on modern datasets, their generalization to historical footage remains unexplored. This paper presents the first systematic evaluation of deep video CMC models on archival film material. We summarize representative methods and datasets, highlighting differences in model design and label definitions. Five standard video classification models are assessed on the HISTORIAN dataset, which includes expert-annotated World War II footage. The best-performing model, Video Swin Transformer, achieves 80.25% accuracy, showing strong convergence despite limited training data. Our findings highlight the challenges and potential of adapting existing models to low-quality video and motivate future work combining diverse input modalities and temporal architectures.

---

### Fast and Scalable Score-Based Kernel Calibration Tests

Authors: Pierre Glaser, David Widmann, Fredrik Lindsten, Arthur Gretton
Published: 2025-10-16
Categories: stat.ML, cs.LG
arXiv: https://arxiv.org/abs/2510.14711v1

We introduce the Kernel Calibration Conditional Stein Discrepancy test (KCCSD test), a non-parametric, kernel-based test for assessing the calibration of probabilistic models with well-defined scores. In contrast to previous methods, our test avoids the need for possibly expensive expectation approximations while providing control over its type-I error. We achieve these improvements by using a new family of kernels for score-based probabilities that can be estimated without probability density samples, and by using a conditional goodness-of-fit criterion for the KCCSD test's U-statistic. We demonstrate the properties of our test on various synthetic settings.

---

### MCbiF: Measuring Topological Autocorrelation in Multiscale Clusterings via 2-Parameter Persistent Homology

Authors: Juni Schindler, Mauricio Barahona
Published: 2025-10-16
Categories: math.AT, cs.LG, physics.data-an
arXiv: https://arxiv.org/abs/2510.14710v2

Datasets often possess an intrinsic multiscale structure with meaningful descriptions at different levels of coarseness. Such datasets are naturally described as multi-resolution clusterings, i.e., not necessarily hierarchical sequences of partitions across scales. To analyse and compare such sequences, we use tools from topological data analysis and define the Multiscale Clustering Bifiltration (MCbiF), a 2-parameter filtration of abstract simplicial complexes that encodes cluster intersection patterns across scales. The MCbiF is a complete invariant of (non-hierarchical) sequences of partitions and can be interpreted as a higher-order extension of Sankey diagrams, which reduce to dendrograms for hierarchical sequences. We show that the multiparameter persistent homology (MPH) of the MCbiF yields a finitely presented and block decomposable module, and its stable Hilbert functions characterise the topological autocorrelation of the sequence of partitions. In particular, at dimension zero, the MPH captures violations of the refinement order of partitions, whereas at dimension one, the MPH captures higher-order inconsistencies between clusters across scales. We then demonstrate through experiments the use of MCbiF Hilbert functions as interpretable topological feature maps for downstream machine learning tasks, and show that MCbiF feature maps outperform both baseline features and representation learning methods on regression and classification tasks for non-hierarchical sequences of partitions. We also showcase an application of MCbiF to real-world data of non-hierarchical wild mice social grouping patterns across time.

---

### Where are the Whales: A Human-in-the-loop Detection Method for Identifying Whales in High-resolution Satellite Imagery

Authors: Caleb Robinson, Kimberly T. Goetz, Christin B. Khan, Meredith Sackett, Kathleen Leonard, Rahul Dodhia, Juan M. Lavista Ferres
Published: 2025-10-16
Categories: cs.CV, cs.AI
arXiv: https://arxiv.org/abs/2510.14709v1

Effective monitoring of whale populations is critical for conservation, but traditional survey methods are expensive and difficult to scale. While prior work has shown that whales can be identified in very high-resolution (VHR) satellite imagery, large-scale automated detection remains challenging due to a lack of annotated imagery, variability in image quality and environmental conditions, and the cost of building robust machine learning pipelines over massive remote sensing archives. We present a semi-automated approach for surfacing possible whale detections in VHR imagery using a statistical anomaly detection method that flags spatial outliers, i.e. "interesting points". We pair this detector with a web-based labeling interface designed to enable experts to quickly annotate the interesting points. We evaluate our system on three benchmark scenes with known whale annotations and achieve recalls of 90.3% to 96.4%, while reducing the area requiring expert inspection by up to 99.8% -- from over 1,000 sq km to less than 2 sq km in some cases. Our method does not rely on labeled training data and offers a scalable first step toward future machine-assisted marine mammal monitoring from space. We have open sourced this pipeline at https://github.com/microsoft/whales.

---

### Leveraging Learned Image Prior for 3D Gaussian Compression

Authors: Seungjoo Shin, Jaesik Park, Sunghyun Cho
Published: 2025-10-16
Categories: cs.CV
arXiv: https://arxiv.org/abs/2510.14705v1

Compression techniques for 3D Gaussian Splatting (3DGS) have recently achieved considerable success in minimizing storage overhead for 3D Gaussians while preserving high rendering quality. Despite the impressive storage reduction, the lack of learned priors restricts further advances in the rate-distortion trade-off for 3DGS compression tasks. To address this, we introduce a novel 3DGS compression framework that leverages the powerful representational capacity of learned image priors to recover compression-induced quality degradation. Built upon initially compressed Gaussians, our restoration network effectively models the compression artifacts in the image space between degraded and original Gaussians. To enhance the rate-distortion performance, we provide coarse rendering residuals into the restoration network as side information. By leveraging the supervision of restored images, the compressed Gaussians are refined, resulting in a highly compact representation with enhanced rendering performance. Our framework is designed to be compatible with existing Gaussian compression methods, making it broadly applicable across different baselines. Extensive experiments validate the effectiveness of our framework, demonstrating superior rate-distortion performance and outperforming the rendering quality of state-of-the-art 3DGS compression methods while requiring substantially less storage.

---

### ToolPRM: Fine-Grained Inference Scaling of Structured Outputs for Function Calling

Authors: Jianghao Lin, Yuanyuan Shi, Xin Peng, Renjie Ding, Hairui Wang, Yuxuan Peng, Bizhe Bai, Weixi Song, Fengshuo Bai, Huacan Chai, Weinan Zhang, Fei Huang et al.
Published: 2025-10-16
Categories: cs.AI
arXiv: https://arxiv.org/abs/2510.14703v2

Large language models (LLMs) excel at function calling, but inference scaling has been explored mainly for unstructured generation. We propose an inference-scaling framework for structured outputs that combines fine-grained beam search with \textbf{ToolPRM}, a process reward model scoring each intra-call decision (function name and argument filling). We build the first fine-grained intra-call supervision dataset via function masking, rollout collection, and step-level annotation. ToolPRM outperforms outcome and coarse-grained reward models in predictive accuracy and yields consistent test-time gains on multiple function-calling benchmarks. We further show that structured generation follows ``\textbf{explore more but retain less}'', since early JSON errors are unrecoverable.

---

### Cognitive-Aligned Spatio-Temporal Large Language Models For Next Point-of-Interest Prediction

Authors: Penglong Zhai, Jie Li, Fanyi Di, Yue Liu, Yifang Yuan, Jie Huang, Peng Wu, Sicong Wang, Mingyang Yin, Tingting Hu, Yao Xu, Xin Li
Published: 2025-10-16
Categories: cs.AI
arXiv: https://arxiv.org/abs/2510.14702v1

The next point-of-interest (POI) recommendation task aims to predict the users' immediate next destinations based on their preferences and historical check-ins, holding significant value in location-based services. Recently, large language models (LLMs) have shown great potential in recommender systems, which treat the next POI prediction in a generative manner. However, these LLMs, pretrained primarily on vast corpora of unstructured text, lack the native understanding of structured geographical entities and sequential mobility patterns required for next POI prediction tasks. Moreover, in industrial-scale POI prediction applications, incorporating world knowledge and alignment of human cognition, such as seasons, weather conditions, holidays, and users' profiles (such as habits, occupation, and preferences), can enhance the user experience while improving recommendation performance. To address these issues, we propose CoAST (Cognitive-Aligned Spatial-Temporal LLMs), a framework employing natural language as an interface, allowing for the incorporation of world knowledge, spatio-temporal trajectory patterns, profiles, and situational information. Specifically, CoAST mainly comprises of 2 stages: (1) Recommendation Knowledge Acquisition through continued pretraining on the enriched spatial-temporal trajectory data of the desensitized users; (2) Cognitive Alignment to align cognitive judgments with human preferences using enriched training data through Supervised Fine-Tuning (SFT) and a subsequent Reinforcement Learning (RL) phase. Extensive offline experiments on various real-world datasets and online experiments deployed in "Guess Where You Go" of AMAP App homepage demonstrate the effectiveness of CoAST.

---

### FedPPA: Progressive Parameter Alignment for Personalized Federated Learning

Authors: Maulidi Adi Prasetia, Muhamad Risqi U. Saputra, Guntur Dharma Putra
Published: 2025-10-16
Categories: cs.LG, cs.AI
arXiv: https://arxiv.org/abs/2510.14698v1

Federated Learning (FL) is designed as a decentralized, privacy-preserving machine learning paradigm that enables multiple clients to collaboratively train a model without sharing their data. In real-world scenarios, however, clients often have heterogeneous computational resources and hold non-independent and identically distributed data (non-IID), which poses significant challenges during training. Personalized Federated Learning (PFL) has emerged to address these issues by customizing models for each client based on their unique data distribution. Despite its potential, existing PFL approaches typically overlook the coexistence of model and data heterogeneity arising from clients with diverse computational capabilities. To overcome this limitation, we propose a novel method, called Progressive Parameter Alignment (FedPPA), which progressively aligns the weights of common layers across clients with the global model's weights. Our approach not only mitigates inconsistencies between global and local models during client updates, but also preserves client's local knowledge, thereby enhancing personalization robustness in non-IID settings. To further enhance the global model performance while retaining strong personalization, we also integrate entropy-based weighted averaging into the FedPPA framework. Experiments on three image classification datasets, including MNIST, FMNIST, and CIFAR-10, demonstrate that FedPPA consistently outperforms existing FL algorithms, achieving superior performance in personalized adaptation.

---

### Purifying Task Vectors in Knowledge-Aware Subspace for Model Merging

Authors: Bang An, Yibo Yang, Philip Torr, Bernard Ghanem
Published: 2025-10-16
Categories: cs.AI
arXiv: https://arxiv.org/abs/2510.14697v1

Model merging aims to integrate task-specific abilities from individually fine-tuned models into a single model without extra training. In recent model merging methods, task vector has become a fundamental building block, as it can encapsulate the residual information from finetuning. However, the merged model often suffers from notable performance degradation due to the conflicts caused by task-irrelevant redundancy in task vectors. Existing efforts in overcoming redundancy by randomly dropping elements in the parameter space involves randomness and lacks knowledge awareness. To address these challenges, in this study, we propose Purifying TAsk Vectors (PAVE) in knowledge-aware subspace. Concretely, we sample some training examples from each task, and feed them into their corresponding fine-tuned models to acquire the covariance matrices before linear layers. We then perform a context-oriented singular value decomposition, which accentuates the weight components most relevant to the target knowledge. As a result, we can split fine-tuned model weights into task-relevant and redundant components in the knowledge-aware subspace, and purify the task vector by pruning the redundant components. To induce fair pruning efforts across models, we further introduce a spectral rank allocation strategy by optimizing a normalized activated pruning error. The task vector purification by our method as a plug-and-play scheme is applicable across various task vector-based merging methods to improve their performance. In experiments, we demonstrate the effectiveness of PAVE across a diverse set of merging methods, tasks, and model architectures.

---